From e75c7ab8d3ed8668cbc125c8dab51fae40369c39 Mon Sep 17 00:00:00 2001
From: gong1 <b.gong@fz-juelich.de>
Date: Tue, 15 Sep 2020 14:13:37 +0200
Subject: [PATCH] remove useless files and move files to new structure
---
.../HPC_scripts/hyperparam_setup.sh | 12 -
.../HPC_scripts/reset_dirs.sh | 11 -
.../bash/download_and_preprocess_dataset.sh | 88 --
.../download_and_preprocess_dataset_era5.sh | 127 ---
.../download_and_preprocess_dataset_v1.sh | 86 --
video_prediction_savp/bash/workflow_era5.sh | 92 --
.../bash/workflow_era5_macOS.sh | 93 --
.../bash/workflow_era5_zam347.sh | 93 --
.../{ => deprecate}/helper/helper.py | 0
.../pretrained_models/download_model.sh | 0
.../scripts/combine_results.py | 0
.../{ => deprecate}/scripts/evaluate.py | 0
.../{ => deprecate}/scripts/evaluate_all.sh | 0
.../{ => deprecate}/scripts/generate_all.sh | 0
.../{ => deprecate}/scripts/generate_orig.py | 0
.../{ => deprecate}/scripts/plot_results.py | 0
.../scripts/plot_results_all.sh | 0
.../{ => deprecate}/scripts/train.py | 0
.../{ => deprecate}/scripts/train_all.sh | 0
.../docs/presentation/presentation.md | 5 -
video_prediction_savp/scripts/Analysis_all.py | 105 ---
video_prediction_savp/scripts/evaluate_svg.sh | 1 -
video_prediction_savp/scripts/generate.py | 537 ------------
.../scripts}/mpi_stager_v2.py | 0
.../scripts}/mpi_stager_v2_process_netCDF.py | 0
video_prediction_savp/scripts/train_v2.py | 362 --------
.../utils}/external_function.py | 0
.../utils}/shiftgrid.py | 0
.../datasets}/prepare_era5_data.py | 0
.../datasets}/process_netCDF_v2.py | 0
.../DataExtraction/helper_single_master.py | 245 ------
.../DataExtraction/main_single_master.py | 257 ------
.../DataExtraction/readme.md | 2 -
.../DataExtraction/submitJob.sh | 22 -
.../Stager_devel_N_24_evaluation.sh | 40 -
.../DataPostprocess/data_utils.py | 164 ----
.../DataPostprocess/keras_utils.py | 58 --
.../kitti_evaluate_parallel.py | 159 ----
.../DataPostprocess/kitti_settings.py | 19 -
.../DataPostprocess/packageInstallation.sh | 9 -
.../parameters_kitti_evaluate.dat | 14 -
.../DataPostprocess/prednet.py | 311 -------
.../Stager_devel_N_24_process_netCDF.sh | 22 -
.../DataPreprocess/external_function.py | 158 ----
.../mpi_split_data_multi_years.py | 58 --
workflow_parallel_frame_prediction/README.md | 56 --
.../Training/data_utils.py | 163 ----
.../Training/devel_horovodJob.sh | 43 -
.../Training/evaluate_multistep.py | 121 ---
.../Training/hickle/bin/f2py | 4 -
.../__pycache__/site.cpython-36.pyc | Bin 1482 -> 0 bytes
.../python3.6/site-packages/easy-install.pth | 4 -
.../hickle-3.4.3-py3.6.egg/EGG-INFO/PKG-INFO | 207 -----
.../EGG-INFO/SOURCES.txt | 52 --
.../EGG-INFO/dependency_links.txt | 1 -
.../EGG-INFO/not-zip-safe | 1 -
.../EGG-INFO/requires.txt | 2 -
.../EGG-INFO/top_level.txt | 2 -
.../hickle-3.4.3-py3.6.egg/hickle/__init__.py | 4 -
.../__pycache__/__init__.cpython-36.pyc | Bin 278 -> 0 bytes
.../hickle/__pycache__/helpers.cpython-36.pyc | Bin 3266 -> 0 bytes
.../hickle/__pycache__/hickle.cpython-36.pyc | Bin 17742 -> 0 bytes
.../__pycache__/hickle_legacy.cpython-36.pyc | Bin 14836 -> 0 bytes
.../__pycache__/hickle_legacy2.cpython-36.pyc | Bin 20746 -> 0 bytes
.../hickle/__pycache__/lookup.cpython-36.pyc | Bin 6661 -> 0 bytes
.../hickle-3.4.3-py3.6.egg/hickle/helpers.py | 113 ---
.../hickle-3.4.3-py3.6.egg/hickle/hickle.py | 611 -------------
.../hickle/hickle_legacy.py | 535 ------------
.../hickle/hickle_legacy2.py | 672 --------------
.../hickle/loaders/__init__.py | 1 -
.../__pycache__/__init__.cpython-36.pyc | Bin 252 -> 0 bytes
.../__pycache__/load_astropy.cpython-36.pyc | Bin 7453 -> 0 bytes
.../__pycache__/load_numpy.cpython-36.pyc | Bin 4199 -> 0 bytes
.../__pycache__/load_pandas.cpython-36.pyc | Bin 249 -> 0 bytes
.../__pycache__/load_python.cpython-36.pyc | Bin 4349 -> 0 bytes
.../__pycache__/load_python3.cpython-36.pyc | Bin 5988 -> 0 bytes
.../__pycache__/load_scipy.cpython-36.pyc | Bin 3032 -> 0 bytes
.../hickle/loaders/load_astropy.py | 237 -----
.../hickle/loaders/load_numpy.py | 145 ---
.../hickle/loaders/load_pandas.py | 4 -
.../hickle/loaders/load_python.py | 141 ---
.../hickle/loaders/load_python3.py | 201 -----
.../hickle/loaders/load_scipy.py | 92 --
.../hickle-3.4.3-py3.6.egg/hickle/lookup.py | 238 -----
.../hickle-3.4.3-py3.6.egg/tests/__init__.py | 0
.../tests/__pycache__/__init__.cpython-36.pyc | Bin 209 -> 0 bytes
.../__pycache__/test_astropy.cpython-36.pyc | Bin 3798 -> 0 bytes
.../__pycache__/test_hickle.cpython-36.pyc | Bin 18861 -> 0 bytes
.../test_hickle_helpers.cpython-36.pyc | Bin 1969 -> 0 bytes
.../test_legacy_load.cpython-36.pyc | Bin 840 -> 0 bytes
.../__pycache__/test_scipy.cpython-36.pyc | Bin 1718 -> 0 bytes
.../tests/test_astropy.py | 133 ---
.../tests/test_hickle.py | 826 ------------------
.../tests/test_hickle_helpers.py | 63 --
.../tests/test_legacy_load.py | 30 -
.../tests/test_scipy.py | 57 --
.../lib/python3.6/site-packages/site.py | 74 --
.../Training/horovodJob.sh | 44 -
.../Training/keras_utils.py | 58 --
.../Training/kitti_settings.py | 19 -
.../Training/kitti_train_horovod.py | 119 ---
.../Training/minMaxExtractor.py | 76 --
.../Training/prednet.py | 311 -------
.../Training/process_netCDF.py | 169 ----
.../Workflow.png | Bin 195317 -> 0 bytes
.../__init__.py | 0
workflow_parallel_frame_prediction/main.sh | 28 -
workflow_parallel_frame_prediction/setup.py | 11 -
108 files changed, 8818 deletions(-)
delete mode 100644 video_prediction_savp/HPC_scripts/hyperparam_setup.sh
delete mode 100644 video_prediction_savp/HPC_scripts/reset_dirs.sh
delete mode 100644 video_prediction_savp/bash/download_and_preprocess_dataset.sh
delete mode 100644 video_prediction_savp/bash/download_and_preprocess_dataset_era5.sh
delete mode 100644 video_prediction_savp/bash/download_and_preprocess_dataset_v1.sh
delete mode 100755 video_prediction_savp/bash/workflow_era5.sh
delete mode 100755 video_prediction_savp/bash/workflow_era5_macOS.sh
delete mode 100755 video_prediction_savp/bash/workflow_era5_zam347.sh
rename video_prediction_savp/{ => deprecate}/helper/helper.py (100%)
rename video_prediction_savp/{ => deprecate}/pretrained_models/download_model.sh (100%)
rename video_prediction_savp/{ => deprecate}/scripts/combine_results.py (100%)
rename video_prediction_savp/{ => deprecate}/scripts/evaluate.py (100%)
rename video_prediction_savp/{ => deprecate}/scripts/evaluate_all.sh (100%)
rename video_prediction_savp/{ => deprecate}/scripts/generate_all.sh (100%)
rename video_prediction_savp/{ => deprecate}/scripts/generate_orig.py (100%)
rename video_prediction_savp/{ => deprecate}/scripts/plot_results.py (100%)
rename video_prediction_savp/{ => deprecate}/scripts/plot_results_all.sh (100%)
rename video_prediction_savp/{ => deprecate}/scripts/train.py (100%)
rename video_prediction_savp/{ => deprecate}/scripts/train_all.sh (100%)
delete mode 100644 video_prediction_savp/docs/presentation/presentation.md
delete mode 100644 video_prediction_savp/scripts/Analysis_all.py
delete mode 100644 video_prediction_savp/scripts/evaluate_svg.sh
delete mode 100644 video_prediction_savp/scripts/generate.py
rename {workflow_parallel_frame_prediction/DataExtraction => video_prediction_savp/scripts}/mpi_stager_v2.py (100%)
rename {workflow_parallel_frame_prediction/DataPreprocess => video_prediction_savp/scripts}/mpi_stager_v2_process_netCDF.py (100%)
delete mode 100644 video_prediction_savp/scripts/train_v2.py
rename {workflow_parallel_frame_prediction/DataExtraction => video_prediction_savp/utils}/external_function.py (100%)
rename {workflow_parallel_frame_prediction/DataExtraction => video_prediction_savp/utils}/shiftgrid.py (100%)
rename {workflow_parallel_frame_prediction/DataExtraction => video_prediction_savp/video_prediction/datasets}/prepare_era5_data.py (100%)
rename {workflow_parallel_frame_prediction/DataPreprocess => video_prediction_savp/video_prediction/datasets}/process_netCDF_v2.py (100%)
delete mode 100644 workflow_parallel_frame_prediction/DataExtraction/helper_single_master.py
delete mode 100644 workflow_parallel_frame_prediction/DataExtraction/main_single_master.py
delete mode 100644 workflow_parallel_frame_prediction/DataExtraction/readme.md
delete mode 100755 workflow_parallel_frame_prediction/DataExtraction/submitJob.sh
delete mode 100755 workflow_parallel_frame_prediction/DataPostprocess/Stager_devel_N_24_evaluation.sh
delete mode 100755 workflow_parallel_frame_prediction/DataPostprocess/data_utils.py
delete mode 100755 workflow_parallel_frame_prediction/DataPostprocess/keras_utils.py
delete mode 100755 workflow_parallel_frame_prediction/DataPostprocess/kitti_evaluate_parallel.py
delete mode 100755 workflow_parallel_frame_prediction/DataPostprocess/kitti_settings.py
delete mode 100644 workflow_parallel_frame_prediction/DataPostprocess/packageInstallation.sh
delete mode 100755 workflow_parallel_frame_prediction/DataPostprocess/parameters_kitti_evaluate.dat
delete mode 100755 workflow_parallel_frame_prediction/DataPostprocess/prednet.py
delete mode 100755 workflow_parallel_frame_prediction/DataPreprocess/Stager_devel_N_24_process_netCDF.sh
delete mode 100755 workflow_parallel_frame_prediction/DataPreprocess/external_function.py
delete mode 100644 workflow_parallel_frame_prediction/DataPreprocess/mpi_split_data_multi_years.py
delete mode 100755 workflow_parallel_frame_prediction/README.md
delete mode 100755 workflow_parallel_frame_prediction/Training/data_utils.py
delete mode 100644 workflow_parallel_frame_prediction/Training/devel_horovodJob.sh
delete mode 100644 workflow_parallel_frame_prediction/Training/evaluate_multistep.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/bin/f2py
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/__pycache__/site.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/easy-install.pth
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/PKG-INFO
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/SOURCES.txt
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/dependency_links.txt
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/not-zip-safe
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/requires.txt
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/top_level.txt
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__init__.py
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/__init__.cpython-36.pyc
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/helpers.cpython-36.pyc
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/hickle.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/hickle_legacy.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/hickle_legacy2.cpython-36.pyc
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/lookup.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/helpers.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle_legacy.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle_legacy2.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__init__.py
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/__init__.cpython-36.pyc
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_astropy.cpython-36.pyc
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_numpy.cpython-36.pyc
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_pandas.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_python.cpython-36.pyc
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_python3.cpython-36.pyc
delete mode 100644 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_scipy.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_astropy.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_numpy.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_pandas.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_python.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_python3.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_scipy.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/lookup.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__init__.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/__init__.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_astropy.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_hickle.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_hickle_helpers.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_legacy_load.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_scipy.cpython-36.pyc
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_astropy.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_hickle.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_hickle_helpers.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_legacy_load.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_scipy.py
delete mode 100755 workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/site.py
delete mode 100644 workflow_parallel_frame_prediction/Training/horovodJob.sh
delete mode 100755 workflow_parallel_frame_prediction/Training/keras_utils.py
delete mode 100755 workflow_parallel_frame_prediction/Training/kitti_settings.py
delete mode 100755 workflow_parallel_frame_prediction/Training/kitti_train_horovod.py
delete mode 100644 workflow_parallel_frame_prediction/Training/minMaxExtractor.py
delete mode 100755 workflow_parallel_frame_prediction/Training/prednet.py
delete mode 100644 workflow_parallel_frame_prediction/Training/process_netCDF.py
delete mode 100644 workflow_parallel_frame_prediction/Workflow.png
delete mode 100755 workflow_parallel_frame_prediction/__init__.py
delete mode 100644 workflow_parallel_frame_prediction/main.sh
delete mode 100755 workflow_parallel_frame_prediction/setup.py
diff --git a/video_prediction_savp/HPC_scripts/hyperparam_setup.sh b/video_prediction_savp/HPC_scripts/hyperparam_setup.sh
deleted file mode 100644
index 34894da8..00000000
--- a/video_prediction_savp/HPC_scripts/hyperparam_setup.sh
+++ /dev/null
@@ -1,12 +0,0 @@
-#!/usr/bin/env bash
-
-# for choosing the model convLSTM vae mcnet savp
-export model=convLSTM
-export model_hparams=../hparams/era5/${model}/model_hparams.json
-
-#create a subfolder with create time and user names, which can be consider as hyperparameter tunning folder. This can avoid overwrite the prevoius trained model using differ#ent hypermeters
-export hyperdir="$(date +"%Y%m%dT%H%M")_"$USER""
-
-echo "model: ${model}"
-echo "hparams: ${model_hparams}"
-echo "experiment dir: ${hyperdir}"
diff --git a/video_prediction_savp/HPC_scripts/reset_dirs.sh b/video_prediction_savp/HPC_scripts/reset_dirs.sh
deleted file mode 100644
index 8de5247e..00000000
--- a/video_prediction_savp/HPC_scripts/reset_dirs.sh
+++ /dev/null
@@ -1,11 +0,0 @@
-#!/usr/bin/env bash
-
-sed -i "s|source_dir=.*|source_dir=${SAVE_DIR}preprocessedData/|g" DataPreprocess_to_tf.sh
-sed -i "s|destination_dir=.*|destination_dir=${SAVE_DIR}preprocessedData/|g" DataPreprocess_to_tf.sh
-
-sed -i "s|source_dir=.*|source_dir=${SAVE_DIR}preprocessedData/|g" train_era5.sh
-sed -i "s|destination_dir=.*|destination_dir=${SAVE_DIR}models/|g" train_era5.sh
-
-sed -i "s|source_dir=.*|source_dir=${SAVE_DIR}preprocessedData/|g" generate_era5.sh
-sed -i "s|checkpoint_dir=.*|checkpoint_dir=${SAVE_DIR}models/|g" generate_era5.sh
-sed -i "s|results_dir=.*|results_dir=${SAVE_DIR}results/|g" generate_era5.sh
diff --git a/video_prediction_savp/bash/download_and_preprocess_dataset.sh b/video_prediction_savp/bash/download_and_preprocess_dataset.sh
deleted file mode 100644
index 5779c2b7..00000000
--- a/video_prediction_savp/bash/download_and_preprocess_dataset.sh
+++ /dev/null
@@ -1,88 +0,0 @@
-#!/usr/bin/env bash
-
-# exit if any command fails
-set -e
-
-if [ "$#" -eq 2 ]; then
- if [ $1 = "bair" ]; then
- echo "IMAGE_SIZE argument is only applicable to kth dataset" >&2
- exit 1
- fi
-elif [ "$#" -ne 1 ]; then
- echo "Usage: $0 DATASET_NAME [IMAGE_SIZE]" >&2
- exit 1
-fi
-if [ $1 = "bair" ]; then
- TARGET_DIR=./data/bair
- mkdir -p ${TARGET_DIR}
- TAR_FNAME=bair_robot_pushing_dataset_v0.tar
- URL=http://rail.eecs.berkeley.edu/datasets/${TAR_FNAME}
- echo "Downloading '$1' dataset (this takes a while)"
- #wget ${URL} -O ${TARGET_DIR}/${TAR_FNAME} Bing: on MacOS system , use curl instead of wget
- curl ${URL} -O ${TARGET_DIR}/${TAR_FNAME}
- tar -xvf ${TARGET_DIR}/${TAR_FNAME} --strip-components=1 -C ${TARGET_DIR}
- rm ${TARGET_DIR}/${TAR_FNAME}
- mkdir -p ${TARGET_DIR}/val
- # reserve a fraction of the training set for validation
- mv ${TARGET_DIR}/train/traj_256_to_511.tfrecords ${TARGET_DIR}/val/
-elif [ $1 = "kth" ]; then
- if [ "$#" -eq 2 ]; then
- IMAGE_SIZE=$2
- TARGET_DIR=./data/kth_${IMAGE_SIZE}
- else
- IMAGE_SIZE=64
- TARGET_DIR=./data/kth
- fi
- echo ${TARGET_DIR} ${IMAGE_SIZE}
- mkdir -p ${TARGET_DIR}
- mkdir -p ${TARGET_DIR}/raw
- echo "Downloading '$1' dataset (this takes a while)"
- # TODO Bing: for save time just use walking, need to change back if all the data are needed
- #for ACTION in walking jogging running boxing handwaving handclapping; do
-# for ACTION in walking; do
-# echo "Action: '$ACTION' "
-# ZIP_FNAME=${ACTION}.zip
-# URL=http://www.nada.kth.se/cvap/actions/${ZIP_FNAME}
-# # wget ${URL} -O ${TARGET_DIR}/raw/${ZIP_FNAME}
-# echo "Start downloading action '$ACTION' ULR '$URL' "
-# curl ${URL} -O ${TARGET_DIR}/raw/${ZIP_FNAME}
-# unzip ${TARGET_DIR}/raw/${ZIP_FNAME} -d ${TARGET_DIR}/raw/${ACTION}
-# echo "Action '$ACTION' data download and unzip "
-# done
- FRAME_RATE=25
-# mkdir -p ${TARGET_DIR}/processed
-# # download files with metadata specifying the subsequences
-# TAR_FNAME=kth_meta.tar.gz
-# URL=http://rail.eecs.berkeley.edu/models/savp/data/${TAR_FNAME}
-# echo "Downloading '${TAR_FNAME}' ULR '$URL' "
-# #wget ${URL} -O ${TARGET_DIR}/processed/${TAR_FNAME}
-# curl ${URL} -O ${TARGET_DIR}/processed/${TAR_FNAME}
-# tar -xzvf ${TARGET_DIR}/processed/${TAR_FNAME} --strip 1 -C ${TARGET_DIR}/processed
- # convert the videos into sequence of downscaled images
- echo "Processing '$1' dataset"
- #TODO Bing, just use walking for test
- #for ACTION in walking jogging running boxing handwaving handclapping; do
- #Todo Bing: remove the comments below after testing
- for ACTION in walking running; do
- for VIDEO_FNAME in ${TARGET_DIR}/raw/${ACTION}/*.avi; do
- FNAME=$(basename ${VIDEO_FNAME})
- FNAME=${FNAME%_uncomp.avi}
- echo "FNAME '$FNAME' "
- # sometimes the directory is not created, so try until it is
- while [ ! -d "${TARGET_DIR}/processed/${ACTION}/${FNAME}" ]; do
- mkdir -p ${TARGET_DIR}/processed/${ACTION}/${FNAME}
- done
- ffmpeg -i ${VIDEO_FNAME} -r ${FRAME_RATE} -f image2 -s ${IMAGE_SIZE}x${IMAGE_SIZE} \
- ${TARGET_DIR}/processed/${ACTION}/${FNAME}/image-%03d_${IMAGE_SIZE}x${IMAGE_SIZE}.png
- done
- done
- python video_prediction/datasets/kth_dataset.py ${TARGET_DIR}/processed ${TARGET_DIR} ${IMAGE_SIZE}
- rm -rf ${TARGET_DIR}/raw
- rm -rf ${TARGET_DIR}/processed
-else
- echo "Invalid dataset name: '$1' (choose from 'bair', 'kth')" >&2
- exit 1
-fi
-echo "Succesfully finished downloadi\
-
-ng and preprocessing dataset '$1'"
diff --git a/video_prediction_savp/bash/download_and_preprocess_dataset_era5.sh b/video_prediction_savp/bash/download_and_preprocess_dataset_era5.sh
deleted file mode 100644
index eacc0180..00000000
--- a/video_prediction_savp/bash/download_and_preprocess_dataset_era5.sh
+++ /dev/null
@@ -1,127 +0,0 @@
-#!/usr/bin/env bash
-
-# exit if any command fails
-set -e
-
-
-#if [ "$#" -eq 2 ]; then
-# if [ $1 = "bair" ]; then
-# echo "IMAGE_SIZE argument is only applicable to kth dataset" >&2
-# exit 1
-# fi
-#elif [ "$#" -ne 1 ]; then
-# echo "Usage: $0 DATASET_NAME [IMAGE_SIZE]" >&2
-# exit 1
-#fi
-#if [ $1 = "bair" ]; then
-# TARGET_DIR=./data/bair
-# mkdir -p ${TARGET_DIR}
-# TAR_FNAME=bair_robot_pushing_dataset_v0.tar
-# URL=http://rail.eecs.berkeley.edu/datasets/${TAR_FNAME}
-# echo "Downloading '$1' dataset (this takes a while)"
-# #wget ${URL} -O ${TARGET_DIR}/${TAR_FNAME} Bing: on MacOS system , use curl instead of wget
-# curl ${URL} -O ${TARGET_DIR}/${TAR_FNAME}
-# tar -xvf ${TARGET_DIR}/${TAR_FNAME} --strip-components=1 -C ${TARGET_DIR}
-# rm ${TARGET_DIR}/${TAR_FNAME}
-# mkdir -p ${TARGET_DIR}/val
-# # reserve a fraction of the training set for validation
-# mv ${TARGET_DIR}/train/traj_256_to_511.tfrecords ${TARGET_DIR}/val/
-#elif [ $1 = "kth" ]; then
-# if [ "$#" -eq 2 ]; then
-# IMAGE_SIZE=$2
-# TARGET_DIR=./data/kth_${IMAGE_SIZE}
-# else
-# IMAGE_SIZE=64
-# fi
-# echo ${TARGET_DIR} ${IMAGE_SIZE}
-# mkdir -p ${TARGET_DIR}
-# mkdir -p ${TARGET_DIR}/raw
-# echo "Downloading '$1' dataset (this takes a while)"
- # TODO Bing: for save time just use walking, need to change back if all the data are needed
- #for ACTION in walking jogging running boxing handwaving handclapping; do
-# for ACTION in walking; do
-# echo "Action: '$ACTION' "
-# ZIP_FNAME=${ACTION}.zip
-# URL=http://www.nada.kth.se/cvap/actions/${ZIP_FNAME}
-# # wget ${URL} -O ${TARGET_DIR}/raw/${ZIP_FNAME}
-# echo "Start downloading action '$ACTION' ULR '$URL' "
-# curl ${URL} -O ${TARGET_DIR}/raw/${ZIP_FNAME}
-# unzip ${TARGET_DIR}/raw/${ZIP_FNAME} -d ${TARGET_DIR}/raw/${ACTION}
-# echo "Action '$ACTION' data download and unzip "
-# done
-# FRAME_RATE=25
-# mkdir -p ${TARGET_DIR}/processed
-# # download files with metadata specifying the subsequences
-# TAR_FNAME=kth_meta.tar.gz
-# URL=http://rail.eecs.berkeley.edu/models/savp/data/${TAR_FNAME}
-# echo "Downloading '${TAR_FNAME}' ULR '$URL' "
-# #wget ${URL} -O ${TARGET_DIR}/processed/${TAR_FNAME}
-# curl ${URL} -O ${TARGET_DIR}/processed/${TAR_FNAME}
-# tar -xzvf ${TARGET_DIR}/processed/${TAR_FNAME} --strip 1 -C ${TARGET_DIR}/processed
- # convert the videos into sequence of downscaled images
-# echo "Processing '$1' dataset"
-# #TODO Bing, just use walking for test
-# #for ACTION in walking jogging running boxing handwaving handclapping; do
-# #Todo Bing: remove the comments below after testing
-# for ACTION in walking; do
-# for VIDEO_FNAME in ${TARGET_DIR}/raw/${ACTION}/*.avi; do
-# FNAME=$(basename ${VIDEO_FNAME})
-# FNAME=${FNAME%_uncomp.avi}
-# echo "FNAME '$FNAME' "
-# # sometimes the directory is not created, so try until it is
-# while [ ! -d "${TARGET_DIR}/processed/${ACTION}/${FNAME}" ]; do
-# mkdir -p ${TARGET_DIR}/processed/${ACTION}/${FNAME}
-# done
-# ffmpeg -i ${VIDEO_FNAME} -r ${FRAME_RATE} -f image2 -s ${IMAGE_SIZE}x${IMAGE_SIZE} \
-# ${TARGET_DIR}/processed/${ACTION}/${FNAME}/image-%03d_${IMAGE_SIZE}x${IMAGE_SIZE}.png
-# done
-# done
-# python video_prediction/datasets/kth_dataset.py ${TARGET_DIR}/processed ${TARGET_DIR} ${IMAGE_SIZE}
-# rm -rf ${TARGET_DIR}/raw
-# rm -rf ${TARGET_DIR}/processed
-
-while [[ $# -gt 0 ]] #of the number of passed argument is greater than 0
-do
-key="$1"
-case $key in
- -d|--data)
- DATA="$2"
- shift
- shift
- ;;
- -i|--input_dir)
- INPUT_DIR="$2"
- shift
- shift
- ;;
- -o|--output_dir)
- OUTPUT_DIR="$2"
- shift
- shift
- ;;
-esac
-done
-
-echo "DATA = ${DATA} "
-
-echo "OUTPUT_DIRECTORY = ${OUTPUT_DIR}"
-
-if [ -d $INPUT_DIR ]; then
- echo "INPUT DIRECTORY = ${INPUT_DIR}"
-
-else
- echo "INPUT DIRECTORY '$INPUT_DIR' DOES NOT EXIST"
- exit 1
-fi
-
-
-if [ $DATA = "era5" ]; then
-
- mkdir -p ${OUTPUT_DIR}
- python video_prediction/datasets/era5_dataset.py $INPUT_DIR ${OUTPUT_DIR}
-else
- echo "dataset name: '$DATA' (choose from 'era5')" >&2
- exit 1
-fi
-
-echo "Succesfully finished downloading and preprocessing dataset '$DATA' "
\ No newline at end of file
diff --git a/video_prediction_savp/bash/download_and_preprocess_dataset_v1.sh b/video_prediction_savp/bash/download_and_preprocess_dataset_v1.sh
deleted file mode 100644
index 3541b4a5..00000000
--- a/video_prediction_savp/bash/download_and_preprocess_dataset_v1.sh
+++ /dev/null
@@ -1,86 +0,0 @@
-#!/usr/bin/env bash
-
-# exit if any command fails
-set -e
-
-if [ "$#" -eq 2 ]; then
- if [ $1 = "bair" ]; then
- echo "IMAGE_SIZE argument is only applicable to kth dataset" >&2
- exit 1
- fi
-elif [ "$#" -ne 1 ]; then
- echo "Usage: $0 DATASET_NAME [IMAGE_SIZE]" >&2
- exit 1
-fi
-if [ $1 = "bair" ]; then
- TARGET_DIR=./data/bair
- mkdir -p ${TARGET_DIR}
- TAR_FNAME=bair_robot_pushing_dataset_v0.tar
- URL=http://rail.eecs.berkeley.edu/datasets/${TAR_FNAME}
- echo "Downloading '$1' dataset (this takes a while)"
- #wget ${URL} -O ${TARGET_DIR}/${TAR_FNAME} Bing: on MacOS system , use curl instead of wget
- curl ${URL} -O ${TARGET_DIR}/${TAR_FNAME}
- tar -xvf ${TARGET_DIR}/${TAR_FNAME} --strip-components=1 -C ${TARGET_DIR}
- rm ${TARGET_DIR}/${TAR_FNAME}
- mkdir -p ${TARGET_DIR}/val
- # reserve a fraction of the training set for validation
- mv ${TARGET_DIR}/train/traj_256_to_511.tfrecords ${TARGET_DIR}/val/
-elif [ $1 = "kth" ]; then
- if [ "$#" -eq 2 ]; then
- IMAGE_SIZE=$2
- TARGET_DIR=./data/kth_${IMAGE_SIZE}
- else
- IMAGE_SIZE=64
- TARGET_DIR=./data/kth
- fi
- echo ${TARGET_DIR} ${IMAGE_SIZE}
- mkdir -p ${TARGET_DIR}
- mkdir -p ${TARGET_DIR}/raw
- echo "Downloading '$1' dataset (this takes a while)"
- # TODO Bing: for save time just use walking, need to change back if all the data are needed
- #for ACTION in walking jogging running boxing handwaving handclapping; do
-# for ACTION in walking; do
-# echo "Action: '$ACTION' "
-# ZIP_FNAME=${ACTION}.zip
-# URL=http://www.nada.kth.se/cvap/actions/${ZIP_FNAME}
-# # wget ${URL} -O ${TARGET_DIR}/raw/${ZIP_FNAME}
-# echo "Start downloading action '$ACTION' ULR '$URL' "
-# curl ${URL} -O ${TARGET_DIR}/raw/${ZIP_FNAME}
-# unzip ${TARGET_DIR}/raw/${ZIP_FNAME} -d ${TARGET_DIR}/raw/${ACTION}
-# echo "Action '$ACTION' data download and unzip "
-# done
- FRAME_RATE=25
-# mkdir -p ${TARGET_DIR}/processed
-# # download files with metadata specifying the subsequences
-# TAR_FNAME=kth_meta.tar.gz
-# URL=http://rail.eecs.berkeley.edu/models/savp/data/${TAR_FNAME}
-# echo "Downloading '${TAR_FNAME}' ULR '$URL' "
-# #wget ${URL} -O ${TARGET_DIR}/processed/${TAR_FNAME}
-# curl ${URL} -O ${TARGET_DIR}/processed/${TAR_FNAME}
-# tar -xzvf ${TARGET_DIR}/processed/${TAR_FNAME} --strip 1 -C ${TARGET_DIR}/processed
- # convert the videos into sequence of downscaled images
- echo "Processing '$1' dataset"
- #TODO Bing, just use walking for test
- #for ACTION in walking jogging running boxing handwaving handclapping; do
- #Todo Bing: remove the comments below after testing
- for ACTION in walking; do
- for VIDEO_FNAME in ${TARGET_DIR}/raw/${ACTION}/*.avi; do
- FNAME=$(basename ${VIDEO_FNAME})
- FNAME=${FNAME%_uncomp.avi}
- echo "FNAME '$FNAME' "
- # sometimes the directory is not created, so try until it is
- while [ ! -d "${TARGET_DIR}/processed/${ACTION}/${FNAME}" ]; do
- mkdir -p ${TARGET_DIR}/processed/${ACTION}/${FNAME}
- done
- ffmpeg -i ${VIDEO_FNAME} -r ${FRAME_RATE} -f image2 -s ${IMAGE_SIZE}x${IMAGE_SIZE} \
- ${TARGET_DIR}/processed/${ACTION}/${FNAME}/image-%03d_${IMAGE_SIZE}x${IMAGE_SIZE}.png
- done
- done
- python video_prediction/datasets/kth_dataset.py ${TARGET_DIR}/processed ${TARGET_DIR} ${IMAGE_SIZE}
- rm -rf ${TARGET_DIR}/raw
- rm -rf ${TARGET_DIR}/processed
-else
- echo "Invalid dataset name: '$1' (choose from 'bair', 'kth')" >&2
- exit 1
-fi
-echo "Succesfully finished downloading and preprocessing dataset '$1'"
diff --git a/video_prediction_savp/bash/workflow_era5.sh b/video_prediction_savp/bash/workflow_era5.sh
deleted file mode 100755
index 01d16bfd..00000000
--- a/video_prediction_savp/bash/workflow_era5.sh
+++ /dev/null
@@ -1,92 +0,0 @@
-#!/usr/bin/env bash
-set -e
-#
-#MODEL=savp
-##train_mode: end_to_end, pre_trained
-#TRAIN_MODE=end_to_end
-#EXP_NAME=era5_size_64_64_3_3t_norm
-
-MODEL=$1
-TRAIN_MODE=$2
-EXP_NAME=$3
-RETRAIN=1 #if we continue training the model or using the existing end-to-end model, 1 means continue training, and 1 means use the existing one
-DATA_ETL_DIR=/p/scratch/deepacf/${USER}/
-DATA_EXTRA_DIR=${DATA_ETL_DIR}/extractedData/${EXP_NAME}
-DATA_PREPROCESS_DIR=${DATA_ETL_DIR}/preprocessedData/${EXP_NAME}
-DATA_PREPROCESS_TF_DIR=./data/${EXP_NAME}
-RESULTS_OUTPUT_DIR=./results_test_samples/${EXP_NAME}/${TRAIN_MODE}/
-
-if [ $MODEL==savp ]; then
- method_dir=ours_savp
-elif [ $MODEL==gan ]; then
- method_dir=ours_gan
-elif [ $MODEL==vae ]; then
- method_dir=ours_vae
-else
- echo "model does not exist" 2>&1
- exit 1
-fi
-
-if [ "$TRAIN_MODE" == pre_trained ]; then
- TRAIN_OUTPUT_DIR=./pretrained_models/kth/${method_dir}
-else
- TRAIN_OUTPUT_DIR=./logs/${EXP_NAME}/${TRAIN_MODE}
-fi
-
-CHECKPOINT_DIR=${TRAIN_OUTPUT_DIR}/${method_dir}
-
-echo "===========================WORKFLOW SETUP===================="
-echo "Model ${MODEL}"
-echo "TRAIN MODE ${TRAIN_MODE}"
-echo "Method_dir ${method_dir}"
-echo "DATA_ETL_DIR ${DATA_ETL_DIR}"
-echo "DATA_EXTRA_DIR ${DATA_EXTRA_DIR}"
-echo "DATA_PREPROCESS_DIR ${DATA_PREPROCESS_DIR}"
-echo "DATA_PREPROCESS_TF_DIR ${DATA_PREPROCESS_TF_DIR}"
-echo "TRAIN_OUTPUT_DIR ${TRAIN_OUTPUT_DIR}"
-echo "============================================================="
-
-##############Datat Preprocessing################
-#To hkl data
-if [ -d "$DATA_PREPROCESS_DIR" ]; then
- echo "The Preprocessed Data (.hkl ) exist"
-else
- python ../workflow_video_prediction/DataPreprocess/benchmark/mpi_stager_v2_process_netCDF.py \
- --input_dir ${DATA_EXTRA_DIR} --destination_dir ${DATA_PREPROCESS_DIR}
-fi
-
-#Change the .hkl data to .tfrecords files
-if [ -d "$DATA_PREPROCESS_TF_DIR" ]
-then
- echo "Step2: The Preprocessed Data (tf.records) exist"
-else
- echo "Step2: start, hkl. files to tf.records"
- python ./video_prediction/datasets/era5_dataset_v2.py --source_dir ${DATA_PREPROCESS_DIR}/splits \
- --destination_dir ${DATA_PREPROCESS_TF_DIR}
- echo "Step2: finish"
-fi
-
-#########Train##########################
-if [ "$TRAIN_MODE" == "pre_trained" ]; then
- echo "step3: Using kth pre_trained model"
-elif [ "$TRAIN_MODE" == "end_to_end" ]; then
- echo "step3: End-to-end training"
- if [ "$RETRAIN" == 1 ]; then
- echo "Using the existing end-to-end model"
- else
- echo "Step3: Training Starts "
- python ./scripts/train_v2.py --input_dir $DATA_PREPROCESS_TF_DIR --dataset era5 \
- --model ${MODEL} --model_hparams_dict hparams/kth/${method_dir}/model_hparams.json \
- --output_dir ${TRAIN_OUTPUT_DIR} --checkpoint ${CHECKPOINT_DIR_DIR}
- echo "Training ends "
- fi
-else
- echo "TRAIN_MODE is end_to_end or pre_trained"
- exit 1
-fi
-
-#########Generate results#################
-echo "Step4: Postprocessing start"
-python ./scripts/generate_transfer_learning_finetune.py --input_dir ${DATA_PREPROCESS_TF_DIR} \
---dataset_hparams sequence_length=20 --checkpoint ${CHECKPOINT_DIR_DIR} --mode test --results_dir ${RESULTS_OUTPUT_DIR} \
---batch_size 4 --dataset era5
\ No newline at end of file
diff --git a/video_prediction_savp/bash/workflow_era5_macOS.sh b/video_prediction_savp/bash/workflow_era5_macOS.sh
deleted file mode 100755
index 1a6ebef3..00000000
--- a/video_prediction_savp/bash/workflow_era5_macOS.sh
+++ /dev/null
@@ -1,93 +0,0 @@
-#!/usr/bin/env bash
-set -e
-#
-#MODEL=savp
-##train_mode: end_to_end, pre_trained
-#TRAIN_MODE=end_to_end
-#EXP_NAME=era5_size_64_64_3_3t_norm
-
-MODEL=$1
-TRAIN_MODE=$2
-EXP_NAME=$3
-RETRAIN=1 #if we continue training the model or using the existing end-to-end model, 1 means continue training, and 1 means use the existing one
-DATA_ETL_DIR=/p/scratch/deepacf/${USER}/
-DATA_ETL_DIR=/p/scratch/deepacf/${USER}/
-DATA_EXTRA_DIR=${DATA_ETL_DIR}/extractedData/${EXP_NAME}
-DATA_PREPROCESS_DIR=${DATA_ETL_DIR}/preprocessedData/${EXP_NAME}
-DATA_PREPROCESS_TF_DIR=./data/${EXP_NAME}
-RESULTS_OUTPUT_DIR=./results_test_samples/${EXP_NAME}/${TRAIN_MODE}/
-
-if [ $MODEL==savp ]; then
- method_dir=ours_savp
-elif [ $MODEL==gan ]; then
- method_dir=ours_gan
-elif [ $MODEL==vae ]; then
- method_dir=ours_vae
-else
- echo "model does not exist" 2>&1
- exit 1
-fi
-
-if [ "$TRAIN_MODE" == pre_trained ]; then
- TRAIN_OUTPUT_DIR=./pretrained_models/kth/${method_dir}
-else
- TRAIN_OUTPUT_DIR=./logs/${EXP_NAME}/${TRAIN_MODE}
-fi
-
-CHECKPOINT_DIR=${TRAIN_OUTPUT_DIR}/${method_dir}
-
-echo "===========================WORKFLOW SETUP===================="
-echo "Model ${MODEL}"
-echo "TRAIN MODE ${TRAIN_MODE}"
-echo "Method_dir ${method_dir}"
-echo "DATA_ETL_DIR ${DATA_ETL_DIR}"
-echo "DATA_EXTRA_DIR ${DATA_EXTRA_DIR}"
-echo "DATA_PREPROCESS_DIR ${DATA_PREPROCESS_DIR}"
-echo "DATA_PREPROCESS_TF_DIR ${DATA_PREPROCESS_TF_DIR}"
-echo "TRAIN_OUTPUT_DIR ${TRAIN_OUTPUT_DIR}"
-echo "============================================================="
-
-##############Datat Preprocessing################
-#To hkl data
-#if [ -d "$DATA_PREPROCESS_DIR" ]; then
-# echo "The Preprocessed Data (.hkl ) exist"
-#else
-# python ../workflow_video_prediction/DataPreprocess/benchmark/mpi_stager_v2_process_netCDF.py \
-# --input_dir ${DATA_EXTRA_DIR} --destination_dir ${DATA_PREPROCESS_DIR}
-#fi
-
-####Change the .hkl data to .tfrecords files
-if [ -d "$DATA_PREPROCESS_TF_DIR" ]
-then
- echo "Step2: The Preprocessed Data (tf.records) exist"
-else
- echo "Step2: start, hkl. files to tf.records"
- python ./video_prediction/datasets/era5_dataset_v2.py --source_dir ${DATA_PREPROCESS_DIR}/splits \
- --destination_dir ${DATA_PREPROCESS_TF_DIR}
- echo "Step2: finish"
-fi
-
-#########Train##########################
-if [ "$TRAIN_MODE" == "pre_trained" ]; then
- echo "step3: Using kth pre_trained model"
-elif [ "$TRAIN_MODE" == "end_to_end" ]; then
- echo "step3: End-to-end training"
- if [ "$RETRAIN" == 1 ]; then
- echo "Using the existing end-to-end model"
- else
- echo "Training Starts "
- python ./scripts/train_v2.py --input_dir $DATA_PREPROCESS_TF_DIR --dataset era5 \
- --model ${MODEL} --model_hparams_dict hparams/kth/${method_dir}/model_hparams.json \
- --output_dir ${TRAIN_OUTPUT_DIR} --checkpoint ${CHECKPOINT_DIR}
- echo "Training ends "
- fi
-else
- echo "TRAIN_MODE is end_to_end or pre_trained"
- exit 1
-fi
-
-#########Generate results#################
-echo "Step4: Postprocessing start"
-python ./scripts/generate_transfer_learning_finetune.py --input_dir ${DATA_PREPROCESS_TF_DIR} \
---dataset_hparams sequence_length=20 --checkpoint ${CHECKPOINT_DIR} --mode test --results_dir ${RESULTS_OUTPUT_DIR} \
---batch_size 4 --dataset era5
diff --git a/video_prediction_savp/bash/workflow_era5_zam347.sh b/video_prediction_savp/bash/workflow_era5_zam347.sh
deleted file mode 100755
index ffe7209b..00000000
--- a/video_prediction_savp/bash/workflow_era5_zam347.sh
+++ /dev/null
@@ -1,93 +0,0 @@
-#!/usr/bin/env bash
-set -e
-#
-#MODEL=savp
-##train_mode: end_to_end, pre_trained
-#TRAIN_MODE=end_to_end
-#EXP_NAME=era5_size_64_64_3_3t_norm
-
-MODEL=$1
-TRAIN_MODE=$2
-EXP_NAME=$3
-RETRAIN=1 #if we continue training the model or using the existing end-to-end model, 1 means continue training, and 1 means use the existing one
-DATA_ETL_DIR=/home/${USER}/
-DATA_ETL_DIR=/p/scratch/deepacf/${USER}/
-DATA_EXTRA_DIR=${DATA_ETL_DIR}/extractedData/${EXP_NAME}
-DATA_PREPROCESS_DIR=${DATA_ETL_DIR}/preprocessedData/${EXP_NAME}
-DATA_PREPROCESS_TF_DIR=./data/${EXP_NAME}
-RESULTS_OUTPUT_DIR=./results_test_samples/${EXP_NAME}/${TRAIN_MODE}/
-
-if [ $MODEL==savp ]; then
- method_dir=ours_savp
-elif [ $MODEL==gan ]; then
- method_dir=ours_gan
-elif [ $MODEL==vae ]; then
- method_dir=ours_vae
-else
- echo "model does not exist" 2>&1
- exit 1
-fi
-
-if [ "$TRAIN_MODE" == pre_trained ]; then
- TRAIN_OUTPUT_DIR=./pretrained_models/kth/${method_dir}
-else
- TRAIN_OUTPUT_DIR=./logs/${EXP_NAME}/${TRAIN_MODE}
-fi
-
-CHECKPOINT_DIR=${TRAIN_OUTPUT_DIR}/${method_dir}
-
-echo "===========================WORKFLOW SETUP===================="
-echo "Model ${MODEL}"
-echo "TRAIN MODE ${TRAIN_MODE}"
-echo "Method_dir ${method_dir}"
-echo "DATA_ETL_DIR ${DATA_ETL_DIR}"
-echo "DATA_EXTRA_DIR ${DATA_EXTRA_DIR}"
-echo "DATA_PREPROCESS_DIR ${DATA_PREPROCESS_DIR}"
-echo "DATA_PREPROCESS_TF_DIR ${DATA_PREPROCESS_TF_DIR}"
-echo "TRAIN_OUTPUT_DIR ${TRAIN_OUTPUT_DIR}"
-echo "============================================================="
-
-##############Datat Preprocessing################
-#To hkl data
-#if [ -d "$DATA_PREPROCESS_DIR" ]; then
-# echo "The Preprocessed Data (.hkl ) exist"
-#else
-# python ../workflow_video_prediction/DataPreprocess/benchmark/mpi_stager_v2_process_netCDF.py \
-# --input_dir ${DATA_EXTRA_DIR} --destination_dir ${DATA_PREPROCESS_DIR}
-#fi
-
-####Change the .hkl data to .tfrecords files
-if [ -d "$DATA_PREPROCESS_TF_DIR" ]
-then
- echo "Step2: The Preprocessed Data (tf.records) exist"
-else
- echo "Step2: start, hkl. files to tf.records"
- python ./video_prediction/datasets/era5_dataset_v2.py --source_dir ${DATA_PREPROCESS_DIR}/splits \
- --destination_dir ${DATA_PREPROCESS_TF_DIR}
- echo "Step2: finish"
-fi
-
-#########Train##########################
-if [ "$TRAIN_MODE" == "pre_trained" ]; then
- echo "step3: Using kth pre_trained model"
-elif [ "$TRAIN_MODE" == "end_to_end" ]; then
- echo "step3: End-to-end training"
- if [ "$RETRAIN" == 1 ]; then
- echo "Using the existing end-to-end model"
- else
- echo "Training Starts "
- python ./scripts/train_v2.py --input_dir $DATA_PREPROCESS_TF_DIR --dataset era5 \
- --model ${MODEL} --model_hparams_dict hparams/kth/${method_dir}/model_hparams.json \
- --output_dir ${TRAIN_OUTPUT_DIR} --checkpoint ${CHECKPOINT_DIR}
- echo "Training ends "
- fi
-else
- echo "TRAIN_MODE is end_to_end or pre_trained"
- exit 1
-fi
-
-#########Generate results#################
-echo "Step4: Postprocessing start"
-python ./scripts/generate_transfer_learning_finetune.py --input_dir ${DATA_PREPROCESS_TF_DIR} \
---dataset_hparams sequence_length=20 --checkpoint ${CHECKPOINT_DIR} --mode test --results_dir ${RESULTS_OUTPUT_DIR} \
---batch_size 4 --dataset era5
diff --git a/video_prediction_savp/helper/helper.py b/video_prediction_savp/deprecate/helper/helper.py
similarity index 100%
rename from video_prediction_savp/helper/helper.py
rename to video_prediction_savp/deprecate/helper/helper.py
diff --git a/video_prediction_savp/pretrained_models/download_model.sh b/video_prediction_savp/deprecate/pretrained_models/download_model.sh
similarity index 100%
rename from video_prediction_savp/pretrained_models/download_model.sh
rename to video_prediction_savp/deprecate/pretrained_models/download_model.sh
diff --git a/video_prediction_savp/scripts/combine_results.py b/video_prediction_savp/deprecate/scripts/combine_results.py
similarity index 100%
rename from video_prediction_savp/scripts/combine_results.py
rename to video_prediction_savp/deprecate/scripts/combine_results.py
diff --git a/video_prediction_savp/scripts/evaluate.py b/video_prediction_savp/deprecate/scripts/evaluate.py
similarity index 100%
rename from video_prediction_savp/scripts/evaluate.py
rename to video_prediction_savp/deprecate/scripts/evaluate.py
diff --git a/video_prediction_savp/scripts/evaluate_all.sh b/video_prediction_savp/deprecate/scripts/evaluate_all.sh
similarity index 100%
rename from video_prediction_savp/scripts/evaluate_all.sh
rename to video_prediction_savp/deprecate/scripts/evaluate_all.sh
diff --git a/video_prediction_savp/scripts/generate_all.sh b/video_prediction_savp/deprecate/scripts/generate_all.sh
similarity index 100%
rename from video_prediction_savp/scripts/generate_all.sh
rename to video_prediction_savp/deprecate/scripts/generate_all.sh
diff --git a/video_prediction_savp/scripts/generate_orig.py b/video_prediction_savp/deprecate/scripts/generate_orig.py
similarity index 100%
rename from video_prediction_savp/scripts/generate_orig.py
rename to video_prediction_savp/deprecate/scripts/generate_orig.py
diff --git a/video_prediction_savp/scripts/plot_results.py b/video_prediction_savp/deprecate/scripts/plot_results.py
similarity index 100%
rename from video_prediction_savp/scripts/plot_results.py
rename to video_prediction_savp/deprecate/scripts/plot_results.py
diff --git a/video_prediction_savp/scripts/plot_results_all.sh b/video_prediction_savp/deprecate/scripts/plot_results_all.sh
similarity index 100%
rename from video_prediction_savp/scripts/plot_results_all.sh
rename to video_prediction_savp/deprecate/scripts/plot_results_all.sh
diff --git a/video_prediction_savp/scripts/train.py b/video_prediction_savp/deprecate/scripts/train.py
similarity index 100%
rename from video_prediction_savp/scripts/train.py
rename to video_prediction_savp/deprecate/scripts/train.py
diff --git a/video_prediction_savp/scripts/train_all.sh b/video_prediction_savp/deprecate/scripts/train_all.sh
similarity index 100%
rename from video_prediction_savp/scripts/train_all.sh
rename to video_prediction_savp/deprecate/scripts/train_all.sh
diff --git a/video_prediction_savp/docs/presentation/presentation.md b/video_prediction_savp/docs/presentation/presentation.md
deleted file mode 100644
index d4923908..00000000
--- a/video_prediction_savp/docs/presentation/presentation.md
+++ /dev/null
@@ -1,5 +0,0 @@
-This is the presentation materials for VP group
-
-
-## 2020-03-01 - 2020-03-31
-https://docs.google.com/presentation/d/18EJKBJJ2LHI7uNU_l8s_Cm-aGZhw9tkoQ8BxqYZfkWk/edit#slide=id.g71f805bc32_0_80
diff --git a/video_prediction_savp/scripts/Analysis_all.py b/video_prediction_savp/scripts/Analysis_all.py
deleted file mode 100644
index 7ed4b6b6..00000000
--- a/video_prediction_savp/scripts/Analysis_all.py
+++ /dev/null
@@ -1,105 +0,0 @@
-import pickle
-import os
-from matplotlib.pylab import plt
-
-# results_path = ["results_test_samples/era5_size_64_64_3_norm_dup_pretrained/ours_savp","results_test_samples/era5_size_64_64_3_norm_dup_pretrained_finetune/ours_savp",
-# "results_test_samples/era5_size_64_64_3_norm_dup_pretrained_gan/kth_ours_gan","results_test_samples/era5_size_64_64_3_norm_dup_pretrained_vae_l1/kth_ours_vae_l1"]
-#
-# model_names = ["SAVP","SAVP_Finetune","GAN","VAE"]
-
-
-# results_path = ["results_test_samples/era5_size_64_64_3_norm_dup_pretrained/ours_savp","results_test_samples/era5_size_64_64_3_norm_msl_gph_pretrained_savp/ours_savp",
-# "results_test_samples/era5_size_64_64_3_norm_dup_pretrained_gan/kth_ours_gan","results_test_samples/era5_size_64_64_3_norm_msl_gph_pretrained_gan/kth_ours_gan"]
-#
-# model_names = ["SAVP_3T","SAVP_T-MSL-GPH","GAN_3T","GAN_T-MSL_GPH"]
-#
-# results_path = ["results_test_samples/era5_size_64_64_3_norm_dup_pretrained/ours_savp","results_test_samples/era5_size_64_64_3_norm_dup/ours_savp",
-# "results_test_samples/era5_size_64_64_3_norm_dup_pretrained/kth_ours_gan","results_test_samples/era5_size_64_64_3_norm_dup/ours_gan",
-# "results_test_samples/era5_size_64_64_3_norm_dup_pretrained/kth_ours_vae_l1","results_test_samples/era5_size_64_64_3_norm_dup/ours_vae_l1"]
-# model_names = ["TF-SAVP(KTH)","SAVP (3T)","TF-GAN(KTH)","GAN (3T)","TF-VAE (KTH)","VAE (3T)"]
-
-##
-##results_path = ["results_test_samples/era5_size_64_64_3_norm_t_msl_gph/ours_savp", "results_test_samples/era5_size_64_64_3_norm_dup/ours_savp",
-## "results_test_samples/era5_size_64_64_3_norm_t_msl_gph/ours_gan","results_test_samples/era5_size_64_64_3_norm_dup/ours_gan"]
-##model_names = ["SAVP(T-MSL-GPH)", "SAVP (3T)", "GAN (T-MSL-GPH)","GAN (3T)"]
-
-##results_path = ["results_test_samples/era5_size_64_64_3_norm_t_msl_gph/ours_savp", "results_test_samples/era5_size_64_64_3_norm_dup/ours_savp",
-## "results_test_samples/era5_size_64_64_3_norm_t_msl_gph/ours_gan","results_test_samples/era5_size_64_64_3_norm_dup/ours_gan"]
-##model_names = ["SAVP(T-MSL-GPH)", "SAVP (3T)", "GAN (T-MSL-GPH)","GAN (3T)"]
-##
-##mse_all = []
-##psnr_all = []
-##ssim_all = []
-##for path in results_path:
-## p = os.path.join(path,"results.pkl")
-## result = pickle.load(open(p,"rb"))
-## mse = result["mse"]
-## psnr = result["psnr"]
-## ssim = result["ssim"]
-## mse_all.append(mse)
-## psnr_all.append(psnr)
-## ssim_all.append(ssim)
-##
-##
-##def get_metric(metrtic):
-## if metric == "mse":
-## return mse_all
-## elif metric == "psnr":
-## return psnr_all
-## elif metric == "ssim":
-## return ssim_all
-## else:
-## raise("Metric error")
-##
-##for metric in ["mse","psnr","ssim"]:
-## evals = get_metric(metric)
-## timestamp = list(range(1,11))
-## fig = plt.figure()
-## plt.plot(timestamp, evals[0],'-.',label=model_names[0])
-## plt.plot(timestamp, evals[1],'--',label=model_names[1])
-## plt.plot(timestamp, evals[2],'-',label=model_names[2])
-## plt.plot(timestamp, evals[3],'--.',label=model_names[3])
-## # plt.plot(timestamp, evals[4],'*-.',label=model_names[4])
-## # plt.plot(timestamp, evals[5],'--*',label=model_names[5])
-## if metric == "mse":
-## plt.legend(loc="upper left")
-## else:
-## plt.legend(loc = "upper right")
-## plt.xlabel("Timestamps")
-## plt.ylabel(metric)
-## plt.title(metric,fontsize=15)
-## plt.savefig(metric + "2.png")
-## plt.clf()
-
-
-
-#persistent analysis
-persistent_mse_all = []
-persistent_psnr_all = []
-persistent_ssim_all = []
-mse_all = []
-psnr_all = []
-ssim_all = []
-results_root_path = "/p/scratch/deepacf/video_prediction_shared_folder/results/era5-Y2017M01to12-64x64-50d00N11d50E-T_T_T/ours_gan"
-p1 = os.path.join(results_root_path,"results.pkl")
-result1 = pickle.load(open(p1,"rb"))
-p2 = os.path.join(results_root_path,"persistent_results.pkl")
-result2 = pickle.load(open(p2,"rb"))
-mse = result1["mse"]
-psnr = result1["psnr"]
-ssim = result1["ssim"]
-mse_all.append(mse)
-psnr_all.append(psnr)
-ssim_all.append(ssim)
-
-persistent_mse = result2["mse"]
-persistent_psnr = result2["psnr"]
-persistent_ssim = result2["ssim"]
-persistent_mse_all.append(persistent_mse)
-persistent_psnr_all.append(persistent_psnr)
-persistent_ssim_all.append(persistent_ssim)
-
-
-
-print("persistent_mse",persistent_mse_all)
-print("mse",mse_all)
diff --git a/video_prediction_savp/scripts/evaluate_svg.sh b/video_prediction_savp/scripts/evaluate_svg.sh
deleted file mode 100644
index 212c4ba2..00000000
--- a/video_prediction_savp/scripts/evaluate_svg.sh
+++ /dev/null
@@ -1 +0,0 @@
-#!/usr/bin/env bash
\ No newline at end of file
diff --git a/video_prediction_savp/scripts/generate.py b/video_prediction_savp/scripts/generate.py
deleted file mode 100644
index e1893194..00000000
--- a/video_prediction_savp/scripts/generate.py
+++ /dev/null
@@ -1,537 +0,0 @@
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-
-import argparse
-import errno
-import json
-import os
-import math
-import random
-import cv2
-import numpy as np
-import tensorflow as tf
-import seaborn as sns
-import pickle
-from random import seed
-import random
-import json
-import numpy as np
-#from six.moves import cPickle
-import matplotlib
-matplotlib.use('Agg')
-import matplotlib.pyplot as plt
-import matplotlib.gridspec as gridspec
-import matplotlib.animation as animation
-import seaborn as sns
-import pandas as pd
-from video_prediction import datasets, models
-from matplotlib.colors import LinearSegmentedColormap
-from matplotlib.ticker import MaxNLocator
-from video_prediction.utils.ffmpeg_gif import save_gif
-
-with open("./splits_size_64_64_1/geo_info.json","r") as json_file:
- geo = json.load(json_file)
- lat = [round(i,2) for i in geo["lat"]]
- lon = [round(i,2) for i in geo["lon"]]
-
-def main():
- parser = argparse.ArgumentParser()
- parser.add_argument("--input_dir", type=str, required=True, help="either a directory containing subdirectories "
- "train, val, test, etc, or a directory containing "
- "the tfrecords")
- parser.add_argument("--results_dir", type=str, default='results', help="ignored if output_gif_dir is specified")
- parser.add_argument("--results_gif_dir", type=str, help="default is results_dir. ignored if output_gif_dir is specified")
- parser.add_argument("--results_png_dir", type=str, help="default is results_dir. ignored if output_png_dir is specified")
- parser.add_argument("--output_gif_dir", help="output directory where samples are saved as gifs. default is "
- "results_gif_dir/model_fname")
- parser.add_argument("--output_png_dir", help="output directory where samples are saved as pngs. default is "
- "results_png_dir/model_fname")
- parser.add_argument("--checkpoint", help="directory with checkpoint or checkpoint name (e.g. checkpoint_dir/model-200000)")
-
- parser.add_argument("--mode", type=str, choices=['val', 'test'], default='val', help='mode for dataset, val or test.')
-
- parser.add_argument("--dataset", type=str, help="dataset class name")
- parser.add_argument("--dataset_hparams", type=str, help="a string of comma separated list of dataset hyperparameters")
- parser.add_argument("--model", type=str, help="model class name")
- parser.add_argument("--model_hparams", type=str, help="a string of comma separated list of model hyperparameters")
-
- parser.add_argument("--batch_size", type=int, default=8, help="number of samples in batch")
- parser.add_argument("--num_samples", type=int, help="number of samples in total (all of them by default)")
- parser.add_argument("--num_epochs", type=int, default=1)
-
- parser.add_argument("--num_stochastic_samples", type=int, default=1) #Bing original is 5, change to 1
- parser.add_argument("--gif_length", type=int, help="default is sequence_length")
- parser.add_argument("--fps", type=int, default=4)
-
- parser.add_argument("--gpu_mem_frac", type=float, default=0, help="fraction of gpu memory to use")
- parser.add_argument("--seed", type=int, default=7)
-
- args = parser.parse_args()
-
- if args.seed is not None:
- tf.set_random_seed(args.seed)
- np.random.seed(args.seed)
- random.seed(args.seed)
-
- args.results_gif_dir = args.results_gif_dir or args.results_dir
- args.results_png_dir = args.results_png_dir or args.results_dir
- dataset_hparams_dict = {}
- model_hparams_dict = {}
- if args.checkpoint:
- checkpoint_dir = os.path.normpath(args.checkpoint)
- if not os.path.isdir(args.checkpoint):
- checkpoint_dir, _ = os.path.split(checkpoint_dir)
- if not os.path.exists(checkpoint_dir):
- raise FileNotFoundError(errno.ENOENT, os.strerror(errno.ENOENT), checkpoint_dir)
- with open(os.path.join(checkpoint_dir, "options.json")) as f:
- print("loading options from checkpoint %s" % args.checkpoint)
- options = json.loads(f.read())
- args.dataset = args.dataset or options['dataset']
- args.model = args.model or options['model']
- try:
- with open(os.path.join(checkpoint_dir, "dataset_hparams.json")) as f:
- dataset_hparams_dict = json.loads(f.read())
- except FileNotFoundError:
- print("dataset_hparams.json was not loaded because it does not exist")
- try:
- with open(os.path.join(checkpoint_dir, "model_hparams.json")) as f:
- model_hparams_dict = json.loads(f.read())
- except FileNotFoundError:
- print("model_hparams.json was not loaded because it does not exist")
- args.output_gif_dir = args.output_gif_dir or os.path.join(args.results_gif_dir, os.path.split(checkpoint_dir)[1])
- args.output_png_dir = args.output_png_dir or os.path.join(args.results_png_dir, os.path.split(checkpoint_dir)[1])
- else:
- if not args.dataset:
- raise ValueError('dataset is required when checkpoint is not specified')
- if not args.model:
- raise ValueError('model is required when checkpoint is not specified')
- args.output_gif_dir = args.output_gif_dir or os.path.join(args.results_gif_dir, 'model.%s' % args.model)
- args.output_png_dir = args.output_png_dir or os.path.join(args.results_png_dir, 'model.%s' % args.model)
-
- print('----------------------------------- Options ------------------------------------')
- for k, v in args._get_kwargs():
- print(k, "=", v)
- print('------------------------------------- End --------------------------------------')
-
- VideoDataset = datasets.get_dataset_class(args.dataset)
- dataset = VideoDataset(
- args.input_dir,
- mode=args.mode,
- num_epochs=args.num_epochs,
- seed=args.seed,
- hparams_dict=dataset_hparams_dict,
- hparams=args.dataset_hparams)
- VideoPredictionModel = models.get_model_class(args.model)
- hparams_dict = dict(model_hparams_dict)
- hparams_dict.update({
- 'context_frames': dataset.hparams.context_frames,
- 'sequence_length': dataset.hparams.sequence_length,
- 'repeat': dataset.hparams.time_shift,
- })
- model = VideoPredictionModel(
- mode=args.mode,
- hparams_dict=hparams_dict,
- hparams=args.model_hparams)
-
- sequence_length = model.hparams.sequence_length
- context_frames = model.hparams.context_frames
- future_length = sequence_length - context_frames
-
- if args.num_samples:
- if args.num_samples > dataset.num_examples_per_epoch():
- raise ValueError('num_samples cannot be larger than the dataset')
- num_examples_per_epoch = args.num_samples
- else:
- #Bing: error occurs here, cheats a little bit here
- #num_examples_per_epoch = dataset.num_examples_per_epoch()
- num_examples_per_epoch = args.batch_size * 8
- if num_examples_per_epoch % args.batch_size != 0:
- #bing
- #raise ValueError('batch_size should evenly divide the dataset size %d' % num_examples_per_epoch)
- pass
- #Bing if it is era 5 data we used dataset.make_batch_v2
- #inputs = dataset.make_batch(args.batch_size)
- inputs = dataset.make_batch(args.batch_size)
- input_phs = {k: tf.placeholder(v.dtype, v.shape, '%s_ph' % k) for k, v in inputs.items()}
- with tf.variable_scope(''):
- model.build_graph(input_phs)
-
- for output_dir in (args.output_gif_dir, args.output_png_dir):
- if not os.path.exists(output_dir):
- os.makedirs(output_dir)
- with open(os.path.join(output_dir, "options.json"), "w") as f:
- f.write(json.dumps(vars(args), sort_keys=True, indent=4))
- with open(os.path.join(output_dir, "dataset_hparams.json"), "w") as f:
- f.write(json.dumps(dataset.hparams.values(), sort_keys=True, indent=4))
- with open(os.path.join(output_dir, "model_hparams.json"), "w") as f:
- f.write(json.dumps(model.hparams.values(), sort_keys=True, indent=4))
-
- gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.gpu_mem_frac)
- config = tf.ConfigProto(gpu_options=gpu_options, allow_soft_placement=True)
- sess = tf.Session(config=config)
- sess.graph.as_default()
- model.restore(sess, args.checkpoint)
- sample_ind = 0
- gen_images_all = []
- input_images_all = []
-
- while True:
- if args.num_samples and sample_ind >= args.num_samples:
- break
- try:
- input_results = sess.run(inputs)
- except tf.errors.OutOfRangeError:
- break
- print("evaluation samples from %d to %d" % (sample_ind, sample_ind + args.batch_size))
- feed_dict = {input_ph: input_results[name] for name, input_ph in input_phs.items()}
- for stochastic_sample_ind in range(args.num_stochastic_samples): #Todo: why use here
- print("Stochastic sample id", stochastic_sample_ind)
- gen_images = sess.run(model.outputs['gen_images'], feed_dict=feed_dict)
- #input_images = sess.run(inputs["images"])
- #Bing: Add evaluation metrics
- # fetches = {'images': model.inputs['images']}
- # fetches.update(model.eval_outputs.items())
- # fetches.update(model.eval_metrics.items())
- # results = sess.run(fetches, feed_dict = feed_dict)
- # input_images = results["images"] #shape (batch_size,future_frames,height,width,channel)
- # only keep the future frames
- #gen_images = gen_images[:, -future_length:] #(8,10,64,64,1) (batch_size, sequences, height, width, channel)
- #input_images = input_results["images"][:,-future_length:,:,:]
- input_images = input_results["images"][:,1:,:,:,:]
- #gen_mse_avg = results["eval_mse/avg"] #shape (batch_size,future_frames)
- print("Finish sample ind",stochastic_sample_ind)
- input_gen_diff_ = input_images - gen_images
- #diff_image_range = pd.cut(input_gen_diff_.flatten(), bins = 4, labels = [-10, -5, 0, 5], right = False)
- #diff_image_range = np.reshape(np.array(diff_image_range),input_gen_diff_.shape)
- gen_images_all.extend(gen_images)
- input_images_all.extend(input_images)
-
- colors = [(1, 0, 0), (0, 1, 0), (0, 0, 1)]
- cmap_name = 'my_list'
- if sample_ind < 100:
- for i in range(len(gen_images)):
- name = 'Batch_id_' + str(sample_ind) + " + Sample_" + str(i)
- gen_images_ = gen_images[i, :]
- gen_mse_avg_ = [np.mean(input_gen_diff_[i, frame, :, :, :]**2) for frame in
- range(19)] # return the list with 10 (sequence) mse
- input_gen_diff = input_gen_diff_ [i,:,:,:,:]
- input_images_ = input_images[i, :]
- #gen_mse_avg_ = gen_mse_avg[i, :]
-
- # Bing: This is to check the difference between the images and next images for debugging the freezon issues
- # gen_images_diff = []
- # for gen_idx in range(len(gen_images_) - 1):
- # img_1 = gen_images_[gen_idx, :, :, :]
- # img_2 = gen_images_[gen_idx + 1, :, :, :]
- # img_diff = img_2 - img_1
- # img_diff_nonzero = [e for img_idx, e in enumerate(img_diff.flatten()) if round(e,3) != 0.000]
- # gen_images_diff.append(img_diff_nonzero)
-
- fig = plt.figure()
- gs = gridspec.GridSpec(4,6)
- gs.update(wspace = 0.7,hspace=0.8)
- ax1 = plt.subplot(gs[0:2,0:3])
- ax2 = plt.subplot(gs[0:2,3:],sharey=ax1)
- ax3 = plt.subplot(gs[2:4,0:3])
- ax4 = plt.subplot(gs[2:4,3:])
- xlables = [round(i,2) for i in list(np.linspace(np.min(lon),np.max(lon),5))]
- ylabels = [round(i,2) for i in list(np.linspace(np.max(lat),np.min(lat),5))]
- plt.setp([ax1,ax2,ax3],xticks=list(np.linspace(0,64,5)), xticklabels=xlables ,yticks=list(np.linspace(0,64,5)),yticklabels=ylabels)
- ax1.title.set_text("(a) Ground Truth")
- ax2.title.set_text("(b) SAVP")
- ax3.title.set_text("(c) Diff.")
- ax4.title.set_text("(d) MSE")
-
- ax1.xaxis.set_tick_params(labelsize=7)
- ax1.yaxis.set_tick_params(labelsize = 7)
- ax2.xaxis.set_tick_params(labelsize=7)
- ax2.yaxis.set_tick_params(labelsize = 7)
- ax3.xaxis.set_tick_params(labelsize=7)
- ax3.yaxis.set_tick_params(labelsize = 7)
-
- init_images = np.zeros((input_images_.shape[1], input_images_.shape[2]))
- print("inti images shape", init_images.shape)
- xdata, ydata = [], []
- plot1 = ax1.imshow(init_images, cmap='jet', vmin = 270, vmax = 300)
- plot2 = ax2.imshow(init_images, cmap='jet', vmin = 270, vmax = 300)
- #x = np.linspace(0, 64, 64)
- #y = np.linspace(0, 64, 64)
- #plot1 = ax1.contourf(x,y,init_images, cmap='jet', vmin = np.min(input_images), vmax = np.max(input_images))
- #plot2 = ax2.contourf(x,y,init_images, cmap='jet', vmin = np.min(input_images), vmax = np.max(input_images))
- fig.colorbar(plot1, ax=ax1).ax.tick_params(labelsize=7)
- fig.colorbar(plot2, ax=ax2).ax.tick_params(labelsize=7)
-
- cm = LinearSegmentedColormap.from_list(
- cmap_name, "bwr", N = 5)
-
- plot3 = ax3.imshow(init_images, vmin=-10, vmax=10, cmap=cm)#cmap = 'PuBu_r',
- plot4, = ax4.plot([], [], color = "r")
- ax4.set_xlim(0, len(gen_mse_avg_)-1)
- ax4.set_ylim(0, 10)
- ax4.set_xlabel("Frames", fontsize=10)
- #ax4.set_ylabel("MSE", fontsize=10)
- ax4.xaxis.set_tick_params(labelsize=7)
- ax4.yaxis.set_tick_params(labelsize=7)
-
-
- plots = [plot1, plot2, plot3, plot4]
-
- #fig.colorbar(plots[1], ax = [ax1, ax2])
-
- fig.colorbar(plots[2], ax=ax3).ax.tick_params(labelsize=7)
- #fig.colorbar(plot1[0], ax=ax1).ax.tick_params(labelsize=7)
- #fig.colorbar(plot2[1], ax=ax2).ax.tick_params(labelsize=7)
-
- def animation_sample(t):
- input_image = input_images_[t, :, :, 0]
- gen_image = gen_images_[t, :, :, 0]
- diff_image = input_gen_diff[t,:,:,0]
-
- data = gen_mse_avg_[:t + 1]
- # x = list(range(len(gen_mse_avg_)))[:t+1]
- xdata.append(t)
- print("xdata", xdata)
- ydata.append(gen_mse_avg_[t])
-
- print("ydata", ydata)
- # p = sns.lineplot(x=x,y=data,color="b")
- # p.tick_params(labelsize=17)
- # plt.setp(p.lines, linewidth=6)
- plots[0].set_data(input_image)
- plots[1].set_data(gen_image)
- #plots[0] = ax1.contourf(x, y, input_image, cmap = 'jet', vmin = np.min(input_images),vmax = np.max(input_images))
- #plots[1] = ax2.contourf(x, y, gen_image, cmap = 'jet', vmin = np.min(input_images),vmax = np.max(input_images))
- plots[2].set_data(diff_image)
- plots[3].set_data(xdata, ydata)
- fig.suptitle("Frame " + str(t+1))
-
- return plots
-
- ani = animation.FuncAnimation(fig, animation_sample, frames=len(gen_mse_avg_), interval = 1000,
- repeat_delay = 2000)
- ani.save(os.path.join(args.output_png_dir, "Sample_" + str(name) + ".mp4"))
-
- else:
- pass
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
- # # for i, gen_mse_avg_ in enumerate(gen_mse_avg):
- # # ims = []
- # # fig = plt.figure()
- # # plt.xlim(0,len(gen_mse_avg_))
- # # plt.ylim(np.min(gen_mse_avg),np.max(gen_mse_avg))
- # # plt.xlabel("Frames")
- # # plt.ylabel("MSE_AVG")
- # # #X = list(range(len(gen_mse_avg_)))
- # # #for t, gen_mse_avg_ in enumerate(gen_mse_avg):
- # # def animate_metric(j):
- # # data = gen_mse_avg_[:(j+1)]
- # # x = list(range(len(gen_mse_avg_)))[:(j+1)]
- # # p = sns.lineplot(x=x,y=data,color="b")
- # # p.tick_params(labelsize=17)
- # # plt.setp(p.lines, linewidth=6)
- # # ani = animation.FuncAnimation(fig, animate_metric, frames=len(gen_mse_avg_), interval = 1000, repeat_delay=2000)
- # # ani.save(os.path.join(args.output_png_dir, "MSE_AVG" + str(i) + ".gif"))
- # #
- # #
- # # for i, input_images_ in enumerate(input_images):
- # # #context_images_ = (input_results['images'][i])
- # # #gen_images_fname = 'gen_image_%05d_%02d.gif' % (sample_ind + i, stochastic_sample_ind)
- # # ims = []
- # # fig = plt.figure()
- # # for t, input_image in enumerate(input_images_):
- # # im = plt.imshow(input_images[i, t, :, :, 0], interpolation = 'none')
- # # ttl = plt.text(1.5, 2,"Frame_" + str(t))
- # # ims.append([im,ttl])
- # # ani = animation.ArtistAnimation(fig, ims, interval= 1000, blit=True,repeat_delay=2000)
- # # ani.save(os.path.join(args.output_png_dir,"groud_true_images_" + str(i) + ".gif"))
- # # #plt.show()
- # #
- # # for i,gen_images_ in enumerate(gen_images):
- # # ims = []
- # # fig = plt.figure()
- # # for t, gen_image in enumerate(gen_images_):
- # # im = plt.imshow(gen_images[i, t, :, :, 0], interpolation = 'none')
- # # ttl = plt.text(1.5, 2, "Frame_" + str(t))
- # # ims.append([im, ttl])
- # # ani = animation.ArtistAnimation(fig, ims, interval = 1000, blit = True, repeat_delay = 2000)
- # # ani.save(os.path.join(args.output_png_dir, "prediction_images_" + str(i) + ".gif"))
- #
- #
- # # for i, gen_images_ in enumerate(gen_images):
- # # #context_images_ = (input_results['images'][i] * 255.0).astype(np.uint8)
- # # #gen_images_ = (gen_images_ * 255.0).astype(np.uint8)
- # # #bing
- # # context_images_ = (input_results['images'][i])
- # # gen_images_fname = 'gen_image_%05d_%02d.gif' % (sample_ind + i, stochastic_sample_ind)
- # # context_and_gen_images = list(context_images_[:context_frames]) + list(gen_images_)
- # # plt.figure(figsize = (10,2))
- # # gs = gridspec.GridSpec(2,10)
- # # gs.update(wspace=0.,hspace=0.)
- # # for t, gen_image in enumerate(gen_images_):
- # # gen_image_fname_pattern = 'gen_image_%%05d_%%02d_%%0%dd.png' % max(2,len(str(len(gen_images_) - 1)))
- # # gen_image_fname = gen_image_fname_pattern % (sample_ind + i, stochastic_sample_ind, t)
- # # plt.subplot(gs[t])
- # # plt.imshow(input_images[i, t, :, :, 0], interpolation = 'none') # the last index sets the channel. 0 = t2
- # # # plt.pcolormesh(X_test[i,t,::-1,:,0], shading='bottom', cmap=plt.cm.jet)
- # # plt.tick_params(axis = 'both', which = 'both', bottom = False, top = False, left = False,
- # # right = False, labelbottom = False, labelleft = False)
- # # if t == 0: plt.ylabel('Actual', fontsize = 10)
- # #
- # # plt.subplot(gs[t + 10])
- # # plt.imshow(gen_images[i, t, :, :, 0], interpolation = 'none')
- # # # plt.pcolormesh(X_hat[i,t,::-1,:,0], shading='bottom', cmap=plt.cm.jet)
- # # plt.tick_params(axis = 'both', which = 'both', bottom = False, top = False, left = False,
- # # right = False, labelbottom = False, labelleft = False)
- # # if t == 0: plt.ylabel('Predicted', fontsize = 10)
- # # plt.savefig(os.path.join(args.output_png_dir, gen_image_fname) + 'plot_' + str(i) + '.png')
- # # plt.clf()
- #
- # # if args.gif_length:
- # # context_and_gen_images = context_and_gen_images[:args.gif_length]
- # # save_gif(os.path.join(args.output_gif_dir, gen_images_fname),
- # # context_and_gen_images, fps=args.fps)
- # #
- # # gen_image_fname_pattern = 'gen_image_%%05d_%%02d_%%0%dd.png' % max(2, len(str(len(gen_images_) - 1)))
- # # for t, gen_image in enumerate(gen_images_):
- # # gen_image_fname = gen_image_fname_pattern % (sample_ind + i, stochastic_sample_ind, t)
- # # if gen_image.shape[-1] == 1:
- # # gen_image = np.tile(gen_image, (1, 1, 3))
- # # else:
- # # gen_image = cv2.cvtColor(gen_image, cv2.COLOR_RGB2BGR)
- # # cv2.imwrite(os.path.join(args.output_png_dir, gen_image_fname), gen_image)
-
- sample_ind += args.batch_size
-
-
- with open(os.path.join(args.output_png_dir, "input_images_all"),"wb") as input_files:
- pickle.dump(input_images_all,input_files)
-
- with open(os.path.join(args.output_png_dir, "gen_images_all"),"wb") as gen_files:
- pickle.dump(gen_images_all,gen_files)
-
- with open(os.path.join(args.output_png_dir, "input_images_all"),"rb") as input_files:
- input_images_all = pickle.load(input_files)
-
- with open(os.path.join(args.output_png_dir, "gen_images_all"),"rb") as gen_files:
- gen_images_all=pickle.load(gen_files)
- ims = []
- fig = plt.figure()
- for frame in range(19):
- input_gen_diff = np.mean((np.array(gen_images_all) - np.array(input_images_all))**2, axis = 0)[frame, :,:, 0] # Get the first prediction frame (batch,height, width, channel)
- #pix_mean = np.mean(input_gen_diff, axis = 0)
- #pix_std = np.std(input_gen_diff, axis=0)
- im = plt.imshow(input_gen_diff, interpolation = 'none',cmap='PuBu')
- if frame == 0:
- fig.colorbar(im)
- ttl = plt.text(1.5, 2, "Frame_" + str(frame +1))
- ims.append([im, ttl])
- ani = animation.ArtistAnimation(fig, ims, interval = 1000, blit = True, repeat_delay = 2000)
- ani.save(os.path.join(args.output_png_dir, "Mean_Frames.mp4"))
- plt.close("all")
-
- ims = []
- fig = plt.figure()
- for frame in range(19):
- pix_std= np.std((np.array(gen_images_all) - np.array(input_images_all))**2, axis = 0)[frame, :,:, 0] # Get the first prediction frame (batch,height, width, channel)
- #pix_mean = np.mean(input_gen_diff, axis = 0)
- #pix_std = np.std(input_gen_diff, axis=0)
- im = plt.imshow(pix_std, interpolation = 'none',cmap='PuBu')
- if frame == 0:
- fig.colorbar(im)
- ttl = plt.text(1.5, 2, "Frame_" + str(frame+1))
- ims.append([im, ttl])
- ani = animation.ArtistAnimation(fig, ims, interval = 1000, blit = True, repeat_delay = 2000)
- ani.save(os.path.join(args.output_png_dir, "Std_Frames.mp4"))
-
- gen_images_all = np.array(gen_images_all)
- input_images_all = np.array(input_images_all)
- # mse_model = np.mean((input_images_all[:, 1:,:,:,0] - gen_images_all[:, 1:,:,:,0])**2) # look at all timesteps except the first
- # mse_model_last = np.mean((input_images_all[:, future_length-1,:,:,0] - gen_images_all[:, future_length-1,:,:,0])**2)
- # mse_prev = np.mean((input_images_all[:, :-1,:,:,0] - gen_images_all[:, 1:,:,:,0])**2 )
-
- mse_model = np.mean((input_images_all[:, :10,:,:,0] - gen_images_all[:, :10,:,:,0])**2) # look at all timesteps except the first
- mse_model_last = np.mean((input_images_all[:,10,:,:,0] - gen_images_all[:, 10,:,:,0])**2)
- mse_prev = np.mean((input_images_all[:, :9, :, :, 0] - input_images_all[:, 1:10, :, :, 0])**2 )
-
- def psnr(img1, img2):
- mse = np.mean((img1 - img2) ** 2)
- if mse == 0: return 100
- PIXEL_MAX = 1
- return 20 * math.log10(PIXEL_MAX / math.sqrt(mse))
-
- psnr_model = psnr(input_images_all[:, :10, :, :, 0], gen_images_all[:, :10, :, :, 0])
- psnr_model_last = psnr(input_images_all[:, 10, :, :, 0], gen_images_all[:,10, :, :, 0])
- psnr_prev = psnr(input_images_all[:, :9, :, :, 0], input_images_all[:, 1:10, :, :, 0])
- f = open(os.path.join(args.output_png_dir,'prediction_scores_4prediction.txt'), 'w')
- f.write("Model MSE: %f\n" % mse_model)
- f.write("Model MSE from only last prediction in sequence: %f\n" % mse_model_last)
- f.write("Previous Frame MSE: %f\n" % mse_prev)
- f.write("Model PSNR: %f\n" % psnr_model)
- f.write("Model PSNR from only last prediction in sequence: %f\n" % psnr_model_last)
- f.write("Previous frame PSNR: %f\n" % psnr_prev)
- f.write("Shape of X_test: " + str(input_images_all.shape))
- f.write("")
- f.write("Shape of X_hat: " + str(gen_images_all.shape))
- f.close()
-
- seed(1)
- s = random.sample(range(len(gen_images_all)), 100)
- print("******KDP******")
- #kernel density plot for checking the model collapse
- fig = plt.figure()
- kdp = sns.kdeplot(gen_images_all[s].flatten(), shade=True, color="r", label = "Generate Images")
- kdp = sns.kdeplot(input_images_all[s].flatten(), shade=True, color="b", label = "Ground True")
- kdp.set(xlabel = 'Temperature (K)', ylabel = 'Probability')
- plt.savefig(os.path.join(args.output_png_dir, "kdp_gen_images.png"), dpi = 400)
- plt.clf()
-
- #line plot for evaluating the prediction and groud-truth
- for i in [0,3,6,9,12,15,18]:
- fig = plt.figure()
- plt.scatter(gen_images_all[:,i,:,:][s].flatten(),input_images_all[:,i,:,:][s].flatten(),s=0.3)
- #plt.scatter(gen_images_all[:,0,:,:].flatten(),input_images_all[:,0,:,:].flatten(),s=0.3)
- plt.xlabel("Prediction")
- plt.ylabel("Real values")
- plt.title("Frame_{}".format(i+1))
- plt.plot([250,300], [250,300],color="black")
- plt.savefig(os.path.join(args.output_png_dir,"pred_real_frame_{}.png".format(str(i))))
- plt.clf()
-
-
- mse_model_by_frames = np.mean((input_images_all[:, :, :, :, 0][s] - gen_images_all[:, :, :, :, 0][s]) ** 2,axis=(2,3)) #return (batch, sequence)
- x = [str(i+1) for i in list(range(19))]
- fig,axis = plt.subplots()
- mean_f = np.mean(mse_model_by_frames, axis = 0)
- median = np.median(mse_model_by_frames, axis=0)
- q_low = np.quantile(mse_model_by_frames, q=0.25, axis=0)
- q_high = np.quantile(mse_model_by_frames, q=0.75, axis=0)
- d_low = np.quantile(mse_model_by_frames,q=0.1, axis=0)
- d_high = np.quantile(mse_model_by_frames, q=0.9, axis=0)
- plt.fill_between(x, d_high, d_low, color="ghostwhite",label="interdecile range")
- plt.fill_between(x,q_high, q_low , color = "lightgray", label="interquartile range")
- plt.plot(x, median, color="grey", linewidth=0.6, label="Median")
- plt.plot(x, mean_f, color="peachpuff",linewidth=1.5, label="Mean")
- plt.title(f'MSE percentile')
- plt.xlabel("Frames")
- plt.legend(loc=2, fontsize=8)
- plt.savefig(os.path.join(args.output_png_dir,"mse_percentiles.png"))
-
-if __name__ == '__main__':
- main()
diff --git a/workflow_parallel_frame_prediction/DataExtraction/mpi_stager_v2.py b/video_prediction_savp/scripts/mpi_stager_v2.py
similarity index 100%
rename from workflow_parallel_frame_prediction/DataExtraction/mpi_stager_v2.py
rename to video_prediction_savp/scripts/mpi_stager_v2.py
diff --git a/workflow_parallel_frame_prediction/DataPreprocess/mpi_stager_v2_process_netCDF.py b/video_prediction_savp/scripts/mpi_stager_v2_process_netCDF.py
similarity index 100%
rename from workflow_parallel_frame_prediction/DataPreprocess/mpi_stager_v2_process_netCDF.py
rename to video_prediction_savp/scripts/mpi_stager_v2_process_netCDF.py
diff --git a/video_prediction_savp/scripts/train_v2.py b/video_prediction_savp/scripts/train_v2.py
deleted file mode 100644
index f83192d6..00000000
--- a/video_prediction_savp/scripts/train_v2.py
+++ /dev/null
@@ -1,362 +0,0 @@
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-
-import argparse
-import errno
-import json
-import os
-import random
-import time
-
-import numpy as np
-import tensorflow as tf
-
-from video_prediction import datasets, models
-
-
-def add_tag_suffix(summary, tag_suffix):
- summary_proto = tf.Summary()
- summary_proto.ParseFromString(summary)
- summary = summary_proto
-
- for value in summary.value:
- tag_split = value.tag.split('/')
- value.tag = '/'.join([tag_split[0] + tag_suffix] + tag_split[1:])
- return summary.SerializeToString()
-
-
-def main():
- parser = argparse.ArgumentParser()
- parser.add_argument("--input_dir", type=str, required=True, help="either a directory containing subdirectories "
- "train, val, test, etc, or a directory containing "
- "the tfrecords")
- parser.add_argument("--val_input_dir", type=str, help="directories containing the tfrecords. default: input_dir")
- parser.add_argument("--logs_dir", default='logs', help="ignored if output_dir is specified")
- parser.add_argument("--output_dir", help="output directory where json files, summary, model, gifs, etc are saved. "
- "default is logs_dir/model_fname, where model_fname consists of "
- "information from model and model_hparams")
- parser.add_argument("--output_dir_postfix", default="")
- parser.add_argument("--checkpoint", help="directory with checkpoint or checkpoint name (e.g. checkpoint_dir/model-200000)")
- parser.add_argument("--resume", action='store_true', help='resume from lastest checkpoint in output_dir.')
-
- parser.add_argument("--dataset", type=str, help="dataset class name")
- parser.add_argument("--dataset_hparams", type=str, help="a string of comma separated list of dataset hyperparameters")
- parser.add_argument("--dataset_hparams_dict", type=str, help="a json file of dataset hyperparameters")
- parser.add_argument("--model", type=str, help="model class name")
- parser.add_argument("--model_hparams", type=str, help="a string of comma separated list of model hyperparameters")
- parser.add_argument("--model_hparams_dict", type=str, help="a json file of model hyperparameters")
-
- parser.add_argument("--summary_freq", type=int, default=1000, help="save frequency of summaries (except for image and eval summaries) for train/validation set")
- parser.add_argument("--image_summary_freq", type=int, default=5000, help="save frequency of image summaries for train/validation set")
- parser.add_argument("--eval_summary_freq", type=int, default=25000, help="save frequency of eval summaries for train/validation set")
- parser.add_argument("--accum_eval_summary_freq", type=int, default=100000, help="save frequency of accumulated eval summaries for validation set only")
- parser.add_argument("--progress_freq", type=int, default=100, help="display progress every progress_freq steps")
- parser.add_argument("--save_freq", type=int, default=5000, help="save frequence of model, 0 to disable")
-
- parser.add_argument("--aggregate_nccl", type=int, default=0, help="whether to use nccl or cpu for gradient aggregation in multi-gpu training")
- parser.add_argument("--gpu_mem_frac", type=float, default=0, help="fraction of gpu memory to use")
- parser.add_argument("--seed", type=int)
-
- args = parser.parse_args()
-
- if args.seed is not None:
- tf.set_random_seed(args.seed)
- np.random.seed(args.seed)
- random.seed(args.seed)
-
- if args.output_dir is None:
- list_depth = 0
- model_fname = ''
- for t in ('model=%s,%s' % (args.model, args.model_hparams)):
- if t == '[':
- list_depth += 1
- if t == ']':
- list_depth -= 1
- if list_depth and t == ',':
- t = '..'
- if t in '=,':
- t = '.'
- if t in '[]':
- t = ''
- model_fname += t
- args.output_dir = os.path.join(args.logs_dir, model_fname) + args.output_dir_postfix
-
- if args.resume:
- if args.checkpoint:
- raise ValueError('resume and checkpoint cannot both be specified')
- args.checkpoint = args.output_dir
-
- dataset_hparams_dict = {}
- model_hparams_dict = {}
- if args.dataset_hparams_dict:
- with open(args.dataset_hparams_dict) as f:
- dataset_hparams_dict.update(json.loads(f.read()))
- if args.model_hparams_dict:
- with open(args.model_hparams_dict) as f:
- model_hparams_dict.update(json.loads(f.read()))
- if args.checkpoint:
- checkpoint_dir = os.path.normpath(args.checkpoint)
- if not os.path.isdir(args.checkpoint):
- checkpoint_dir, _ = os.path.split(checkpoint_dir)
- if not os.path.exists(checkpoint_dir):
- raise FileNotFoundError(errno.ENOENT, os.strerror(errno.ENOENT), checkpoint_dir)
- with open(os.path.join(checkpoint_dir, "options.json")) as f:
- print("loading options from checkpoint %s" % args.checkpoint)
- options = json.loads(f.read())
- args.dataset = args.dataset or options['dataset']
- args.model = args.model or options['model']
- try:
- with open(os.path.join(checkpoint_dir, "dataset_hparams.json")) as f:
- dataset_hparams_dict.update(json.loads(f.read()))
- except FileNotFoundError:
- print("dataset_hparams.json was not loaded because it does not exist")
- try:
- with open(os.path.join(checkpoint_dir, "model_hparams.json")) as f:
- model_hparams_dict.update(json.loads(f.read()))
- except FileNotFoundError:
- print("model_hparams.json was not loaded because it does not exist")
-
- print('----------------------------------- Options ------------------------------------')
- for k, v in args._get_kwargs():
- print(k, "=", v)
- print('------------------------------------- End --------------------------------------')
-
- VideoDataset = datasets.get_dataset_class(args.dataset)
- train_dataset = VideoDataset(
- args.input_dir,
- mode='train',
- hparams_dict=dataset_hparams_dict,
- hparams=args.dataset_hparams)
- val_dataset = VideoDataset(
- args.val_input_dir or args.input_dir,
- mode='val',
- hparams_dict=dataset_hparams_dict,
- hparams=args.dataset_hparams)
- if val_dataset.hparams.long_sequence_length != val_dataset.hparams.sequence_length:
- # the longer dataset is only used for the accum_eval_metrics
- long_val_dataset = VideoDataset(
- args.val_input_dir or args.input_dir,
- mode='val',
- hparams_dict=dataset_hparams_dict,
- hparams=args.dataset_hparams)
- long_val_dataset.set_sequence_length(val_dataset.hparams.long_sequence_length)
- else:
- long_val_dataset = None
-
- variable_scope = tf.get_variable_scope()
- variable_scope.set_use_resource(True)
-
- VideoPredictionModel = models.get_model_class(args.model)
- hparams_dict = dict(model_hparams_dict)
- hparams_dict.update({
- 'context_frames': train_dataset.hparams.context_frames,
- 'sequence_length': train_dataset.hparams.sequence_length,
- 'repeat': train_dataset.hparams.time_shift,
- })
- model = VideoPredictionModel(
- hparams_dict=hparams_dict,
- hparams=args.model_hparams,
- aggregate_nccl=args.aggregate_nccl)
-
- batch_size = model.hparams.batch_size
- train_tf_dataset = train_dataset.make_dataset_v2(batch_size)#Bing: adopt the meteo data prepartion here
- train_iterator = train_tf_dataset.make_one_shot_iterator()#Bing:for era5, the problem happen in sess.run(feches) should come from here
- # The `Iterator.string_handle()` method returns a tensor that can be evaluated
- # and used to feed the `handle` placeholder.
- train_handle = train_iterator.string_handle()
- val_tf_dataset = val_dataset.make_dataset_v2(batch_size)
- val_iterator = val_tf_dataset.make_one_shot_iterator()
- val_handle = val_iterator.string_handle()
- #iterator = tf.data.Iterator.from_string_handle(
- # train_handle, train_tf_dataset.output_types, train_tf_dataset.output_shapes)
- inputs = train_iterator.get_next()
-
- # inputs comes from the training dataset by default, unless train_handle is remapped to the val_handles
- model.build_graph(inputs, finetune=True)
-
- if long_val_dataset is not None:
- # separately build a model for the longer sequence.
- # this is needed because the model doesn't support dynamic shapes.
- long_hparams_dict = dict(hparams_dict)
- long_hparams_dict['sequence_length'] = long_val_dataset.hparams.sequence_length
- # use smaller batch size for longer model to prevenet running out of memory
- long_hparams_dict['batch_size'] = model.hparams.batch_size // 2
- long_model = VideoPredictionModel(
- mode="test", # to not build the losses and discriminators
- hparams_dict=long_hparams_dict,
- hparams=args.model_hparams,
- aggregate_nccl=args.aggregate_nccl)
- tf.get_variable_scope().reuse_variables()
- long_model.build_graph(long_val_dataset.make_batch(batch_size))
- else:
- long_model = None
-
- if not os.path.exists(args.output_dir):
- os.makedirs(args.output_dir)
- with open(os.path.join(args.output_dir, "options.json"), "w") as f:
- f.write(json.dumps(vars(args), sort_keys=True, indent=4))
- with open(os.path.join(args.output_dir, "dataset_hparams.json"), "w") as f:
- f.write(json.dumps(train_dataset.hparams.values(), sort_keys=True, indent=4))
- with open(os.path.join(args.output_dir, "model_hparams.json"), "w") as f:
- f.write(json.dumps(model.hparams.values(), sort_keys=True, indent=4))
-
- with tf.name_scope("parameter_count"):
- # exclude trainable variables that are replicas (used in multi-gpu setting)
- trainable_variables = set(tf.trainable_variables()) & set(model.saveable_variables)
- parameter_count = tf.reduce_sum([tf.reduce_prod(tf.shape(v)) for v in trainable_variables])
-
- saver = tf.train.Saver(var_list=model.saveable_variables, max_to_keep=2)
-
- # None has the special meaning of evaluating at the end, so explicitly check for non-equality to zero
- if (args.summary_freq != 0 or args.image_summary_freq != 0 or
- args.eval_summary_freq != 0 or args.accum_eval_summary_freq != 0):
- summary_writer = tf.summary.FileWriter(args.output_dir)
-
- gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.gpu_mem_frac, allow_growth=True)
- config = tf.ConfigProto(gpu_options=gpu_options, allow_soft_placement=True)
- global_step = tf.train.get_or_create_global_step()
- max_steps = model.hparams.max_steps
- with tf.Session(config=config) as sess:
- print("parameter_count =", sess.run(parameter_count))
- sess.run(tf.global_variables_initializer())
- sess.run(tf.local_variables_initializer())
- #coord = tf.train.Coordinator()
- #threads = tf.train.start_queue_runners(sess = sess, coord = coord)
- print("Init done: {sess.run(tf.local_variables_initializer())}%")
- model.restore(sess, args.checkpoint)
- print("Restore processed finished")
- sess.run(model.post_init_ops)
- print("Model run started")
- val_handle_eval = sess.run(val_handle)
- print("val handle done")
- sess.graph.finalize()
- print("graph inalize done")
- start_step = sess.run(global_step)
- print("global step done")
-
- def should(step, freq):
- if freq is None:
- return (step + 1) == (max_steps - start_step)
- else:
- return freq and ((step + 1) % freq == 0 or (step + 1) in (0, max_steps - start_step))
-
- def should_eval(step, freq):
- # never run eval summaries at the beginning since it's expensive, unless it's the last iteration
- return should(step, freq) and (step >= 0 or (step + 1) == (max_steps - start_step))
-
- # start at one step earlier to log everything without doing any training
- # step is relative to the start_step
- for step in range(-1, max_steps - start_step):
- if step == 1:
- # skip step -1 and 0 for timing purposes (for warmstarting)
- start_time = time.time()
-
- fetches = {"global_step": global_step}
- if step >= 0:
- fetches["train_op"] = model.train_op
- if should(step, args.progress_freq):
- fetches['d_loss'] = model.d_loss
- fetches['g_loss'] = model.g_loss
- fetches['d_losses'] = model.d_losses
- fetches['g_losses'] = model.g_losses
- if isinstance(model.learning_rate, tf.Tensor):
- fetches["learning_rate"] = model.learning_rate
- if should(step, args.summary_freq):
- fetches["summary"] = model.summary_op
- if should(step, args.image_summary_freq):
- fetches["image_summary"] = model.image_summary_op
- if should_eval(step, args.eval_summary_freq):
- fetches["eval_summary"] = model.eval_summary_op
-
- run_start_time = time.time()
- results = sess.run(fetches) #fetch the elements in dictinoary fetch
-
- run_elapsed_time = time.time() - run_start_time
- if run_elapsed_time > 1.5 and step > 0 and set(fetches.keys()) == {"global_step", "train_op"}:
- print('running train_op took too long (%0.1fs)' % run_elapsed_time)
-
- if (should(step, args.summary_freq) or
- should(step, args.image_summary_freq) or
- should_eval(step, args.eval_summary_freq)):
- val_fetches = {"global_step": global_step}
- if should(step, args.summary_freq):
- val_fetches["summary"] = model.summary_op
- if should(step, args.image_summary_freq):
- val_fetches["image_summary"] = model.image_summary_op
- if should_eval(step, args.eval_summary_freq):
- val_fetches["eval_summary"] = model.eval_summary_op
- val_results = sess.run(val_fetches, feed_dict={train_handle: val_handle_eval})
- for name, summary in val_results.items():
- if name == 'global_step':
- continue
- val_results[name] = add_tag_suffix(summary, '_1')
-
- if should(step, args.summary_freq):
- print("recording summary")
- summary_writer.add_summary(results["summary"], results["global_step"])
- summary_writer.add_summary(val_results["summary"], val_results["global_step"])
- print("done")
- if should(step, args.image_summary_freq):
- print("recording image summary")
- summary_writer.add_summary(results["image_summary"], results["global_step"])
- summary_writer.add_summary(val_results["image_summary"], val_results["global_step"])
- print("done")
- if should_eval(step, args.eval_summary_freq):
- print("recording eval summary")
- summary_writer.add_summary(results["eval_summary"], results["global_step"])
- summary_writer.add_summary(val_results["eval_summary"], val_results["global_step"])
- print("done")
- if should_eval(step, args.accum_eval_summary_freq):
- val_datasets = [val_dataset]
- val_models = [model]
- if long_model is not None:
- val_datasets.append(long_val_dataset)
- val_models.append(long_model)
- for i, (val_dataset_, val_model) in enumerate(zip(val_datasets, val_models)):
- sess.run(val_model.accum_eval_metrics_reset_op)
- # traverse (roughly up to rounding based on the batch size) all the validation dataset
- accum_eval_summary_num_updates = val_dataset_.num_examples_per_epoch() // val_model.hparams.batch_size
- val_fetches = {"global_step": global_step, "accum_eval_summary": val_model.accum_eval_summary_op}
- for update_step in range(accum_eval_summary_num_updates):
- print('evaluating %d / %d' % (update_step + 1, accum_eval_summary_num_updates))
- val_results = sess.run(val_fetches, feed_dict={train_handle: val_handle_eval})
- accum_eval_summary = add_tag_suffix(val_results["accum_eval_summary"], '_%d' % (i + 1))
- print("recording accum eval summary")
- summary_writer.add_summary(accum_eval_summary, val_results["global_step"])
- print("done")
- if (should(step, args.summary_freq) or should(step, args.image_summary_freq) or
- should_eval(step, args.eval_summary_freq) or should_eval(step, args.accum_eval_summary_freq)):
- summary_writer.flush()
- if should(step, args.progress_freq):
- # global_step will have the correct step count if we resume from a checkpoint
- # global step is read before it's incremented
- steps_per_epoch = train_dataset.num_examples_per_epoch() / batch_size
- train_epoch = results["global_step"] / steps_per_epoch
- print("progress global step %d epoch %0.1f" % (results["global_step"] + 1, train_epoch))
- if step > 0:
- elapsed_time = time.time() - start_time
- average_time = elapsed_time / step
- images_per_sec = batch_size / average_time
- remaining_time = (max_steps - (start_step + step + 1)) * average_time
- print(" image/sec %0.1f remaining %dm (%0.1fh) (%0.1fd)" %
- (images_per_sec, remaining_time / 60, remaining_time / 60 / 60, remaining_time / 60 / 60 / 24))
-
- if results['d_losses']:
- print("d_loss", results["d_loss"])
- for name, loss in results['d_losses'].items():
- print(" ", name, loss)
- if results['g_losses']:
- print("g_loss", results["g_loss"])
- for name, loss in results['g_losses'].items():
- print(" ", name, loss)
- if isinstance(model.learning_rate, tf.Tensor):
- print("learning_rate", results["learning_rate"])
-
- if should(step, args.save_freq):
- print("saving model to", args.output_dir)
- saver.save(sess, os.path.join(args.output_dir, "model"), global_step=global_step)
- print("done")
-
-if __name__ == '__main__':
- main()
diff --git a/workflow_parallel_frame_prediction/DataExtraction/external_function.py b/video_prediction_savp/utils/external_function.py
similarity index 100%
rename from workflow_parallel_frame_prediction/DataExtraction/external_function.py
rename to video_prediction_savp/utils/external_function.py
diff --git a/workflow_parallel_frame_prediction/DataExtraction/shiftgrid.py b/video_prediction_savp/utils/shiftgrid.py
similarity index 100%
rename from workflow_parallel_frame_prediction/DataExtraction/shiftgrid.py
rename to video_prediction_savp/utils/shiftgrid.py
diff --git a/workflow_parallel_frame_prediction/DataExtraction/prepare_era5_data.py b/video_prediction_savp/video_prediction/datasets/prepare_era5_data.py
similarity index 100%
rename from workflow_parallel_frame_prediction/DataExtraction/prepare_era5_data.py
rename to video_prediction_savp/video_prediction/datasets/prepare_era5_data.py
diff --git a/workflow_parallel_frame_prediction/DataPreprocess/process_netCDF_v2.py b/video_prediction_savp/video_prediction/datasets/process_netCDF_v2.py
similarity index 100%
rename from workflow_parallel_frame_prediction/DataPreprocess/process_netCDF_v2.py
rename to video_prediction_savp/video_prediction/datasets/process_netCDF_v2.py
diff --git a/workflow_parallel_frame_prediction/DataExtraction/helper_single_master.py b/workflow_parallel_frame_prediction/DataExtraction/helper_single_master.py
deleted file mode 100644
index a26d7639..00000000
--- a/workflow_parallel_frame_prediction/DataExtraction/helper_single_master.py
+++ /dev/null
@@ -1,245 +0,0 @@
-from mpi4py import MPI
-from os import walk
-import os
-import sys
-import subprocess
-import logging
-import time
-import hashlib
-import argparse
-from os import listdir
-from os.path import isfile, join
-
-# ini. MPI
-comm = MPI.COMM_WORLD
-my_rank = comm.Get_rank() # rank of the node
-p = comm.Get_size() # number of assigned nods
-my_rank = comm.Get_rank() # rank of the node
-
-
-# ======================= List of functions ====================================== #
-if my_rank == 0: # node is master
-
- logger = logging.getLogger(__file__)
- logger.addHandler(logging.StreamHandler(sys.stdout))
-
-
-def directory_scanner(source_path,load_level):
- # Take a look inside a directories and make a list of ll the folders, sub directories, number of the files and size
- # NOTE : It will neglect if there is a sub-directories inside directories!!!
- # NOTE : It will discriminate between the load level : sub-directories / Files
-
- dir_detail_list = [] # directories details
- list_items_to_process = []
- total_size_source = 0
- total_num_files = 0
- list_directories = []
-
- ## =================== Here will be for the Files ================= ##
-
- if load_level == 1:
-
- # Listing all the files in the directory
- for dirpath, dirnames, filenames in os.walk(source_path):
- list_items_to_process.extend(filenames)
-
- for f in list_items_to_process :
- path = source_path +"/"+ str(f)
- statinfo = os.stat(path)
- size = statinfo.st_size
- total_size_source = total_size_source + int(size)
-
- total_num_files = len(list_items_to_process) # number of the files in the source
- total_num_directories = int(0) # TODO need to unify the concept as the number of items
-
- ## ===================== Here will be for the directories ========== ##
-
- if load_level == 0:
- list_directories = os.listdir(source_path)
-
- for d in list_directories:
- path = source_path + d
- if os.path.isdir(path):
- list_items_to_process.append(d)
- list_items_to_process.sort()
- num_files = 0
- # size of the files and subdirectories
- size_dir = subprocess.check_output(['du', '-sc', path])
- splitted = size_dir.split() # fist item is the size of the folder
- size = (splitted[0])
- num_files = len([f for f in os.listdir(path) if os.path.isfile(os.path.join(path, f))])
- dir_detail_list.extend([d, size, num_files])
- total_num_files = total_num_files + int(num_files)
- total_size_source = total_size_source + int(size)
-
- else:
- message = path,'does not exist'
- logging.error(message)
-
-
- total_num_directories = int(len(list_directories))
-
- ## ======================= End of the Directory case =================== ##
- total_size_source = float(total_size_source / 1000000) # human readable size source
-
- logger.info("=== Directory Scanner output ===")
- message = 'Total size of the source directory is:' + str(total_size_source) + 'Gb.'
- logger.info(message)
- message = "Total number of the files in the source directory is: " + str(total_num_files)
- logger.info(message)
- message = "Total number of the directories in the source directory is: " + str(total_num_directories)
- logger.info(message)
-
- # Unifying the naming of this section for both cases : Sub - Directory or File
- # dir_detail_list == > Including the name of the directories, size and number of teh files in each directory / for files is empty
- # list_items_to_process === > List of items to process (Sub-Directories / Files)
- # total_size_source === > Total size of the items to process
- # total_num_files === > for Sub - Directories : sum of all files in different directories / for Files is sum of all
- # total_num_directories === > for Files = 0
-
- return dir_detail_list, list_items_to_process, total_size_source, total_num_files, total_num_directories
-
-# Source - Directoy
-# Destination Rirectory
-# Dir_detail_list
-# list_items_to_process
-# load level
-
-def data_structure_builder (source_dir, destination_dir, dir_detail_list, list_items_to_process,load_level):
-
-
- if not os.path.exists(destination_dir): # check if the Destination dir. is existing
- os_command = ("mkdir " + destination_dir)
- os.system(os_command)
- logger.info('destination path is created')
- else:
- logger.info('The destination path exists')
-
-
- os.chdir(destination_dir) # chnage the directory to the destination
-
- if load_level == 0:
- logging.info('Load Level = 0 : Data Sctructure will be build')
-
- for dir_name in list_items_to_process:
- #print(dir_name)
- dir_path = destination_dir + dir_name
-
- # TODO : os.mkdir() it can be cleaned up to use the OS predifnie functions
- if not os.path.exists(dir_path):
- #print(dir_name + " will be created ")
- os_command = ("mkdir " + dir_name)
- os.system(os_command)
- logging.info(dir_name + " is created ")
-
-
- if load_level == 1:
- logger.info('Load Level = 1 : File will be processed')
-
- return
-
-
-
-def load_distributor(dir_detail_list, list_items_to_process, total_size_source, total_num_files, total_num_directories,load_level, processor_num):
- firs_slave_processor_id = 1
- # create a dictionary with p number of keys
- # for each directory they add the name to one of the keys
- # here we define the first availabe slave node as well
- transfer_dict = dict.fromkeys(list(range(firs_slave_processor_id, processor_num)))
- print(transfer_dict)
- logger.info("The follwoing is in the load Balancer ")
- logger.info(transfer_dict)
- logger.info(list_items_to_process)
- logger.info(total_num_directories)
- logger.info(total_num_files)
-
- # package_counter = 0 possibility to use the counter to fill
- counter = firs_slave_processor_id # this is the ID of the first available slave to p!
-
- if load_level == 0:
- for Directory_counter in range(0, total_num_directories):
- if transfer_dict[counter] is None: # if the value for the key is None add to it
- transfer_dict[counter] = list_items_to_process[Directory_counter]
- else: # if key has a value join the new value to the old value
- transfer_dict[counter] = "{};{}".format(transfer_dict[counter], list_items_to_process[Directory_counter])
- counter = counter + 1
- if counter == processor_num:
- counter = firs_slave_processor_id
-
- if load_level == 1:
- for File_counter in range(0, total_num_files):
- if transfer_dict[counter] is None: # if the value for the key is None add to it
- #print(" M1: New key made for a free processor number {my_rank}".format(my_rank = counter))
- # statemnet if we have more than number of the files processor available
- if counter > len(list_items_to_process) + (firs_slave_processor_id - 1 ):
- transfer_dict[counter] = None
- else:
- transfer_dict[counter] = list_items_to_process[File_counter]
-
-
-
- else: # if key has a value join the new value to the old value
- transfer_dict[counter] = "{};{}".format(transfer_dict[counter], list_items_to_process[File_counter])
- counter = counter + 1
- if counter == processor_num:
- counter = firs_slave_processor_id
-
- logging.info(transfer_dict)
- return transfer_dict
-
-def sync_file(source_path, destination_dir, job_name, rsync_status):
- rsync_msg = ("rsync -r " + source_path + job_name + "/" + " " + destination_dir + "/" + job_name)
- # print('Node:', str(my_rank),'will execute :', rsync_str,'\r\n')
- # sync the assigned folder
-
- if rsync_status == 1:
- os.system(rsync_msg)
-
- return
-
-
-
-def hash_directory(source_path,job_name,hash_rep_file,input_status):
- #sha256_hash = hashlib.sha256()
- md5_hash = hashlib.md5()
-
- ########## Create a hashed file repasitory for direcotry(ies) assigned to node #######
- hash_repo_text = input_status + "_"+job_name +"_hashed.txt"
- os.chdir(hash_rep_file)
- hashed_text_note=open(hash_repo_text,"w+")
-
- # job_name is the name of the subdirectory that is going to be processed
- directory_to_process = source_path + job_name
- # print(directory_to_process)
- files_list = []
- for dirpath, dirnames, filenames in os.walk(directory_to_process):
- files_list.extend(filenames)
-
- os.chdir(directory_to_process) # change to the working directory
-
- for file_to_process in filenames:
-
- ## ======= this is the sha256 checksum ========= #
- #with open(file_to_process,"rb") as f:
- # # Read and update hash in chunks of 4K
- # for byte_block in iter(lambda: f.read(4096),b""):
- # sha256_hash.update(byte_block)
- # hashed_file = sha256_hash.hexdigest()
-
- with open(file_to_process,"rb") as f:
- # Read and update hash in chunks of 4K
- for byte_block in iter(lambda: f.read(4096),b""):
- md5_hash.update(byte_block)
- hashed_file = md5_hash.hexdigest()
-
- hashed_text_note.write(hashed_file)
-
- return
-
-def md5(fname):
- md5_hash = hashlib.md5()
- with open(fname,"rb") as f:
- # Read and update hash in chunks of 4K
- for byte_block in iter(lambda: f.read(4096),b""):
- md5_hash.update(byte_block)
- return md5_hash.hexdigest()
diff --git a/workflow_parallel_frame_prediction/DataExtraction/main_single_master.py b/workflow_parallel_frame_prediction/DataExtraction/main_single_master.py
deleted file mode 100644
index fda72c67..00000000
--- a/workflow_parallel_frame_prediction/DataExtraction/main_single_master.py
+++ /dev/null
@@ -1,257 +0,0 @@
-from mpi4py import MPI
-from os import walk
-import sys
-import subprocess
-import logging
-import time
-import shutil
-import glob
-import argparse
-import os
-
-
-from helper_single_master import directory_scanner
-from helper_single_master import load_distributor
-from helper_single_master import hash_directory
-from helper_single_master import data_structure_builder
-from helper_single_master import md5
-
-from prepare_era5_data import prepare_era5_data_one_file
-
-# How to Run it!
-# mpirun -np 6 python mpi_stager_v2.py
-# mpiexec -np 6 python mpi_stager_v2.py
-
-
-def main():
- parser=argparse.ArgumentParser()
- parser.add_argument("--job_id",type=int,default=100)
- parser.add_argument("--source_dir",type=str,default="//home/a.mozaffari/data_era5/2017/")
- parser.add_argument("--destination_dir",type=str,default="/home/a.mozaffari/data_dest/")
- parser.add_argument("--log_temp",type=str,default="log_temp")
- parser.add_argument("--checksum_status",type=int,default = 0)
- parser.add_argument("--rsync_status",type=int,default=0)
- parser.add_argument("--load_level",type=int,default=0)
- parser.add_argument("--clear_destination",type=int,default=1)
- args = parser.parse_args()
- # for the local machine test
- current_path = os.getcwd()
- job_id = args.job_id
- source_dir = args.source_dir
- destination_dir = args.destination_dir
- checksum_status = args.checksum_status
- rsync_status = args.rsync_status
- clear_destination = args.clear_destination
- log_temp = args.log_temp
-
-
- # for the local machine test
- current_path = os.path.dirname(os.path.abspath(__file__))
- os.chdir(current_path)
- time.sleep(0)
-
-# ini. MPI
- comm = MPI.COMM_WORLD
- my_rank = comm.Get_rank() # rank of the node
- p = comm.Get_size() # number of assigned nods
- firs_slave_processor_id = 1
-
-
- # ==================================== Master Logging ==================================================== #
- # DEBUG: Detailed information, typically of interest only when diagnosing problems.
- # INFO: Confirmation that things are working as expected.
- # WARNING: An indication that something unexpected happened, or indicative of some problem in the near
- # ERROR: Due to a more serious problem, the software has not been able to perform some function.
- # CRITICAL: A serious error, indicating that the program itself may be unable to continue running.
- # It will copy the logging messages to the stdout, for the case of container version on HPC
-
- if my_rank == 0: # node is master
-
- # delete the general logger if exist
- logger_path = current_path + '/distribution_job_{job_id}.log'.format(job_id=job_id)
- if os.path.isfile(logger_path):
- print("Logger Exists -> Logger Deleted")
- os.remove(logger_path)
- logging.basicConfig(filename='distribution_job_{job_id}.log'.format(job_id=job_id), level=logging.DEBUG,
- format='%(asctime)s:%(levelname)s:%(message)s')
- logger = logging.getLogger(__file__)
- logger.addHandler(logging.StreamHandler(sys.stdout))
- start = time.time() # start of the MPI
-
-# check the existence of the source path :
- if not os.path.exists(source_dir): # check if the source dir. is existing
- if my_rank == 0:
- logger.critical('The source does not exist')
- message_out = "Source : {source} is not existing -> Abort".format(source=source_dir)
- logger.info('exit status : 1')
- sys.exit(1)
-
-# Check if the destination is existing, if so, it will delete and recreate the destination_dir
- if os.path.exists(destination_dir):
- if my_rank == 0:
- logger.info('The destination exist')
- if clear_destination == 1:
- shutil.rmtree(destination_dir)
- os.mkdir(destination_dir)
- logger.critical("Destination : {destination} exist -> Remove and Re-Create".format(destination=destination_dir))
- print("Destination : {destination} exist -> Remove and Re-Create".format(destination=destination_dir))
-
- else:
- logger.critical("Destination : {destination} exist -> will not be removed (caution : overwrite)".format(destination=destination_dir))
- print("Destination : {destination} exist -> will not be rmeoved (caution : overwrite)".format(destination=destination_dir))
-
-
-
- # 20200630 +++ Scarlet
- else:
- if my_rank == 0:
- os.makedirs(destination_dir) #, exist_ok=True)
- logger.info("Destination : {destination} does not exist -> Create".format(destination=destination_dir))
- print("Destination : {destination} does not exist -> Create".format(destination=destination_dir))
-
- # 20200630 --- Scarlet
-
-
- # Create a log folder for slave-nodes to write down their processes
- slave_log_path = os.path.join(destination_dir,log_temp)
-
- if my_rank == 0:
- if os.path.exists(slave_log_path) == False:
- # 20200630 Scarlet
- #os.mkdir(slave_log_path)
- os.makedirs(slave_log_path)
-
- if my_rank == 0: # node is master
-
- # ==================================== Master : Directory scanner {Parent level load level = 0} ================================= #
-
- logger.info("The source path is : {path}".format(path=source_dir))
- logger.info("The destination path is : {path}".format(path=destination_dir))
- logger.info("==== Directory scanner : start ====")
- load_level = 0
- ret_dir_scanner = directory_scanner(source_dir,load_level)
- #print(ret_dir_scanner)
-
- # Unifying the naming of this section for both cases : Sub - Directory or File
- # dir_detail_list == > Including the name of the directories, size and number of teh files in each directory / for files is empty
- # list_items_to_process === > List of items to process (Sub-Directories / Files)
- # total_size_source === > Total size of the items to process
- # total_num_files === > for Sub - Directories : sum of all files in different directories / for Files is sum of all
- # total_num_directories === > for Files = 0
-
- dir_detail_list = ret_dir_scanner[0]
- list_items_to_process = ret_dir_scanner[1]
- total_size_source = ret_dir_scanner[2]
- total_num_files = ret_dir_scanner[3]
- total_num_dir = ret_dir_scanner[4]
- logger.info("==== Directory scanner : end ====")
-
- # ================================= Master : Data Structure Builder {Parent level load level = 0} ========================= #
-
- logger.info("==== Data Structure Builder : start ====")
- data_structure_builder(source_dir, destination_dir, dir_detail_list, list_items_to_process,load_level)
- logger.info("==== Data Structure Builder : end ====")
- # message to inform the slaves that they will recive #Batch of messages including the logger_p
- batch_info = list_items_to_process
- for slaves in range (1,p):
- comm.send(batch_info, dest=slaves)
-
- for batch_counter in range (0,len(batch_info)):
- #relative_source = source_dir + str(batch_info[batch_counter]) +"/"
- relative_source = os.path.join(source_dir,str(batch_info[batch_counter]))
- print(relative_source)
- logger.info("MA{my_rank}: Next to be processed is {task} loacted in {path} ".format(my_rank = my_rank,task=batch_info[batch_counter], path=relative_source))
- load_level = 1 # it will process the files in the relative source
-
- #________ Directory Scanner ______#
- relative_ret_dir_scanner = directory_scanner(relative_source,load_level)
- relative_dir_detail_list = relative_ret_dir_scanner[0]
- relative_list_items_to_process = relative_ret_dir_scanner[1]
- relative_total_size_source = relative_ret_dir_scanner[2]
- relative_total_num_files = relative_ret_dir_scanner[3]
- relative_total_num_dir = relative_ret_dir_scanner[4]
- #________ Load Distribution ________#
- relative_ret_load_balancer = load_distributor(relative_dir_detail_list, relative_list_items_to_process, relative_total_size_source, relative_total_num_files, relative_total_num_dir,load_level, p)
- relative_transfer_dict = relative_ret_load_balancer
- logger.info(relative_transfer_dict)
-
- #________ Communication ________#
-
- for processor in range(firs_slave_processor_id, p):
- broadcast_list = relative_transfer_dict[processor]
- comm.send(broadcast_list, dest=processor)
-
- receive_counter = 0
- total_number_messages = (p-1) * len(batch_info) - 1
- while receive_counter <= total_number_messages:
- message_in = comm.recv()
- logger.info("MA{my_rank}: S{message_in} ".format(my_rank=my_rank,message_in=message_in))
- receive_counter = receive_counter + 1
-
-
- # Cleaning up the slaves temprory log file, if it is empty.
- if len(os.listdir(slave_log_path) ) == 0:
- print("Temprory log file is empty, it is deleted")
- os.removedirs(slave_log_path)
-
-
- end = time.time()
- termination_message = "MA{my_rank}: Sucssfully terminated with total time : {wall_time}".format(my_rank=my_rank,wall_time= end-start)
- logger.info(termination_message)
- sys.exit(0)
-
- else: # Processor is slave
-
- # ============================================= Slave : Send / Receive ============================================ #
- # recive the #Batch process that will be recived
- batch_info = comm.recv(source = 0)
- #print("S{my_rank} will receive {todo_message} batch of task to process".format(my_rank=my_rank, todo_message=len(batch_info)))
- batch_counter = 0
-
- # here will be a loop around all the #batchs
-
- while batch_counter <= len(batch_info) -1:
- message_in = comm.recv(source = 0)
- relative_source_directory = os.path.join(source_dir,str(batch_info[batch_counter]))
- relative_destination_directory = os.path.join(destination_dir,str(batch_info[batch_counter]))
-
- if message_in is None: # in case more than number of the dir. processor is assigned !
- slave_out_message = "{my_rank} is idle".format(my_rank=my_rank)
- # comm.send(message_out, dest=1)
-
- else: # if the Slave node has joblist to do
- job_list = message_in.split(';')
- for job_count in range(0, len(job_list)):
- job = job_list[job_count] # job is the name of the directory(ies) assigned to slave_node
- #print(job)
- if rsync_status == 1:
- # prepare the rsync commoand to be excexuted by the worker node
- rsync_message = "rsync {relative_source_directory}/{job} {relative_destination_directory}/{job}".format(relative_source_directory=relative_source_directory,job=job, relative_destination_directory=relative_destination_directory)
- os.system(rsync_message)
- #slave_out_message= " RSYNC process"
- else :
- ## @Bing here is the job for the slaves
- print("S{my_rank} will execute era5 preperation on {job}".format(my_rank=my_rank, job=job))
- prepare_era5_data_one_file(src_file=job,directory_to_process=relative_source_directory, target=job, target_dir=relative_destination_directory)
-
-
-
- #if job.endswith(".nc"):
- # if os.path.exists(os.path.join(relative_destination_directory, job)):
- # print("{job} is has been processed in directory {directory}".format(job=job,directory=relative_destination_directory))
- #else:
- # prepare_era5_data_one_file(src_file=job,directory_to_process=relative_source_directory, target=job, target_dir=relative_destination_directory)
- # print("File {job} in directory {directory} has been processed in directory".format(job=job,directory=relative_destination_directory))
- #
- #slave_out_message = " {in_message} process".format(in_message=my_rank)
- # Generate a hash of the output
-
- message_out = "{my_rank}: is finished the {in_message} .".format(my_rank=my_rank,in_message=batch_info[batch_counter])
- comm.send(message_out, dest=0)
- batch_counter = batch_counter + 1
-
- MPI.Finalize()
-
-
-if __name__ == "__main__":
- main()
diff --git a/workflow_parallel_frame_prediction/DataExtraction/readme.md b/workflow_parallel_frame_prediction/DataExtraction/readme.md
deleted file mode 100644
index 9e97dae8..00000000
--- a/workflow_parallel_frame_prediction/DataExtraction/readme.md
+++ /dev/null
@@ -1,2 +0,0 @@
-`source create_env_zam347.sh {MPI}` <br/>
-`mpirun -np {number of processors max 13 on zam347} python main_single_master.py`
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/DataExtraction/submitJob.sh b/workflow_parallel_frame_prediction/DataExtraction/submitJob.sh
deleted file mode 100755
index 2e55e377..00000000
--- a/workflow_parallel_frame_prediction/DataExtraction/submitJob.sh
+++ /dev/null
@@ -1,22 +0,0 @@
-#!/bin/bash -x
-#SBATCH --account=deepacf
-#SBATCH --nodes=1
-#SBATCH --ntasks=1
-##SBATCH --ntasks-per-node=1
-#SBATCH --cpus-per-task=1
-#SBATCH --output=pystager-out.%j
-#SBATCH --error=pystager-err.%j
-#SBATCH --time=00:20:00
-#SBATCH --partition=devel
-#SBATCH --mail-type=ALL
-#SBATCH --mail-user=b.gong@fz-juelich.de
-##jutil env activate -p deepacf
-
-module --force purge
-module /usr/local/software/jureca/OtherStages
-module load Stages/2019a
-module load Intel/2019.3.199-GCC-8.3.0 ParaStationMPI/5.2.2-1
-module load mpi4py/3.0.1-Python-3.6.8
-
-#srun python mpi_stager_v2.py --source_dir /p/fastdata/slmet/slmet111/met_data/ecmwf/era5/#nc/2017/ --destination_dir /p/scratch/deepacf/bing/extractedData
-srun python Extract_data_for_testing.py
diff --git a/workflow_parallel_frame_prediction/DataPostprocess/Stager_devel_N_24_evaluation.sh b/workflow_parallel_frame_prediction/DataPostprocess/Stager_devel_N_24_evaluation.sh
deleted file mode 100755
index 1aa0ae9a..00000000
--- a/workflow_parallel_frame_prediction/DataPostprocess/Stager_devel_N_24_evaluation.sh
+++ /dev/null
@@ -1,40 +0,0 @@
-#!/bin/bash
-#SBATCH --account=jjsc42
-# budget account where contingent is taken from# TASKS = NODES * GPUS_PER_NODE
-#SBATCH --nodes=3
-#SBATCH --ntasks-per-node=4
-#SBATCH --ntasks=12
-# can be omitted if --nodes and --ntasks-per-node
-# are given
-# SBATCH --cpus-per-task=1
-# for OpenMP/hybrid jobs only
-#SBATCH --output=horovod-4ntasks%j.out
-# if keyword omitted: Default is slurm-%j.out in
-# the submission directory (%j is replaced by
-# the job ID).
-#SBATCH --error=horovod-4ntasks%j.err
-# if keyword omitted: Default is slurm-%j.out in
-# the submission directory.
-#SBATCH --time=20:00:00
-#SBATCH --gres=gpu:4
-#SBATCH --partition=gpus
-#SBATCH --mail-user=b.gong@fz-juelich.de
-#SBATCH --mail-type=ALL
-
-#create a folder to save the output
-
-module --force purge
-module --force purge
-module load Stages/Devel-2019a
-module load GCC/8.3.0
-module load MVAPICH2/2.3.2-GDR
-#module /usr/local/software/jureca/OtherStages
-module load Stages/2019a
-module load GCCcore/.8.3.0
-module load cuDNN/7.5.1.10-CUDA-10.1.105
-module load Horovod/0.16.2-GPU-Python-3.6.8
-module load Keras/2.2.4-GPU-Python-3.6.8
-#module load Intel/2019.3.199-GCC-8.3.0 ParaStationMPI/5.2.2-1-mt
-#module load mpi4py/3.0.1-Python-3.6.8
-
-srun python3.6 kitti_evaluate_parallel.py
diff --git a/workflow_parallel_frame_prediction/DataPostprocess/data_utils.py b/workflow_parallel_frame_prediction/DataPostprocess/data_utils.py
deleted file mode 100755
index 8751fc42..00000000
--- a/workflow_parallel_frame_prediction/DataPostprocess/data_utils.py
+++ /dev/null
@@ -1,164 +0,0 @@
-import hickle as hkl
-import numpy as np
-from keras import backend as K
-from keras.preprocessing.image import Iterator
-
-
-import inspect
-print(inspect.getmembers(hkl,predicate=inspect.ismethod))
-
-# Data generator that creates sequences for input into PredNet.
-class SequenceGenerator(Iterator):
- def __init__(self, data_file, source_file, nt,
- batch_size=8, shuffle=False, seed=None,
- output_mode='error', sequence_start_mode='all', N_seq=None,
- data_format=K.image_data_format()):
- self.X = hkl.load(data_file) # X will be like (n_images, nb_cols, nb_rows, nb_channels)
- self.sources = hkl.load(source_file) # source for each image so when creating sequences can assure that consecutive frames are from same video
- self.nt = nt
- self.batch_size = batch_size
- self.data_format = data_format
- assert sequence_start_mode in {'all', 'unique'}, 'sequence_start_mode must be in {all, unique}'
- self.sequence_start_mode = sequence_start_mode
- assert output_mode in {'error', 'prediction'}, 'output_mode must be in {error, prediction}'
- self.output_mode = output_mode
-
- if self.data_format == 'channels_first':
- self.X = np.transpose(self.X, (0, 3, 1, 2))
- self.im_shape = self.X[0].shape
-
- if self.sequence_start_mode == 'all': # allow for any possible sequence, starting from any frame
- #bing
- #self.possible_starts = np.array([i for i in range(self.X.shape[0] - self.nt) if self.sources[i] == self.sources[i + self.nt - 1]])
- self.possible_starts = np.array([i for i in range(self.X.shape[0] - self.nt)])
- elif self.sequence_start_mode == 'unique': #create sequences where each unique frame is in at most one sequence
- curr_location = 0
- possible_starts = []
- while curr_location < self.X.shape[0] - self.nt + 1:
- if self.sources[curr_location] == self.sources[curr_location + self.nt - 1]:
- possible_starts.append(curr_location)
- curr_location += self.nt
- else:
- curr_location += 1
- self.possible_starts = possible_starts
-
- if shuffle:
- self.possible_starts = np.random.permutation(self.possible_starts)
- if N_seq is not None and len(self.possible_starts) > N_seq: # select a subset of sequences if want to
- self.possible_starts = self.possible_starts[:N_seq]
- self.N_sequences = len(self.possible_starts)
- print("N_sequences", self.N_sequences)
- super(SequenceGenerator, self).__init__(len(self.possible_starts), batch_size, shuffle, seed)
-
- def __getitem__(self, null):
- return self.next()
-
- def next(self):
- with self.lock:
- current_index = (self.batch_index * self.batch_size) % self.n
- index_array, current_batch_size = next(self.index_generator), self.batch_size
- batch_x = np.zeros((current_batch_size, self.nt) + self.im_shape, np.float32)
- for i, idx in enumerate(index_array):
- idx = self.possible_starts[idx]
- batch_x[i] = self.preprocess(self.X[idx:idx+self.nt])
- if self.output_mode == 'error': # model outputs errors, so y should be zeros
- batch_y = np.zeros(current_batch_size, np.float32)
- elif self.output_mode == 'prediction': # output actual pixels
- batch_y = batch_x
- return batch_x, batch_y
-
- def preprocess(self, X):
- ### Normalization after extrema cut off: ###
- #cut maxs & mins to mean+3*std & mean-3*std of training set for each parameter
- #x_cut = np.zeros(shape=X.shape)
- #x_cut = X*1 #pass X by value and not by reference
- #x_cut[:,:,:,0][X[:,:,:,0]>311.5]=311.5 #set T2 upper limit
- #x_cut[:,:,:,0][X[:,:,:,0]<258.9]=258.9 #set T2 lower limit
- #x_cut[:,:,:,1][X[:,:,:,1]>104635.2]=104635.2 #set GP upper limit
- #x_cut[:,:,:,1][X[:,:,:,1]<98205.6]=98205.6 #set GP lower limit ###Caution: Drastical cut ###
- #x_cut[:,:,:,2][X[:,:,:,2]>6209.5]=6209.5 #set GPH upper limit ###Caution: Unnecessary as it succeeds max GPH ###
- #x_cut[:,:,:,2][X[:,:,:,2]<5005.8]=5005.8 #set GPH lower limit
- #normalize X based on max and min values(equals upper and lower limits except highCutGPH)
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (x_cut[:,:,:,0]-258.9)/(311.5-258.9)
- #x_processed[:,:,:,1] = (x_cut[:,:,:,1]-98205.6)/(104635.2-98205.6)
- #x_processed[:,:,:,2] = (x_cut[:,:,:,2]-5005.8)/(6007.097417091836-5005.8) #GPH max stays; see above
-
- ### 'Standard' normalization: (x-min(x))/(max(x)-min(x)) ###
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-93401.125)/(105391.4375-93401.125)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-4836.070232780612)/(6007.097417091836-4836.070232780612)
-
- ### t2only 'Standard' normalization: (x-min(x))/(max(x)-min(x)) ###
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-235.2141571044922)/(321.46630859375-235.2141571044922)
-
- ### t2_2MSL_1 'standard' normalization:
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-93401.125)/(105391.4375-93401.125)
-
- ### t2_1MSL_2 'standard' normalization:
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-93401.125)/(105391.4375-93401.125)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-93401.125)/(105391.4375-93401.125)
-
- ### t2_2gph500_1 'standard' normalization:
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-4836.070232780612)/(6007.097417091836-4836.070232780612)
- ## t2_1gph500_2 'standard' normalization:
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-4836.070232780612)/(6007.097417091836-4836.070232780612)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-4836.070232780612)/(6007.097417091836-4836.070232780612)
-
- ### No standardization for moving Objects test set: Just 0s and 1s
- #x_processed = np.zeros(shape=X.shape)
- #x_processed = X
-
- ### t2_1 'standard' normalization (got one dimension less, due to just one channel)
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = X[:,:,:,1]
- #x_processed[:,:,:,2] = X[:,:,:,2]
-
- ### t2_1 'standard' normalization (got one dimension less, due to just one channel)
- x_processed = np.zeros(shape=X.shape)
- x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- x_processed[:,:,:,1] = (X[:,:,:,1]-235.2141571044922)/(321.46630859375-235.2141571044922)
- x_processed[:,:,:,2] = X[:,:,:,2]
-
- ### Standardization: (x-mean)/standard_deviation ###
- #Doesn't work due to some activation functions
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-285.1751264870658)/8.770013367617763
- #x_processed[:,:,:,1] = (X[:,:,:,1]-101420.4382666807)/1071.5999818175521
- #x_processed[:,:,:,2] = (X[:,:,:,2]-5607.662795353248)/200.62593105865764
-
- ### Standardization+Normalization ###
- # standardize:(x-mean)/standard_deviation
- #x_preprocessed = np.zeros(shape=X.shape)
- #x_preprocessed[:,:,:,0] = (X[:,:,:,0]-285.1751264870658)/8.770013367617763
- #x_preprocessed[:,:,:,1] = (X[:,:,:,1]-101420.4382666807)/1071.5999818175521
- #x_preprocessed[:,:,:,2] = (X[:,:,:,2]-5607.662795353248)/200.62593105865764
- # normalize:(x-min(x))/(max(x)-min(x))
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (x_preprocessed[:,:,:,0]-np.amin(x_preprocessed[:,:,:,0]))/(np.amax(x_preprocessed[:,:,:,0])-np.amin(x_preprocessed[:,:,:,0]))
- #x_processed[:,:,:,1] = (x_preprocessed[:,:,:,1]-np.amin(x_preprocessed[:,:,:,1]))/(np.amax(x_preprocessed[:,:,:,1])-np.amin(x_preprocessed[:,:,:,1]))
- #x_processed[:,:,:,2] = (x_preprocessed[:,:,:,2]-np.amin(x_preprocessed[:,:,:,2]))/(np.amax(x_preprocessed[:,:,:,2])-np.amin(x_preprocessed[:,:,:,2]))
-
- return x_processed.astype(np.float32)
- #return X.astype(np.float32) / 255
-
- def create_all(self):
- X_all = np.zeros((self.N_sequences, self.nt) + self.im_shape, np.float32)
- for i, idx in enumerate(self.possible_starts):
- X_all[i] = self.preprocess(self.X[idx:idx+self.nt])
- return X_all
diff --git a/workflow_parallel_frame_prediction/DataPostprocess/keras_utils.py b/workflow_parallel_frame_prediction/DataPostprocess/keras_utils.py
deleted file mode 100755
index ededcc74..00000000
--- a/workflow_parallel_frame_prediction/DataPostprocess/keras_utils.py
+++ /dev/null
@@ -1,58 +0,0 @@
-import os
-import numpy as np
-
-from keras import backend as K
-from keras.legacy.interfaces import generate_legacy_interface, recurrent_args_preprocessor
-from keras.models import model_from_json
-
-legacy_prednet_support = generate_legacy_interface(
- allowed_positional_args=['stack_sizes', 'R_stack_sizes',
- 'A_filt_sizes', 'Ahat_filt_sizes', 'R_filt_sizes'],
- conversions=[('dim_ordering', 'data_format'),
- ('consume_less', 'implementation')],
- value_conversions={'dim_ordering': {'tf': 'channels_last',
- 'th': 'channels_first',
- 'default': None},
- 'consume_less': {'cpu': 0,
- 'mem': 1,
- 'gpu': 2}},
- preprocessor=recurrent_args_preprocessor)
-
-# Convert old Keras (1.2) json models and weights to Keras 2.0
-def convert_model_to_keras2(old_json_file, old_weights_file, new_json_file, new_weights_file):
- from prednet import PredNet
- # If using tensorflow, it doesn't allow you to load the old weights.
- if K.backend() != 'theano':
- os.environ['KERAS_BACKEND'] = backend
- reload(K)
-
- f = open(old_json_file, 'r')
- json_string = f.read()
- f.close()
- model = model_from_json(json_string, custom_objects = {'PredNet': PredNet})
- model.load_weights(old_weights_file)
-
- weights = model.layers[1].get_weights()
- if weights[0].shape[0] == model.layers[1].stack_sizes[1]:
- for i, w in enumerate(weights):
- if w.ndim == 4:
- weights[i] = np.transpose(w, (2, 3, 1, 0))
- model.set_weights(weights)
-
- model.save_weights(new_weights_file)
- json_string = model.to_json()
- with open(new_json_file, "w") as f:
- f.write(json_string)
-
-
-if __name__ == '__main__':
- old_dir = './model_data/'
- new_dir = './model_data_keras2/'
- if not os.path.exists(new_dir):
- os.mkdir(new_dir)
- for w_tag in ['', '-Lall', '-extrapfinetuned']:
- m_tag = '' if w_tag == '-Lall' else w_tag
- convert_model_to_keras2(old_dir + 'prednet_kitti_model' + m_tag + '.json',
- old_dir + 'prednet_kitti_weights' + w_tag + '.hdf5',
- new_dir + 'prednet_kitti_model' + m_tag + '.json',
- new_dir + 'prednet_kitti_weights' + w_tag + '.hdf5')
diff --git a/workflow_parallel_frame_prediction/DataPostprocess/kitti_evaluate_parallel.py b/workflow_parallel_frame_prediction/DataPostprocess/kitti_evaluate_parallel.py
deleted file mode 100755
index 1c1166ee..00000000
--- a/workflow_parallel_frame_prediction/DataPostprocess/kitti_evaluate_parallel.py
+++ /dev/null
@@ -1,159 +0,0 @@
-'''
-Evaluate trained PredNet on KITTI sequences.
-Calculates mean-squared error and plots predictions.
-'''
-import math
-import os
-import numpy as np
-#from six.moves import cPickle
-import matplotlib
-matplotlib.use('Agg')
-import matplotlib.pyplot as plt
-import matplotlib.gridspec as gridspec
-from keras import backend as K
-from keras.models import Model, model_from_json
-from keras.layers import Input, Dense, Flatten
-from prednet import PredNet
-from data_utils import SequenceGenerator
-from kitti_settings import *
-##Just for checking how the shape is after generator.create_all() from Sequence Generator
-#import hickle as hkl
-import horovod.keras as hvd
-import tensorflow as tf
-
-#Horovod:initialize horovod
-hvd.init()
-#Horovod: pin GPU to be used for process local rank (one GPU per process)
-config = tf.ConfigProto()
-config.gpu_options.allow_growth = True
-config.gpu_options.visible_device_list = str(hvd.local_rank())
-K.set_session(tf.Session(config=config))
-
-n_plot = 10
-batch_size = 5
-nt = 10
-
-##Configure the horovod
-
-#weights_file = os.path.join(WEIGHTS_DIR, 'tensorflow_weights/prednet_kitti_weights.hdf5')
-weights_file = os.path.join(WEIGHTS_DIR, 'prednet_kitti_weights.hdf5')
-print("weights loaded")
-json_file = os.path.join(WEIGHTS_DIR, 'prednet_kitti_model.json')
-print("model loaded")
-test_file = os.path.join(DATA_DIR, 'X_train.hkl')
-print("test file loaded")
-test_sources = os.path.join(DATA_DIR, 'sources_test.hkl')
-print("test source loaded")
-
-# Load trained model
-f = open(json_file, 'r')
-json_string = f.read()
-f.close()
-train_model = model_from_json(json_string, custom_objects = {'PredNet': PredNet})
-train_model.load_weights(weights_file)
-print("Trained model is loaded")
-
-# Create testing model (to output predictions)
-layer_config = train_model.layers[1].get_config()
-layer_config['output_mode'] = 'prediction'
-data_format = layer_config['data_format'] if 'data_format' in layer_config else layer_config['dim_ordering']
-test_prednet = PredNet(weights=train_model.layers[1].get_weights(), **layer_config)
-input_shape = list(train_model.layers[0].batch_input_shape[1:])
-input_shape[0] = nt
-inputs = Input(shape=tuple(input_shape))
-predictions = test_prednet(inputs)
-test_model = Model(inputs=inputs, outputs=predictions)
-
-#Bing: change the unique to all,need to ask severin why training data use all , but for testing use unique
-test_generator = SequenceGenerator(test_file, test_sources, nt, sequence_start_mode='all', data_format=data_format)
-print ("test_generator finished")
-X_test = test_generator.create_all()
-print(X_test.shape)
-#Bing, replace this part with horovod because of the memeory issue
-X_test_batch_size = round(X_test.shape[0]/hvd.size())
-
-def post_process(rank = 0):
- print ("Rank {}".format(rank))
- X_test_batch = X_test[rank * X_test_batch_size:(rank + 1) * X_test_batch_size]
- print ("X_test_batch size",X_test_batch.shape)
- X_hat = test_model.predict(X_test_batch, batch_size)
-
- if data_format == 'channels_first':
- X_test_batch = np.transpose(X_test_batch, (0, 1, 3, 4, 2))
- X_hat = np.transpose(X_hat, (0, 1, 3, 4, 2))
-
- ##Just for checking how the shape is after generator.create_all() from Sequence Generator
- #hkl.dump(X_test, os.path.join(RESULTS_SAVE_DIR, 'X_AfterGeneratorStandardized.hkl'))
- #hkl.dump(X_hat, os.path.join(RESULTS_SAVE_DIR, 'X_hatStandardized.hkl'))
- #
-
- # Compare MSE of PredNet predictions vs. using last frame. Write results to prediction_scores.txt
- # Furthermore, calculate Model MSE from the last prediction of the sequence only
- # as the model improves after several frames (mse_model_last)
- # Typical shape of X_test and X_hat: (263, 10, 128, 160, 3)
- # where 263 are the sequences, 10 ist the amount of frames in one sequence,
- # 128 & 160 are the image sice and 3 the number of layers.
- # For our case only take layer 0 (= T2) into account.
- shapeXhat = str(X_hat.shape) #Just have a look at the shapes to be sure we are calculating the right MSE
- shapeXtest = str(X_test_batch.shape)
- mse_model = np.mean((X_test_batch[:, 1:,:,:,0] - X_hat[:, 1:,:,:,0])**2) # look at all timesteps except the first
- mse_model_last = np.mean((X_test_batch[:, 9,:,:,0] - X_hat[:, 9,:,:,0])**2 )
- mse_prev = np.mean((X_test_batch[:, :-1,:,:,0] - X_test_batch[:, 1:,:,:,0])**2)
-
- # Calculate PSNR
- # Function to calculate PSNR
- # In the absence of noise, the two images I and K are identical, and thus the MSE is zero. In this case the PSNR is infinite.
- # Or here the best value: 100
- def psnr(img1, img2):
- mse = np.mean((img1-img2) ** 2)
- if mse == 0: return 100
- PIXEL_MAX = 1
- return 20 * math.log10(PIXEL_MAX/math.sqrt(mse))
-
- psnr_model = psnr(X_test_batch[:, 1:,:,:,0], X_hat[:, 1:,:,:,0])
- psnr_model_last = psnr(X_test_batch[:, 9,:,:,0], X_hat[:, 9,:,:,0])
- psnr_prev = psnr(X_test_batch[:, :-1,:,:,0], X_test_batch[:, 1:,:,:,0])
- print("Evaluations are completed")
-
- if not os.path.exists(RESULTS_SAVE_DIR): os.mkdir(RESULTS_SAVE_DIR)
- f = open(RESULTS_SAVE_DIR + '/prediction_scores_rank_{}.txt'.format(rank), 'w')
-
- f.write("X_test_batch_size:{} ; Rank: {}\n".format(X_test_batch_size,hvd.rank()))
- f.write("Model MSE: %f\n" % mse_model)
- f.write("Model MSE from only last prediction in sequence: %f\n" % mse_model_last)
- f.write("Previous Frame MSE: %f\n" % mse_prev)
- f.write("Model PSNR: %f\n" % psnr_model)
- f.write("Model PSNR from only last prediction in sequence: %f\n" % psnr_model_last)
- f.write("Previous frame PSNR: %f\n" % psnr_prev)
- f.write("Shape of X_test: " + shapeXtest)
- f.write("")
- f.write("Shape of X_hat: " + shapeXhat)
- f.close()
- print("Results are saved to {}\n".format(RESULTS_SAVE_DIR + "/prediction_scores_rank_{}.txt".format(rank)))
-
- # Plot some predictions
- aspect_ratio = float(X_hat.shape[2]) / X_hat.shape[3]
- plt.figure(figsize = (nt, 2*aspect_ratio))
- gs = gridspec.GridSpec(2, nt)
- gs.update(wspace=0., hspace=0.)
- plot_save_dir = os.path.join(RESULTS_SAVE_DIR, 'prediction_plots_rank_{}/'.format(rank))
- if not os.path.exists(plot_save_dir): os.mkdir(plot_save_dir)
- plot_idx = np.random.permutation(X_test_batch.shape[0])[:n_plot]
- for i in plot_idx:
- for t in range(nt):
- plt.subplot(gs[t])
- plt.imshow(X_test_batch[i,t,:,:,0], interpolation='none') #the last index sets the channel. 0 = t2
- #plt.pcolormesh(X_test[i,t,::-1,:,0], shading='bottom', cmap=plt.cm.jet)
- plt.tick_params(axis='both', which='both', bottom=False, top=False, left=False, right=False, labelbottom=False, labelleft=False)
- if t==0: plt.ylabel('Actual', fontsize=10)
-
- plt.subplot(gs[t + nt])
- plt.imshow(X_hat[i,t,:,:,0], interpolation='none')
- #plt.pcolormesh(X_hat[i,t,::-1,:,0], shading='bottom', cmap=plt.cm.jet)
- plt.tick_params(axis='both', which='both', bottom=False, top=False, left=False, right=False, labelbottom=False, labelleft=False)
- if t==0: plt.ylabel('Predicted', fontsize=10)
- plt.savefig(plot_save_dir + 'plot_' + str(i) + '.png')
- plt.clf()
-
-
-post_process(rank = hvd.rank())
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/DataPostprocess/kitti_settings.py b/workflow_parallel_frame_prediction/DataPostprocess/kitti_settings.py
deleted file mode 100755
index 489989ee..00000000
--- a/workflow_parallel_frame_prediction/DataPostprocess/kitti_settings.py
+++ /dev/null
@@ -1,19 +0,0 @@
-# Where KITTI data will be saved if you run process_kitti.py
-# If you directly download the processed data, change to the path of the data.
-## Changed logic: Now this is the path where the processed data lies: X_train,val,test
-#DATA_DIR = './kitti_data/'
-#data directory for training data 2015 and 2016
-#DATA_DIR = '/p/project/cjjsc42/severin/try3'
-#data directory for moving objects:
-#DATA_DIR = '/p/home/jusers/hussmann1/jureca/movingObjects/se_nw'
-#data directory for featuretesting:
-##DATA_DIR = './testTry2'
-DATA_DIR = '/p/scratch/cjjsc42/bing/pystager-development/processData/splits'
-# Where model weights and config will be saved if you run kitti_train.py
-# If you directly download the trained weights, change to appropriate path.
-WEIGHTS_DIR = '/p/project/cjjsc42/bing/ml-severin/model_data_keras2'
-#WEIGHTS_DIR = '/p/project/cjjsc42/bing/ml-severin/model_data_keras2'
-
-# Where results (prediction plots and evaluation file) will be saved.
-RESULTS_SAVE_DIR = '/p/project/cjjsc42/bing/ml-severin/kitti_results'
-
diff --git a/workflow_parallel_frame_prediction/DataPostprocess/packageInstallation.sh b/workflow_parallel_frame_prediction/DataPostprocess/packageInstallation.sh
deleted file mode 100644
index 1039e5fd..00000000
--- a/workflow_parallel_frame_prediction/DataPostprocess/packageInstallation.sh
+++ /dev/null
@@ -1,9 +0,0 @@
-module --force purge
-module /usr/local/software/jureca/OtherStages
-module load Stages/2019a
-module load Intel/2019.3.199-GCC-8.3.0 ParaStationMPI/5.2.2-1-mt
-module load mpi4py/3.0.1-Python-3.6.8
-pip3 install --user netCDF4
-pip3 install --user numpy
-
-
diff --git a/workflow_parallel_frame_prediction/DataPostprocess/parameters_kitti_evaluate.dat b/workflow_parallel_frame_prediction/DataPostprocess/parameters_kitti_evaluate.dat
deleted file mode 100755
index ac64b704..00000000
--- a/workflow_parallel_frame_prediction/DataPostprocess/parameters_kitti_evaluate.dat
+++ /dev/null
@@ -1,14 +0,0 @@
-import os
-
-# ============ input parameters =================== #
-# 0:deactivate 1: active
-
-Pleas fill the following parameters list for PyStager
-Source_Directory = /p/fastdata/slmet/slmet111/met_data/ecmwf/era5/nc/2017/
-Destination_Directory = /p/scratch/cjjsc42/bing/pystager-development/tryData/
-Log_Directory = /p/project/cjjsc42/bing/pystager-development/log
-Rsync_Status = 1
-Checksum_Status = 0
-
-
-
diff --git a/workflow_parallel_frame_prediction/DataPostprocess/prednet.py b/workflow_parallel_frame_prediction/DataPostprocess/prednet.py
deleted file mode 100755
index b5a0208a..00000000
--- a/workflow_parallel_frame_prediction/DataPostprocess/prednet.py
+++ /dev/null
@@ -1,311 +0,0 @@
-import numpy as np
-
-from keras import backend as K
-from keras import activations
-from keras.layers import Recurrent
-from keras.layers import Conv2D, UpSampling2D, MaxPooling2D
-from keras.engine import InputSpec
-from keras_utils import legacy_prednet_support
-
-class PredNet(Recurrent):
- '''PredNet architecture - Lotter 2016.
- Stacked convolutional LSTM inspired by predictive coding principles.
-
- # Arguments
- stack_sizes: number of channels in targets (A) and predictions (Ahat) in each layer of the architecture.
- Length is the number of layers in the architecture.
- First element is the number of channels in the input.
- Ex. (3, 16, 32) would correspond to a 3 layer architecture that takes in RGB images and has 16 and 32
- channels in the second and third layers, respectively.
- R_stack_sizes: number of channels in the representation (R) modules.
- Length must equal length of stack_sizes, but the number of channels per layer can be different.
- A_filt_sizes: filter sizes for the target (A) modules.
- Has length of 1 - len(stack_sizes).
- Ex. (3, 3) would mean that targets for layers 2 and 3 are computed by a 3x3 convolution of the errors (E)
- from the layer below (followed by max-pooling)
- Ahat_filt_sizes: filter sizes for the prediction (Ahat) modules.
- Has length equal to length of stack_sizes.
- Ex. (3, 3, 3) would mean that the predictions for each layer are computed by a 3x3 convolution of the
- representation (R) modules at each layer.
- R_filt_sizes: filter sizes for the representation (R) modules.
- Has length equal to length of stack_sizes.
- Corresponds to the filter sizes for all convolutions in the LSTM.
- pixel_max: the maximum pixel value.
- Used to clip the pixel-layer prediction.
- error_activation: activation function for the error (E) units.
- A_activation: activation function for the target (A) and prediction (A_hat) units.
- LSTM_activation: activation function for the cell and hidden states of the LSTM.
- LSTM_inner_activation: activation function for the gates in the LSTM.
- output_mode: either 'error', 'prediction', 'all' or layer specification (ex. R2, see below).
- Controls what is outputted by the PredNet.
- If 'error', the mean response of the error (E) units of each layer will be outputted.
- That is, the output shape will be (batch_size, nb_layers).
- If 'prediction', the frame prediction will be outputted.
- If 'all', the output will be the frame prediction concatenated with the mean layer errors.
- The frame prediction is flattened before concatenation.
- Nomenclature of 'all' is kept for backwards compatibility, but should not be confused with returning all of the layers of the model
- For returning the features of a particular layer, output_mode should be of the form unit_type + layer_number.
- For instance, to return the features of the LSTM "representational" units in the lowest layer, output_mode should be specificied as 'R0'.
- The possible unit types are 'R', 'Ahat', 'A', and 'E' corresponding to the 'representation', 'prediction', 'target', and 'error' units respectively.
- extrap_start_time: time step for which model will start extrapolating.
- Starting at this time step, the prediction from the previous time step will be treated as the "actual"
- data_format: 'channels_first' or 'channels_last'.
- It defaults to the `image_data_format` value found in your
- Keras config file at `~/.keras/keras.json`.
-
- # References
- - [Deep predictive coding networks for video prediction and unsupervised learning](https://arxiv.org/abs/1605.08104)
- - [Long short-term memory](http://deeplearning.cs.cmu.edu/pdfs/Hochreiter97_lstm.pdf)
- - [Convolutional LSTM network: a machine learning approach for precipitation nowcasting](http://arxiv.org/abs/1506.04214)
- - [Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects](http://www.nature.com/neuro/journal/v2/n1/pdf/nn0199_79.pdf)
- '''
- @legacy_prednet_support
- def __init__(self, stack_sizes, R_stack_sizes,
- A_filt_sizes, Ahat_filt_sizes, R_filt_sizes,
- pixel_max=1., error_activation='relu', A_activation='relu',
- LSTM_activation='tanh', LSTM_inner_activation='hard_sigmoid',
- output_mode='error', extrap_start_time=None,
- data_format=K.image_data_format(), **kwargs):
- self.stack_sizes = stack_sizes
- self.nb_layers = len(stack_sizes)
- assert len(R_stack_sizes) == self.nb_layers, 'len(R_stack_sizes) must equal len(stack_sizes)'
- self.R_stack_sizes = R_stack_sizes
- assert len(A_filt_sizes) == (self.nb_layers - 1), 'len(A_filt_sizes) must equal len(stack_sizes) - 1'
- self.A_filt_sizes = A_filt_sizes
- assert len(Ahat_filt_sizes) == self.nb_layers, 'len(Ahat_filt_sizes) must equal len(stack_sizes)'
- self.Ahat_filt_sizes = Ahat_filt_sizes
- assert len(R_filt_sizes) == (self.nb_layers), 'len(R_filt_sizes) must equal len(stack_sizes)'
- self.R_filt_sizes = R_filt_sizes
-
- self.pixel_max = pixel_max
- self.error_activation = activations.get(error_activation)
- self.A_activation = activations.get(A_activation)
- self.LSTM_activation = activations.get(LSTM_activation)
- self.LSTM_inner_activation = activations.get(LSTM_inner_activation)
-
- default_output_modes = ['prediction', 'error', 'all']
- layer_output_modes = [layer + str(n) for n in range(self.nb_layers) for layer in ['R', 'E', 'A', 'Ahat']]
- assert output_mode in default_output_modes + layer_output_modes, 'Invalid output_mode: ' + str(output_mode)
- self.output_mode = output_mode
- if self.output_mode in layer_output_modes:
- self.output_layer_type = self.output_mode[:-1]
- self.output_layer_num = int(self.output_mode[-1])
- else:
- self.output_layer_type = None
- self.output_layer_num = None
- self.extrap_start_time = extrap_start_time
-
- assert data_format in {'channels_last', 'channels_first'}, 'data_format must be in {channels_last, channels_first}'
- self.data_format = data_format
- self.channel_axis = -3 if data_format == 'channels_first' else -1
- self.row_axis = -2 if data_format == 'channels_first' else -3
- self.column_axis = -1 if data_format == 'channels_first' else -2
- super(PredNet, self).__init__(**kwargs)
- self.input_spec = [InputSpec(ndim=5)]
-
- def compute_output_shape(self, input_shape):
- if self.output_mode == 'prediction':
- out_shape = input_shape[2:]
- elif self.output_mode == 'error':
- out_shape = (self.nb_layers,)
- elif self.output_mode == 'all':
- out_shape = (np.prod(input_shape[2:]) + self.nb_layers,)
- else:
- stack_str = 'R_stack_sizes' if self.output_layer_type == 'R' else 'stack_sizes'
- stack_mult = 2 if self.output_layer_type == 'E' else 1
- out_stack_size = stack_mult * getattr(self, stack_str)[self.output_layer_num]
- out_nb_row = input_shape[self.row_axis] / 2**self.output_layer_num
- out_nb_col = input_shape[self.column_axis] / 2**self.output_layer_num
- if self.data_format == 'channels_first':
- out_shape = (out_stack_size, out_nb_row, out_nb_col)
- else:
- out_shape = (out_nb_row, out_nb_col, out_stack_size)
-
- if self.return_sequences:
- return (input_shape[0], input_shape[1]) + out_shape
- else:
- return (input_shape[0],) + out_shape
-
- def get_initial_state(self, x):
- input_shape = self.input_spec[0].shape
- init_nb_row = input_shape[self.row_axis]
- init_nb_col = input_shape[self.column_axis]
-
- base_initial_state = K.zeros_like(x) # (samples, timesteps) + image_shape
- non_channel_axis = -1 if self.data_format == 'channels_first' else -2
- for _ in range(2):
- base_initial_state = K.sum(base_initial_state, axis=non_channel_axis)
- base_initial_state = K.sum(base_initial_state, axis=1) # (samples, nb_channels)
-
- initial_states = []
- states_to_pass = ['r', 'c', 'e']
- nlayers_to_pass = {u: self.nb_layers for u in states_to_pass}
- if self.extrap_start_time is not None:
- states_to_pass.append('ahat') # pass prediction in states so can use as actual for t+1 when extrapolating
- nlayers_to_pass['ahat'] = 1
- for u in states_to_pass:
- for l in range(nlayers_to_pass[u]):
- ds_factor = 2 ** l
- nb_row = init_nb_row // ds_factor
- nb_col = init_nb_col // ds_factor
- if u in ['r', 'c']:
- stack_size = self.R_stack_sizes[l]
- elif u == 'e':
- stack_size = 2 * self.stack_sizes[l]
- elif u == 'ahat':
- stack_size = self.stack_sizes[l]
- output_size = stack_size * nb_row * nb_col # flattened size
-
- reducer = K.zeros((input_shape[self.channel_axis], output_size)) # (nb_channels, output_size)
- initial_state = K.dot(base_initial_state, reducer) # (samples, output_size)
- if self.data_format == 'channels_first':
- output_shp = (-1, stack_size, nb_row, nb_col)
- else:
- output_shp = (-1, nb_row, nb_col, stack_size)
- initial_state = K.reshape(initial_state, output_shp)
- initial_states += [initial_state]
-
- if K._BACKEND == 'theano':
- from theano import tensor as T
- # There is a known issue in the Theano scan op when dealing with inputs whose shape is 1 along a dimension.
- # In our case, this is a problem when training on grayscale images, and the below line fixes it.
- initial_states = [T.unbroadcast(init_state, 0, 1) for init_state in initial_states]
-
- if self.extrap_start_time is not None:
- initial_states += [K.variable(0, int if K.backend() != 'tensorflow' else 'int32')] # the last state will correspond to the current timestep
- return initial_states
-
- def build(self, input_shape):
- self.input_spec = [InputSpec(shape=input_shape)]
- self.conv_layers = {c: [] for c in ['i', 'f', 'c', 'o', 'a', 'ahat']}
-
- for l in range(self.nb_layers):
- for c in ['i', 'f', 'c', 'o']:
- act = self.LSTM_activation if c == 'c' else self.LSTM_inner_activation
- self.conv_layers[c].append(Conv2D(self.R_stack_sizes[l], self.R_filt_sizes[l], padding='same', activation=act, data_format=self.data_format))
-
- act = 'relu' if l == 0 else self.A_activation
- self.conv_layers['ahat'].append(Conv2D(self.stack_sizes[l], self.Ahat_filt_sizes[l], padding='same', activation=act, data_format=self.data_format))
-
- if l < self.nb_layers - 1:
- self.conv_layers['a'].append(Conv2D(self.stack_sizes[l+1], self.A_filt_sizes[l], padding='same', activation=self.A_activation, data_format=self.data_format))
-
- self.upsample = UpSampling2D(data_format=self.data_format)
- self.pool = MaxPooling2D(data_format=self.data_format)
-
- self.trainable_weights = []
- nb_row, nb_col = (input_shape[-2], input_shape[-1]) if self.data_format == 'channels_first' else (input_shape[-3], input_shape[-2])
- for c in sorted(self.conv_layers.keys()):
- for l in range(len(self.conv_layers[c])):
- ds_factor = 2 ** l
- if c == 'ahat':
- nb_channels = self.R_stack_sizes[l]
- elif c == 'a':
- nb_channels = 2 * self.stack_sizes[l]
- else:
- nb_channels = self.stack_sizes[l] * 2 + self.R_stack_sizes[l]
- if l < self.nb_layers - 1:
- nb_channels += self.R_stack_sizes[l+1]
- in_shape = (input_shape[0], nb_channels, nb_row // ds_factor, nb_col // ds_factor)
- if self.data_format == 'channels_last': in_shape = (in_shape[0], in_shape[2], in_shape[3], in_shape[1])
- with K.name_scope('layer_' + c + '_' + str(l)):
- self.conv_layers[c][l].build(in_shape)
- self.trainable_weights += self.conv_layers[c][l].trainable_weights
-
- self.states = [None] * self.nb_layers*3
-
- if self.extrap_start_time is not None:
- self.t_extrap = K.variable(self.extrap_start_time, int if K.backend() != 'tensorflow' else 'int32')
- self.states += [None] * 2 # [previous frame prediction, timestep]
-
- def step(self, a, states):
- r_tm1 = states[:self.nb_layers]
- c_tm1 = states[self.nb_layers:2*self.nb_layers]
- e_tm1 = states[2*self.nb_layers:3*self.nb_layers]
-
- if self.extrap_start_time is not None:
- t = states[-1]
- a = K.switch(t >= self.t_extrap, states[-2], a) # if past self.extrap_start_time, the previous prediction will be treated as the actual
-
- c = []
- r = []
- e = []
-
- # Update R units starting from the top
- for l in reversed(range(self.nb_layers)):
- inputs = [r_tm1[l], e_tm1[l]]
- if l < self.nb_layers - 1:
- inputs.append(r_up)
-
- inputs = K.concatenate(inputs, axis=self.channel_axis)
- i = self.conv_layers['i'][l].call(inputs)
- f = self.conv_layers['f'][l].call(inputs)
- o = self.conv_layers['o'][l].call(inputs)
- _c = f * c_tm1[l] + i * self.conv_layers['c'][l].call(inputs)
- _r = o * self.LSTM_activation(_c)
- c.insert(0, _c)
- r.insert(0, _r)
-
- if l > 0:
- r_up = self.upsample.call(_r)
-
- # Update feedforward path starting from the bottom
- for l in range(self.nb_layers):
- ahat = self.conv_layers['ahat'][l].call(r[l])
- if l == 0:
- ahat = K.minimum(ahat, self.pixel_max)
- frame_prediction = ahat
-
- # compute errors
- e_up = self.error_activation(ahat - a)
- e_down = self.error_activation(a - ahat)
-
- e.append(K.concatenate((e_up, e_down), axis=self.channel_axis))
-
- if self.output_layer_num == l:
- if self.output_layer_type == 'A':
- output = a
- elif self.output_layer_type == 'Ahat':
- output = ahat
- elif self.output_layer_type == 'R':
- output = r[l]
- elif self.output_layer_type == 'E':
- output = e[l]
-
- if l < self.nb_layers - 1:
- a = self.conv_layers['a'][l].call(e[l])
- a = self.pool.call(a) # target for next layer
-
- if self.output_layer_type is None:
- if self.output_mode == 'prediction':
- output = frame_prediction
- else:
- for l in range(self.nb_layers):
- layer_error = K.mean(K.batch_flatten(e[l]), axis=-1, keepdims=True)
- all_error = layer_error if l == 0 else K.concatenate((all_error, layer_error), axis=-1)
- if self.output_mode == 'error':
- output = all_error
- else:
- output = K.concatenate((K.batch_flatten(frame_prediction), all_error), axis=-1)
-
- states = r + c + e
- if self.extrap_start_time is not None:
- states += [frame_prediction, t + 1]
- return output, states
-
- def get_config(self):
- config = {'stack_sizes': self.stack_sizes,
- 'R_stack_sizes': self.R_stack_sizes,
- 'A_filt_sizes': self.A_filt_sizes,
- 'Ahat_filt_sizes': self.Ahat_filt_sizes,
- 'R_filt_sizes': self.R_filt_sizes,
- 'pixel_max': self.pixel_max,
- 'error_activation': self.error_activation.__name__,
- 'A_activation': self.A_activation.__name__,
- 'LSTM_activation': self.LSTM_activation.__name__,
- 'LSTM_inner_activation': self.LSTM_inner_activation.__name__,
- 'data_format': self.data_format,
- 'extrap_start_time': self.extrap_start_time,
- 'output_mode': self.output_mode}
- base_config = super(PredNet, self).get_config()
- return dict(list(base_config.items()) + list(config.items()))
diff --git a/workflow_parallel_frame_prediction/DataPreprocess/Stager_devel_N_24_process_netCDF.sh b/workflow_parallel_frame_prediction/DataPreprocess/Stager_devel_N_24_process_netCDF.sh
deleted file mode 100755
index 98333610..00000000
--- a/workflow_parallel_frame_prediction/DataPreprocess/Stager_devel_N_24_process_netCDF.sh
+++ /dev/null
@@ -1,22 +0,0 @@
-#!/bin/bash -x
-#SBATCH --account=deepacf
-#SBATCH --nodes=1
-#SBATCH --ntasks=12
-##SBATCH --ntasks-per-node=12
-#SBATCH --cpus-per-task=1
-#SBATCH --output=process_netcdf-out.%j
-#SBATCH --error=process_netcdf-err.%j
-#SBATCH --time=00:20:00
-#SBATCH --partition=devel
-#SBATCH --mail-type=ALL
-#SBATCH --mail-user=b.gong@fz-juelich.de
-
-module --force purge
-module use $OTHERSTAGES
-module load Stages/2019a
-module load Intel/2019.3.199-GCC-8.3.0 ParaStationMPI/5.2.2-1
-module load h5py/2.9.0-Python-3.6.8
-module load mpi4py/3.0.1-Python-3.6.8
-
-srun python mpi_stager_v2_process_netCDF.py --source_dir /p/scratch/deepacf/video_prediction_shared_folder/extractedData/\
---destination_dir /p/scratch/deepacf/bing/processData_size_64_64_3_3t_norm_test2
diff --git a/workflow_parallel_frame_prediction/DataPreprocess/external_function.py b/workflow_parallel_frame_prediction/DataPreprocess/external_function.py
deleted file mode 100755
index c360c0e0..00000000
--- a/workflow_parallel_frame_prediction/DataPreprocess/external_function.py
+++ /dev/null
@@ -1,158 +0,0 @@
-#from mpi4py import MPI
-from os import walk
-import os
-import sys
-import subprocess
-import logging
-import time
-import hashlib
-
-# ======================= List of functions ====================================== #
-
-# check the rank and print it
-
-def logger(file_name, logger_level, program_name):
- # Log file starter
-
- logging.basicConfig(filename=file_name, level=logger_level,
- format='%(asctime)s:%(levelname)s:%(message)s')
- logging.debug(' === PyStager is started === ')
- print(str(program_name) + ' is Running .... ')
-
-
-def config_file(config_file_name):
- params = {}
- for line in open(config_file_name):
- line = line.strip()
- read_in_value = line.split("=")
- if len(read_in_value) == 2:
- params[read_in_value[0].strip()] = read_in_value[1].strip()
-
- source_dir = str(params["Source_Directory"])
- print(source_dir)
- destination_dir = str(params["Destination_Directory"])
- log_dir = str(params["Log_Directory"])
- rsync_status = int(params["Rsync_Status"])
- return source_dir, destination_dir, log_dir, rsync_status
-
-
-def directory_scanner(source_path):
- # Take a look inside a directories and make a list of ll the folders, sub directories, number of the files and size
- # NOTE : It will neglect if there is a sub-directories inside directories!!!
-
- dir_detail_list = [] # directories details
- sub_dir_list = []
- total_size_source = 0
- total_num_files = 0
- list_directories = []
-
- list_directories = os.listdir(source_path)
- print(list_directories)
- print(int(len(list_directories)))
-
- for d in list_directories:
- print(d)
- path = source_path + d
- print(path)
- if os.path.isdir(path):
- sub_dir_list.append(d)
- sub_dir_list.sort()
- num_files = 0
- # size of the files and subdirectories
- size_dir = subprocess.check_output(['du', '-sc', path])
- splitted = size_dir.split() # fist item is the size of the folder
- size = (splitted[0])
- num_files = len([f for f in os.listdir(path) if os.path.isfile(os.path.join(path, f))])
- dir_detail_list.extend([d, size, num_files])
- total_num_files = total_num_files + int(num_files)
- total_size_source = total_size_source + int(size)
- else:
- print(path, 'does not exist')
- print("===== Debug here =====")
-
- total_num_directories = int(len(list_directories))
- total_size_source = float(total_size_source / 1000000)
-
- message = 'Total size of the source directory is:' + str(total_size_source) + 'Gb.'
- print(message)
- message = "Total number of the files in the source directory is: " + str(total_num_files)
- print(message)
- message = "Total number of the directories in the source directory is: " + str(total_num_directories)
- print(message)
-
- return dir_detail_list, sub_dir_list, total_size_source, total_num_files, total_num_directories
-
-
-def load_distributor(dir_detail_list, sub_dir_list, total_size_source, total_num_files, total_num_directories, p):
- # create a dictionary with p number of keys
- # for each directory they add the name to one of the keys
- transfer_dict = dict.fromkeys(list(range(1, p)))
- print(transfer_dict)
- # package_counter = 0 possibility to use the counter to fill
- counter = 1
- for Directory_counter in range(0, total_num_directories):
-
- if transfer_dict[counter] is None: # if the value for the key is None add to it
- transfer_dict[counter] = sub_dir_list[Directory_counter]
- else: # if key has a value join the new value to the old value
- transfer_dict[counter] = "{};{}".format(transfer_dict[counter], sub_dir_list[Directory_counter])
- counter = counter + 1
- if counter == p:
- counter = 1
-
- return transfer_dict
-
-def sync_file(source_path, destination_dir, job_name, rsync_status):
- rsync_msg = ("rsync -r " + source_path + job_name + "/" + " " + destination_dir + "/" + job_name)
- # print('Node:', str(my_rank),'will execute :', rsync_str,'\r\n')
- # sync the assigned folder
-
- if rsync_status == 1:
- os.system(rsync_msg)
-
-
-
-def hash_directory(source_path,job_name,hash_rep_file,input_status):
- #sha256_hash = hashlib.sha256()
- md5_hash = hashlib.md5()
-
- ########## Create a hashed file repasitory for direcotry(ies) assigned to node #######
- hash_repo_text = input_status + "_"+job_name +"_hashed.txt"
- os.chdir(hash_rep_file)
- hashed_text_note=open(hash_repo_text,"w+")
-
- # job_name is the name of the subdirectory that is going to be processed
- directory_to_process = source_path + job_name
- # print(directory_to_process)
- files_list = []
- for dirpath, dirnames, filenames in os.walk(directory_to_process):
- files_list.extend(filenames)
-
- os.chdir(directory_to_process) # change to the working directory
-
- for file_to_process in filenames:
-
- ## ======= this is the sha256 checksum ========= #
- #with open(file_to_process,"rb") as f:
- # # Read and update hash in chunks of 4K
- # for byte_block in iter(lambda: f.read(4096),b""):
- # sha256_hash.update(byte_block)
- # hashed_file = sha256_hash.hexdigest()
-
- with open(file_to_process,"rb") as f:
- # Read and update hash in chunks of 4K
- for byte_block in iter(lambda: f.read(4096),b""):
- md5_hash.update(byte_block)
- hashed_file = md5_hash.hexdigest()
-
- hashed_text_note.write(hashed_file)
-
- return
-
-def md5(fname):
- md5_hash = hashlib.md5()
- with open(fname,"rb") as f:
- # Read and update hash in chunks of 4K
- for byte_block in iter(lambda: f.read(4096),b""):
- md5_hash.update(byte_block)
- return md5_hash.hexdigest()
diff --git a/workflow_parallel_frame_prediction/DataPreprocess/mpi_split_data_multi_years.py b/workflow_parallel_frame_prediction/DataPreprocess/mpi_split_data_multi_years.py
deleted file mode 100644
index 68d1ddfb..00000000
--- a/workflow_parallel_frame_prediction/DataPreprocess/mpi_split_data_multi_years.py
+++ /dev/null
@@ -1,58 +0,0 @@
-from mpi4py import MPI
-import argparse
-from process_netCDF_v2 import *
-from metadata import MetaData
-import json
-
-#add parser arguments
-parser = argparse.ArgumentParser()
-#parser.add_argument("--source_dir", type=str, default="/p/scratch/deepacf/bing/extractedData/")
-parser.add_argument("--destination_dir","-dest",dest="destination_dir",type=str, default="/p/scratch/deepacf/bing/processData_size_64_64_3_3t_norm")
-parser.add_argument("--varnames","-vars",dest="varnames", nargs = '+')
-#parser.add_argument("--partition","-part",dest="partition",type=json.loads)
-# help="--partition allows to control the splitting of the processed data in training, test and validation data. Pass a dictionary-like string.")
-
-args = parser.parse_args()
-# ML 2020/06/08: Dirty workaround as long as data-splitting is done with this seperate Python-script
-# called from the same parent Shell-/Batch-script as 'mpi_stager_v2_process_netCDF.py'
-target_dir = os.path.join(MetaData.get_destdir_jsontmp(),"hickle")
-varnames = args.varnames
-
-#partition = args.partition
-#all_keys = partition.keys()
-#for key in all_keys:
-# print(partition[key])
-
-cv ={}
-partition1 = {
- "train":{
- #"2222":[1,2,3,5,6,7,8,9,10,11,12],
- #"2010_1":[1,2,3,4,5,6,7,8,9,10,11,12],
- #"2012":[1,2,3,4,5,6,7,8,9,10,11,12],
- #"2013_complete":[1,2,3,4,5,6,7,8,9,10,11,12],
- #"2015":[1,2,3,4,5,6,7,8,9,10,11,12],
- #"2017":[1,2,3,4,5,6,7,8,9,10,11,12]
- "2015":[1,2,3,4,5,6,7,8,9,10,11,12]
- },
- "val":
- {"2016":[1,2,3,4,5,6,7,8,9,10,11,12]
- },
- "test":
- {"2017":[1,2,3,4,5,6,7,8,9,10,11,12]
- }
- }
-
-
-
-
-
-#cv["1"] = partition1
-#cv2["2"] = partition2
-# ini. MPI
-comm = MPI.COMM_WORLD
-my_rank = comm.Get_rank() # rank of the node
-p = comm.Get_size() # number of assigned nods
-if my_rank == 0: # node is master
- split_data_multiple_years(target_dir=target_dir,partition=partition1,varnames=varnames)
-else:
- pass
diff --git a/workflow_parallel_frame_prediction/README.md b/workflow_parallel_frame_prediction/README.md
deleted file mode 100755
index 434e591f..00000000
--- a/workflow_parallel_frame_prediction/README.md
+++ /dev/null
@@ -1,56 +0,0 @@
-# Workflow for Frame Prediction by Parallel Deep Learning
-
-
-
-## Workflow for parallel deep learning
-
-This project implements a workflow for parallel deep learning to predict the 2m temperature based on Severin's master thesis [code link](https://github.com/severin1992/airtemprednet) [thesis link](https://b2drop.eudat.eu/s/RmTd8K3pLsDMFw6) .
-
-
-The workflow consists of a sqeuence of steps (Data Extraction, Data Preprocessing, Training and Data Postprocess)to implement video prediction, and In each step try to Parallel for accelerating the whole prediction process.
-
-
-The wokflow have been tested on the supercomputers from JSC, [JURECA](https://www.fz-juelich.de/ias/jsc/EN/Expertise/Supercomputers/JURECA/JURECA_node.html) and [JUWELS](https://www.fz-juelich.de/ias/jsc/EN/Expertise/Supercomputers/JUWELS/JUWELS_node.html)
-
-
-## Requirement:
-* Keras
-* Horovod
-* Python3.6
-* mpi4py
-
-
-
-## Usage
-
-1. Clone or download this repository,
-2. Install the required modules/packages on JURECA/JUWELS.
-
- ```shell
- source packageInstallation.sh
- ```
-
- Add package directory to 'PYTHONPATH'
- ```shell
- export PYTHONPATH=/p/home/jusers/USERNAME/jureca/.local/lib/python3.6/site-packages:$PYTHONPATH
- ```
-
-
- ```shell
- source packageInstallation.sh
- ```
- Add the packages directory to the PYTHONPATH
- ```shell
- export PATHONPATH=/p/home/jusers/USERNAME/jureca/.local/lib/python3.6/site-packages:$PYTHONPATH
- ```
-
-3. Configure your input directory, output and log directory in .dat file for each step.
-
-
-4. Run .sh file for submitting job
-
-
-## Workflow example
-
-
-
diff --git a/workflow_parallel_frame_prediction/Training/data_utils.py b/workflow_parallel_frame_prediction/Training/data_utils.py
deleted file mode 100755
index d048236f..00000000
--- a/workflow_parallel_frame_prediction/Training/data_utils.py
+++ /dev/null
@@ -1,163 +0,0 @@
-import hickle as hkl
-import numpy as np
-from keras import backend as K
-from keras.preprocessing.image import Iterator
-
-
-import inspect
-print(inspect.getmembers(hkl,predicate=inspect.ismethod))
-
-# Data generator that creates sequences for input into PredNet.
-class SequenceGenerator(Iterator):
- def __init__(self, data_file, source_file, nt,
- batch_size=8, shuffle=False, seed=None,
- output_mode='error', sequence_start_mode='all', N_seq=None,
- data_format=K.image_data_format()):
- self.X = hkl.load(data_file) # X will be like (n_images, nb_cols, nb_rows, nb_channels)
- self.sources = hkl.load(source_file) # source for each image so when creating sequences can assure that consecutive frames are from same video
- self.nt = nt
- self.batch_size = batch_size
- self.data_format = data_format
- assert sequence_start_mode in {'all', 'unique'}, 'sequence_start_mode must be in {all, unique}'
- self.sequence_start_mode = sequence_start_mode
- assert output_mode in {'error', 'prediction'}, 'output_mode must be in {error, prediction}'
- self.output_mode = output_mode
-
- if self.data_format == 'channels_first':
- self.X = np.transpose(self.X, (0, 3, 1, 2))
- self.im_shape = self.X[0].shape
- if self.sequence_start_mode == 'all': # allow for any possible sequence, starting from any frame
- #bing
- #self.possible_starts = np.array([i for i in range(self.X.shape[0] - self.nt) if self.sources[i] == self.sources[i + self.nt - 1]])
- self.possible_starts = np.array([i for i in range(self.X.shape[0] - self.nt)])
- elif self.sequence_start_mode == 'unique': #create sequences where each unique frame is in at most one sequence
- curr_location = 0
- possible_starts = []
- while curr_location < self.X.shape[0] - self.nt + 1:
- if self.sources[curr_location] == self.sources[curr_location + self.nt - 1]:
- possible_starts.append(curr_location)
- curr_location += self.nt
- else:
- curr_location += 1
- self.possible_starts = possible_starts
-
- if shuffle:
- self.possible_starts = np.random.permutation(self.possible_starts)
- if N_seq is not None and len(self.possible_starts) > N_seq: # select a subset of sequences if want to
- self.possible_starts = self.possible_starts[:N_seq]
- self.N_sequences = len(self.possible_starts)
- print("N_sequences",self.N_sequences)
- super(SequenceGenerator, self).__init__(len(self.possible_starts), batch_size, shuffle, seed)
-
- def __getitem__(self, null):
- return self.next()
-
- def next(self):
- with self.lock:
- current_index = (self.batch_index * self.batch_size) % self.n
- index_array, current_batch_size = next(self.index_generator), self.batch_size
- batch_x = np.zeros((current_batch_size, self.nt) + self.im_shape, np.float32)
- for i, idx in enumerate(index_array):
- idx = self.possible_starts[idx]
- batch_x[i] = self.preprocess(self.X[idx:idx+self.nt])
- if self.output_mode == 'error': # model outputs errors, so y should be zeros
- batch_y = np.zeros(current_batch_size, np.float32)
- elif self.output_mode == 'prediction': # output actual pixels
- batch_y = batch_x
- return batch_x, batch_y
-
- def preprocess(self, X):
- ### Normalization after extrema cut off: ###
- #cut maxs & mins to mean+3*std & mean-3*std of training set for each parameter
- #x_cut = np.zeros(shape=X.shape)
- #x_cut = X*1 #pass X by value and not by reference
- #x_cut[:,:,:,0][X[:,:,:,0]>311.5]=311.5 #set T2 upper limit
- #x_cut[:,:,:,0][X[:,:,:,0]<258.9]=258.9 #set T2 lower limit
- #x_cut[:,:,:,1][X[:,:,:,1]>104635.2]=104635.2 #set GP upper limit
- #x_cut[:,:,:,1][X[:,:,:,1]<98205.6]=98205.6 #set GP lower limit ###Caution: Drastical cut ###
- #x_cut[:,:,:,2][X[:,:,:,2]>6209.5]=6209.5 #set GPH upper limit ###Caution: Unnecessary as it succeeds max GPH ###
- #x_cut[:,:,:,2][X[:,:,:,2]<5005.8]=5005.8 #set GPH lower limit
- #normalize X based on max and min values(equals upper and lower limits except highCutGPH)
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (x_cut[:,:,:,0]-258.9)/(311.5-258.9)
- #x_processed[:,:,:,1] = (x_cut[:,:,:,1]-98205.6)/(104635.2-98205.6)
- #x_processed[:,:,:,2] = (x_cut[:,:,:,2]-5005.8)/(6007.097417091836-5005.8) #GPH max stays; see above
-
- ### 'Standard' normalization: (x-min(x))/(max(x)-min(x)) ###
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-93401.125)/(105391.4375-93401.125)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-4836.070232780612)/(6007.097417091836-4836.070232780612)
-
- ### t2only 'Standard' normalization: (x-min(x))/(max(x)-min(x)) ###
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-235.2141571044922)/(321.46630859375-235.2141571044922)
-
- ### t2_2MSL_1 'standard' normalization:
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-93401.125)/(105391.4375-93401.125)
-
- ### t2_1MSL_2 'standard' normalization:
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-93401.125)/(105391.4375-93401.125)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-93401.125)/(105391.4375-93401.125)
-
- ### t2_2gph500_1 'standard' normalization:
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-4836.070232780612)/(6007.097417091836-4836.070232780612)
- ## t2_1gph500_2 'standard' normalization:
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = (X[:,:,:,1]-4836.070232780612)/(6007.097417091836-4836.070232780612)
- #x_processed[:,:,:,2] = (X[:,:,:,2]-4836.070232780612)/(6007.097417091836-4836.070232780612)
-
- ### No standardization for moving Objects test set: Just 0s and 1s
- #x_processed = np.zeros(shape=X.shape)
- #x_processed = X
-
- ### t2_1 'standard' normalization (got one dimension less, due to just one channel)
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- #x_processed[:,:,:,1] = X[:,:,:,1]
- #x_processed[:,:,:,2] = X[:,:,:,2]
-
- ### t2_1 'standard' normalization (got one dimension less, due to just one channel)
- x_processed = np.zeros(shape=X.shape)
- x_processed[:,:,:,0] = (X[:,:,:,0]-235.2141571044922)/(321.46630859375-235.2141571044922)
- x_processed[:,:,:,1] = (X[:,:,:,1]-235.2141571044922)/(321.46630859375-235.2141571044922)
- x_processed[:,:,:,2] = X[:,:,:,2]
-
- ### Standardization: (x-mean)/standard_deviation ###
- #Doesn't work due to some activation functions
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (X[:,:,:,0]-285.1751264870658)/8.770013367617763
- #x_processed[:,:,:,1] = (X[:,:,:,1]-101420.4382666807)/1071.5999818175521
- #x_processed[:,:,:,2] = (X[:,:,:,2]-5607.662795353248)/200.62593105865764
-
- ### Standardization+Normalization ###
- # standardize:(x-mean)/standard_deviation
- #x_preprocessed = np.zeros(shape=X.shape)
- #x_preprocessed[:,:,:,0] = (X[:,:,:,0]-285.1751264870658)/8.770013367617763
- #x_preprocessed[:,:,:,1] = (X[:,:,:,1]-101420.4382666807)/1071.5999818175521
- #x_preprocessed[:,:,:,2] = (X[:,:,:,2]-5607.662795353248)/200.62593105865764
- # normalize:(x-min(x))/(max(x)-min(x))
- #x_processed = np.zeros(shape=X.shape)
- #x_processed[:,:,:,0] = (x_preprocessed[:,:,:,0]-np.amin(x_preprocessed[:,:,:,0]))/(np.amax(x_preprocessed[:,:,:,0])-np.amin(x_preprocessed[:,:,:,0]))
- #x_processed[:,:,:,1] = (x_preprocessed[:,:,:,1]-np.amin(x_preprocessed[:,:,:,1]))/(np.amax(x_preprocessed[:,:,:,1])-np.amin(x_preprocessed[:,:,:,1]))
- #x_processed[:,:,:,2] = (x_preprocessed[:,:,:,2]-np.amin(x_preprocessed[:,:,:,2]))/(np.amax(x_preprocessed[:,:,:,2])-np.amin(x_preprocessed[:,:,:,2]))
-
- return x_processed.astype(np.float32)
- #return X.astype(np.float32) / 255
-
- def create_all(self):
- X_all = np.zeros((self.N_sequences, self.nt) + self.im_shape, np.float32)
- for i, idx in enumerate(self.possible_starts):
- X_all[i] = self.preprocess(self.X[idx:idx+self.nt])
- return X_all
diff --git a/workflow_parallel_frame_prediction/Training/devel_horovodJob.sh b/workflow_parallel_frame_prediction/Training/devel_horovodJob.sh
deleted file mode 100644
index 6b442c18..00000000
--- a/workflow_parallel_frame_prediction/Training/devel_horovodJob.sh
+++ /dev/null
@@ -1,43 +0,0 @@
-#!/bin/bash
-#SBATCH --account=deepacf
-# budget account where contingent is taken from# TASKS = NODES * GPUS_PER_NODE
-#SBATCH --nodes=1
-#SBATCH --ntasks-per-node=2
-#SBATCH --ntasks=2
-# can be omitted if --nodes and --ntasks-per-node
-# are given
-# SBATCH --cpus-per-task=1
-# for OpenMP/hybrid jobs only
-#SBATCH --output=horovod-%j.out
-# if keyword omitted: Default is slurm-%j.out in
-# the submission directory (%j is replaced by
-# the job ID).
-#SBATCH --error=horovod-%j.err
-# if keyword omitted: Default is slurm-%j.out in
-# the submission directory.
-#SBATCH --time=00:20:00
-#SBATCH --gres=gpu:2
-#SBATCH --partition=develgpus
-#SBATCH --mail-user=b.gong@fz-juelich.de
-#SBATCH --mail-type=ALL
-
-#create a folder to save the output
-jutil env activate -p deepacf
-module --force purge
-module load Stages/Devel-2019a
-module load GCC/8.3.0
-module load MVAPICH2/2.3.2-GDR
-module load Stages/2019a
-module load Horovod/0.16.2-GPU-Python-3.6.8
-module load Keras/2.2.4-GPU-Python-3.6.8
-#module load ParaStationMPI/5.2.2-1
-#module load h5py/2.9.0-Python-3.6.8
-# *** start of job script ***:
-# Note: The current working directory at this point is
-# the directory where sbatch was executed.
-# export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK}
-# *** start of job script ***
-# Note: The current working directory at this point is
-# the directory where sbatch was executed.
-# export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK}
-srun --cpu_bind=none python3.6 kitti_train_horovod.py
diff --git a/workflow_parallel_frame_prediction/Training/evaluate_multistep.py b/workflow_parallel_frame_prediction/Training/evaluate_multistep.py
deleted file mode 100644
index a555a929..00000000
--- a/workflow_parallel_frame_prediction/Training/evaluate_multistep.py
+++ /dev/null
@@ -1,121 +0,0 @@
-'''
-Evaluate trained PredNet
-Calculates mean-squared error and plots predictions.
-'''
-
-import os
-#import sys, argparse
-import numpy as np
-#from six.moves import cPickle
-import matplotlib
-matplotlib.use('Agg')
-import matplotlib.pyplot as plt
-import matplotlib.gridspec as gridspec
-
-from keras import backend as K
-from keras.models import Model, model_from_json
-from keras.layers import Input, Dense, Flatten
-
-from prednet import PredNet
-from data_utils import SequenceGenerator
-from kitti_settings import *
-#from scipy.misc import imsave
-
-##Just for checking how the shape is after generator.create_all() from Sequence Generator
-#import hickle as hkl
-##
-n_plot = 10 #number of plots
-batch_size = 10
-nt = 15 #number of timesteps used for sequences in training
-numtests = 18
-extrap = 10 #frame number from where extrapolation will start to be produced
-
-#parser = argparse.ArgumentParser()
-#parser.add_argument('-ft', help="fine-tune multistep: add extrap time")
-#args=parser.parse_args()
-
-weights_file = os.path.join(WEIGHTS_DIR, 'tensorflow_weights/prednet_kitti_weights-extrapfinetuned.hdf5')
-json_file = os.path.join(WEIGHTS_DIR, 'prednet_kitti_model-extrapfinetuned.json')
-test_file = os.path.join(DATA_DIR, 'X_test.hkl')
-test_sources = os.path.join(DATA_DIR, 'sources_test.hkl')
-
-#if args.ft is not None:
-# extrap = int(args.ft)
-# nt = extrap + 5
-# weights_file = os.path.join(MODELS_DIR, 'prednet_ee_weights-extrapfinetuned.hdf5')
-# json_file = os.path.join(MODELS_DIR, 'prednet_ee_model-extrapfinetuned.json')
-
-# Load trained model
-f = open(json_file, 'r')
-json_string = f.read()
-f.close()
-train_model = model_from_json(json_string, custom_objects = {'PredNet': PredNet})
-train_model.load_weights(weights_file)
-
-# Create testing model (to output predictions)
-layer_config = train_model.layers[1].get_config()
-layer_config['output_mode'] = 'prediction' #'prediction'
-layer_config['extrap_start_time'] = extrap;
-data_format = layer_config['data_format'] if 'data_format' in layer_config else layer_config['dim_ordering']
-test_prednet = PredNet(weights=train_model.layers[1].get_weights(), **layer_config)
-input_shape = list(train_model.layers[0].batch_input_shape[1:])
-input_shape[0] = nt
-inputs = Input(shape=tuple(input_shape))
-predictions = test_prednet(inputs)
-test_model = Model(inputs=inputs, outputs=predictions)
-
-test_generator = SequenceGenerator(test_file, test_sources, nt, sequence_start_mode='unique', data_format=data_format) # orig: unique
-X_test = test_generator.create_all()
-X_hat = test_model.predict(X_test, batch_size)
-if data_format == 'channels_first':
- X_test = np.transpose(X_test, (0, 1, 3, 4, 2))
- X_hat = np.transpose(X_hat, (0, 1, 3, 4, 2))
-
-# Compare MSE of PredNet predictions vs. using last frame. Write results to prediction_scores.txt
-shapeXhat = str(X_hat.shape) #Just have a look at the shapes to be sure we are calculating the right MSE
-shapeXtest = str(X_test.shape)
-mse_model = np.mean( (X_test[:, 1:,:,:,0] - X_hat[:, 1:,:,:,0])**2 ) # look at all timesteps except the first
-mse_model_last = np.mean( (X_test[:, 9,:,:,0] - X_hat[:, 14,:,:,0])**2 )
-#mse_prev = np.mean( (X_test[:, :-1,:,:,0] - X_test[:, 1:,:,:,0])**2 )
-mse_prev = np.mean( (X_test[:, 9,:,:,0] - X_test[:, 14,:,:,0])**2 )
-if not os.path.exists(RESULTS_SAVE_DIR): os.mkdir(RESULTS_SAVE_DIR)
-f = open(os.path.join(RESULTS_SAVE_DIR, 'prediction_scores.txt'), 'w')
-f.write("Model MSE: %f\n" % mse_model)
-f.write("Model MSE from only last prediction in sequence in comparison with extrap start time: %f\n" % mse_model_last)
-f.write("Previous Frame MSE last frame vs extrap start time: %f" % mse_prev)
-f.write("Shape of X_test: " + shapeXtest)
-f.write("")
-f.write("Shape of X_hat: " + shapeXhat)
-f.close()
-
-# Plot some predictions
-aspect_ratio = float(X_hat.shape[2]) / X_hat.shape[3]
-plt.figure(figsize = (nt, 2*aspect_ratio))
-gs = gridspec.GridSpec(2, nt)
-gs.update(wspace=0., hspace=0.)
-plot_save_dir = os.path.join(RESULTS_SAVE_DIR, 'prediction_plots/')
-if not os.path.exists(plot_save_dir): os.mkdir(plot_save_dir)
-plot_idx = np.random.permutation(X_test.shape[0])[:n_plot]
-for i in plot_idx:
- for t in range(nt):
- plt.subplot(gs[t])
- plt.imshow(X_test[i,t,:,:,0], interpolation='none')
- plt.tick_params(axis='both', which='both', bottom='off', top='off', left='off', right='off', labelbottom='off', labelleft='off')
- if t==0: plt.ylabel('Actual', fontsize=10)
-
- plt.subplot(gs[t + nt])
- plt.imshow(X_hat[i,t,:,:,0], interpolation='none')
- plt.tick_params(axis='both', which='both', bottom='off', top='off', left='off', right='off', labelbottom='off', labelleft='off')
- if t==0: plt.ylabel('Predicted', fontsize=10)
-
- plt.savefig(plot_save_dir + 'plot_' + str(i) + '.jpg')
- plt.clf()
-
-#abe
-#for test in range(numtests):
-# testdir = "tile-" + str(test)
-# testdir = os.path.join(plot_save_dir, testdir)
-# if not os.path.exists( testdir ) : os.mkdir( testdir )
-# for t in range(nt):
-# imsave( testdir + "/pred-%02d.jpg" % (t,), X_hat[test,t] )
-# imsave( testdir + "/orig-%02d.jpg" % (t,), X_test[test,t])
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/Training/hickle/bin/f2py b/workflow_parallel_frame_prediction/Training/hickle/bin/f2py
deleted file mode 100755
index fcc774fb..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/bin/f2py
+++ /dev/null
@@ -1,4 +0,0 @@
-#!/usr/local/software/jureca/Stages/2018b/software/Python/3.6.6-GCCcore-7.3.0/bin/python
-# EASY-INSTALL-SCRIPT: 'numpy==1.15.2','f2py'
-__requires__ = 'numpy==1.15.2'
-__import__('pkg_resources').run_script('numpy==1.15.2', 'f2py')
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/__pycache__/site.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/__pycache__/site.cpython-36.pyc
deleted file mode 100644
index abc2e28feeb37e7e377c2cf5eadc076e5edb1e93..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 1482
zcmXr!<>i{1ZxPGP$-wZK0SPcMFfceUFfbI$FfcHrFr+ZHFhnt=GNv$PGZjgsGBh(V
zGDI_^Fb6YevLwT_i83%Sh%hiPID_;FFflNsFs3l%Fw{b5##+W2#w?~R<`kx8##*Kt
zrWA&3j$)G<hAfs8=4_@Ss~RRAh9awUhFYdPo*Jes)*8kvHjv5^_8Nv1mQ03b=33?&
zhAfT}&Kjm>#uSEDCP{``mKw$^jx4Sc?q<eX){0#<EG`VOt7_S5*lJj_nTpG5*mxL<
zx(dr`n6sE`K&G+Ru%|F&a~AEXVb9`8VQprt<*4BRi|nf5$l?WwFfs%%6qZ#gfL+g5
z%UQz-Qdz^6&e+Tp!(7W%%U#P8&X6a@!cfAW%~kY?kpTqD%OV-_m?9WzxIq3Y`d3(1
zB~rr)GO>oMhP#HRnT3%doFRpwhAT?|6nYV0Tf!Mi_*2-L8JieO1XI{kI9eD$K6GJd
zW^87x<pukkGlj94xt1@5v4$aDs6woU$%P@7ua>Pu7~~&Dh6#*?5#bCq3@i*;LN$Em
z3``91LM0+KEGdkV49$!&Ott*A0x1l&g6RykLKQg>(_@or*-At}`eT@Cg=<A>L}~=G
znI<q69jOrnxgNwS;j9sAW~>#g5vk$LW}CoRd;>1`q(&%96r`JxAy1@6C`+tHu!gOf
ziIJg3FonyUp;mALW8sW&h8iJ|Z7B?(n6hC=VF+f><o0_BN-Vdy0wO~^{QUwPLp-8b
z%QN$gjcy4SB$i~vXXX~<7nP(I#V02wXQbX@Db6fOtx|E$FU?8GQ!i0S%gjqrD9K1w
zC`wJtQBViTsw?E?r<CTTCNqLeW?*0drCb&U28LizDg>q8bcPy+Sdm)B5{4Q^P`Z(1
zNMT50n#feh63n2<_!6W~lkpZuZen&S$PJoIx0v({Zn0#Rq~@+<DB@&bVEE;$U!Y%{
zT$EUnoS~nbl~tT<Vx*sxnU}7go1<HtT9#UrnWvwTnVg-Is-KgYq+d{3l98WhtY@YV
za-nWPVsdt3dTKG4r&mz<i_0b_v$!NVKexb66XZ&eli3)mq~ha~^7BjdY;y9G6LX5~
z^bpE!amUA}rl)7-6{i-J6tOZeFldV3VlJ*MzQvSZe2YCbuPn1DKkpWEdTPlnj)I)T
zlC=Dy+*_>01v!}|QS2aZ6{i;5Voy&kNh~QTy2Sz#xy785ns<vmC9^0mF*o%VM|^x*
zW=?8+{4H)!FvmlJ^%i$deqsuY!<?C0aEsd$97(Q4MfpXySd(+|i&Jm0q+}+S6!9=H
zFhp@Brlf$poswBp#KyqDaEmRmpdd9b<rW*n^_rqZybKHsMSLIv<nt)zg3RP7_TrKh
zkgK9tb1OkwIY4G{z|zAlHjpz?i;AScCKi_zr6%SUfeM=_)|AxZ<f16<?7aN)ym*ii
z#ZheeIVlDC#kUw!qL}hhqBtQo$0wFVapa|zgB9IkEQn&tD*(qYgx~@j4T%F#q9{%P
zr6@)YMh-?EMhQkXCLSgpMj=KXCN4%kMkYosMh*~W;$mcBVq)ZAlwjlot7T$jViIBE
zVq|0DU=m^CVPpfTXX0ZLU}RxrVd7!rVG;q!F!C_-F!C`Ksepn3lwSQbAskRdfFdM5
e9^x!e0)()@4&<<bM2sESiJ$=HaOYrTV*~(<@OyOt
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/easy-install.pth b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/easy-install.pth
deleted file mode 100755
index 09ac2825..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/easy-install.pth
+++ /dev/null
@@ -1,4 +0,0 @@
-./hickle-3.4.3-py3.6.egg
-/usr/local/software/jureca/Stages/2018b/software/h5py/2.8.0-ipsmpi-2018b-Python-3.6.6/lib/python3.6/site-packages/h5py-2.8.0-py3.6-linux-x86_64.egg
-/usr/local/software/jureca/Stages/2018b/software/SciPy-Stack/2018b-gcccoremkl-7.3.0-2019.0.117-Python-3.6.6/lib/python3.6/site-packages/numpy-1.15.2-py3.6-linux-x86_64.egg
-/usr/local/software/jureca/Stages/2018b/software/Python/3.6.6-GCCcore-7.3.0/lib/python3.6/site-packages/six-1.11.0-py3.6.egg
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/PKG-INFO b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/PKG-INFO
deleted file mode 100755
index 5f821450..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/PKG-INFO
+++ /dev/null
@@ -1,207 +0,0 @@
-Metadata-Version: 2.1
-Name: hickle
-Version: 3.4.3
-Summary: Hickle - a HDF5 based version of pickle
-Home-page: http://github.com/telegraphic/hickle
-Author: Danny Price
-Author-email: dan@thetelegraphic.com
-License: UNKNOWN
-Download-URL: https://github.com/telegraphic/hickle/archive/3.4.3.tar.gz
-Description: [](https://travis-ci.org/telegraphic/hickle)
- [](http://joss.theoj.org/papers/0c6638f84a1a574913ed7c6dd1051847)
-
-
- Hickle
- ======
-
- Hickle is a [HDF5](https://www.hdfgroup.org/solutions/hdf5/) based clone of `pickle`, with a twist: instead of serializing to a pickle file,
- Hickle dumps to a HDF5 file (Hierarchical Data Format). It is designed to be a "drop-in" replacement for pickle (for common data objects), but is
- really an amalgam of `h5py` and `dill`/`pickle` with extended functionality.
-
- That is: `hickle` is a neat little way of dumping python variables to HDF5 files that can be read in most programming
- languages, not just Python. Hickle is fast, and allows for transparent compression of your data (LZF / GZIP).
-
- Why use Hickle?
- ---------------
-
- While `hickle` is designed to be a drop-in replacement for `pickle` (or something like `json`), it works very differently.
- Instead of serializing / json-izing, it instead stores the data using the excellent [h5py](https://www.h5py.org/) module.
-
- The main reasons to use hickle are:
-
- 1. It's faster than pickle and cPickle.
- 2. It stores data in HDF5.
- 3. You can easily compress your data.
-
- The main reasons not to use hickle are:
-
- 1. You don't want to store your data in HDF5. While hickle can serialize arbitrary python objects, this functionality is provided only for convenience, and you're probably better off just using the pickle module.
- 2. You want to convert your data in human-readable JSON/YAML, in which case, you should do that instead.
-
- So, if you want your data in HDF5, or if your pickling is taking too long, give hickle a try.
- Hickle is particularly good at storing large numpy arrays, thanks to `h5py` running under the hood.
-
- Documentation
- -------------
-
- Documentation for hickle can be found at [telegraphic.github.io/hickle/](http://telegraphic.github.io/hickle/).
-
-
- Usage example
- -------------
-
- Hickle is nice and easy to use, and should look very familiar to those of you who have pickled before.
-
- In short, `hickle` provides two methods: a [hickle.load](http://telegraphic.github.io/hickle/toc.html#hickle.load)
- method, for loading hickle files, and a [hickle.dump](http://telegraphic.github.io/hickle/toc.html#hickle.dump)
- method, for dumping data into HDF5. Here's a complete example:
-
- ```python
- import os
- import hickle as hkl
- import numpy as np
-
- # Create a numpy array of data
- array_obj = np.ones(32768, dtype='float32')
-
- # Dump to file
- hkl.dump(array_obj, 'test.hkl', mode='w')
-
- # Dump data, with compression
- hkl.dump(array_obj, 'test_gzip.hkl', mode='w', compression='gzip')
-
- # Compare filesizes
- print('uncompressed: %i bytes' % os.path.getsize('test.hkl'))
- print('compressed: %i bytes' % os.path.getsize('test_gzip.hkl'))
-
- # Load data
- array_hkl = hkl.load('test_gzip.hkl')
-
- # Check the two are the same file
- assert array_hkl.dtype == array_obj.dtype
- assert np.all((array_hkl, array_obj))
- ```
-
- ### HDF5 compression options
-
- A major benefit of `hickle` over `pickle` is that it allows fancy HDF5 features to
- be applied, by passing on keyword arguments on to `h5py`. So, you can do things like:
- ```python
- hkl.dump(array_obj, 'test_lzf.hkl', mode='w', compression='lzf', scaleoffset=0,
- chunks=(100, 100), shuffle=True, fletcher32=True)
- ```
- A detailed explanation of these keywords is given at http://docs.h5py.org/en/latest/high/dataset.html,
- but we give a quick rundown below.
-
- In HDF5, datasets are stored as B-trees, a tree data structure that has speed benefits over contiguous
- blocks of data. In the B-tree, data are split into [chunks](http://docs.h5py.org/en/latest/high/dataset.html#chunked-storage),
- which is leveraged to allow [dataset resizing](http://docs.h5py.org/en/latest/high/dataset.html#resizable-datasets) and
- compression via [filter pipelines](http://docs.h5py.org/en/latest/high/dataset.html#filter-pipeline). Filters such as
- `shuffle` and `scaleoffset` move your data around to improve compression ratios, and `fletcher32` computes a checksum.
- These file-level options are abstracted away from the data model.
-
- Recent changes
- --------------
-
- * December 2018: Accepted to Journal of Open-Source Software (JOSS).
- * June 2018: Major refactor and support for Python 3.
- * Aug 2016: Added support for scipy sparse matrices `bsr_matrix`, `csr_matrix` and `csc_matrix`.
-
- Performance comparison
- ----------------------
-
- Hickle runs a lot faster than pickle with its default settings, and a little faster than pickle with `protocol=2` set:
-
- ```Python
- In [1]: import numpy as np
-
- In [2]: x = np.random.random((2000, 2000))
-
- In [3]: import pickle
-
- In [4]: f = open('foo.pkl', 'w')
-
- In [5]: %time pickle.dump(x, f) # slow by default
- CPU times: user 2 s, sys: 274 ms, total: 2.27 s
- Wall time: 2.74 s
-
- In [6]: f = open('foo.pkl', 'w')
-
- In [7]: %time pickle.dump(x, f, protocol=2) # actually very fast
- CPU times: user 18.8 ms, sys: 36 ms, total: 54.8 ms
- Wall time: 55.6 ms
-
- In [8]: import hickle
-
- In [9]: f = open('foo.hkl', 'w')
-
- In [10]: %time hickle.dump(x, f) # a bit faster
- dumping <type 'numpy.ndarray'> to file <HDF5 file "foo.hkl" (mode r+)>
- CPU times: user 764 us, sys: 35.6 ms, total: 36.4 ms
- Wall time: 36.2 ms
- ```
-
- So if you do continue to use pickle, add the `protocol=2` keyword (thanks @mrocklin for pointing this out).
-
- For storing python dictionaries of lists, hickle beats the python json encoder, but is slower than uJson. For a dictionary with 64 entries, each containing a 4096 length list of random numbers, the times are:
-
-
- json took 2633.263 ms
- uJson took 138.482 ms
- hickle took 232.181 ms
-
-
- It should be noted that these comparisons are of course not fair: storing in HDF5 will not help you convert something into JSON, nor will it help you serialize a string. But for quick storage of the contents of a python variable, it's a pretty good option.
-
- Installation guidelines (for Linux and Mac OS).
- -----------------------------------------------
-
- ### Easy method
- Install with `pip` by running `pip install hickle` from the command line.
-
- ### Manual install
-
- 1. You should have Python 2.7 and above installed
-
- 2. Install h5py
- (Official page: http://docs.h5py.org/en/latest/build.html)
-
- 3. Install hdf5
- (Official page: http://www.hdfgroup.org/ftp/HDF5/current/src/unpacked/release_docs/INSTALL)
-
- 4. Download `hickle`:
- via terminal: git clone https://github.com/telegraphic/hickle.git
- via manual download: Go to https://github.com/telegraphic/hickle and on right hand side you will find `Download ZIP` file
-
- 5. cd to your downloaded `hickle` directory
-
- 6. Then run the following command in the `hickle` directory:
- `python setup.py install`
-
- ### Testing
-
- Once installed from source, run `python setup.py test` to check it's all working.
-
-
- Bugs & contributing
- --------------------
-
- Contributions and bugfixes are very welcome. Please check out our [contribution guidelines](https://github.com/telegraphic/hickle/blob/master/CONTRIBUTING.md)
- for more details on how to contribute to development.
-
-
- Referencing hickle
- ------------------
-
- If you use `hickle` in academic research, we would be grateful if you could reference [our paper](http://joss.theoj.org/papers/0c6638f84a1a574913ed7c6dd1051847) in the [Journal of Open-Source Software (JOSS)](http://joss.theoj.org/about).
-
- ```
- Price et al., (2018). Hickle: A HDF5-based python pickle replacement. Journal of Open Source Software, 3(32), 1115, https://doi.org/10.21105/joss.01115
- ```
-
-Keywords: pickle,hdf5,data storage,data export
-Platform: Cross platform (Linux
-Platform: Mac OSX
-Platform: Windows)
-Requires-Python: >=2.7
-Description-Content-Type: text/markdown
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/SOURCES.txt b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/SOURCES.txt
deleted file mode 100755
index bf56f059..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/SOURCES.txt
+++ /dev/null
@@ -1,52 +0,0 @@
-.gitignore
-.nojekyll
-.pylintrc
-.travis.yml
-CODE_OF_CONDUCT.md
-CONTRIBUTING.md
-LICENSE
-README.md
-_config.yml
-paper.bib
-paper.md
-requirements.txt
-setup.cfg
-setup.py
-docs/Makefile
-docs/make_docs.sh
-docs/source/conf.py
-docs/source/index.md
-docs/source/toc.rst
-docs/source/_static/empty.txt
-docs/source/_templates/empty.txt
-hickle/__init__.py
-hickle/helpers.py
-hickle/hickle.py
-hickle/hickle_legacy.py
-hickle/hickle_legacy2.py
-hickle/lookup.py
-hickle.egg-info/PKG-INFO
-hickle.egg-info/SOURCES.txt
-hickle.egg-info/dependency_links.txt
-hickle.egg-info/not-zip-safe
-hickle.egg-info/requires.txt
-hickle.egg-info/top_level.txt
-hickle/loaders/__init__.py
-hickle/loaders/load_astropy.py
-hickle/loaders/load_numpy.py
-hickle/loaders/load_pandas.py
-hickle/loaders/load_python.py
-hickle/loaders/load_python3.py
-hickle/loaders/load_scipy.py
-tests/__init__.py
-tests/test_astropy.py
-tests/test_hickle.py
-tests/test_hickle_helpers.py
-tests/test_legacy_load.py
-tests/test_scipy.py
-tests/legacy_hkls/generate_test_hickle.py
-tests/legacy_hkls/hickle_1_1_0.hkl
-tests/legacy_hkls/hickle_1_3_2.hkl
-tests/legacy_hkls/hickle_1_4_0.hkl
-tests/legacy_hkls/hickle_2_0_5.hkl
-tests/legacy_hkls/hickle_2_1_0.hkl
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/dependency_links.txt b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/dependency_links.txt
deleted file mode 100755
index 8b137891..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/dependency_links.txt
+++ /dev/null
@@ -1 +0,0 @@
-
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/not-zip-safe b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/not-zip-safe
deleted file mode 100755
index 8b137891..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/not-zip-safe
+++ /dev/null
@@ -1 +0,0 @@
-
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/requires.txt b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/requires.txt
deleted file mode 100755
index 8ccd5558..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/requires.txt
+++ /dev/null
@@ -1,2 +0,0 @@
-numpy
-h5py
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/top_level.txt b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/top_level.txt
deleted file mode 100755
index ce3b9fb8..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/EGG-INFO/top_level.txt
+++ /dev/null
@@ -1,2 +0,0 @@
-hickle
-tests
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__init__.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__init__.py
deleted file mode 100755
index 46e2ea2c..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__init__.py
+++ /dev/null
@@ -1,4 +0,0 @@
-from .hickle import dump, load
-from .hickle import __version__
-
-
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/__init__.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/__init__.cpython-36.pyc
deleted file mode 100644
index 638a67eaaa3ab784f6e31d96cc63e6c3a1acc1e7..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 278
zcmXr!<>i`^S{>`iz`*dB0SPcMFfceUFfbG=FfcHrFr+Z%Fyt~uG3GKwF)@PpOgRj>
z%u&pY3@OaP44N!285tNDG?{L(q?G0s++xYePfXEdyu}?KUzS=_oSB~&AMdBha*Hh^
zGdVjawFqQ*5i<h=LlFxD1H(#&A~p~kMEuIvFVHV2%FjwoF40fU$|_DaG15=U%uCnL
z&CxARElVxR%+rUMqMwtQq+d{3l98WhtY@ZQoLQ2pTacKXotU0l3{|3QtY@NUtXoiN
qtY@Z|nw}07){l?R%*!l^kJl@xyv1RYo1apelWGU@K{3c>0*nCn_DL-O
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/helpers.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/helpers.cpython-36.pyc
deleted file mode 100644
index 944de5cd7f681a49a8d9fbf2024be8e218cadb71..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 3266
zcmXr!<>i`^S{=Kan}Ojm0}^0jU|?`yU|=YAVPIfLVMt-jVTgjzj8Tj!OexGQ3{gxe
zEGeul3{lJ}Y$@z53{fm894VYF3{k8pTq)cw3{h+;JSn^_3{mVUjKK_=d@n)9`z14i
zC<X=wW(EcZkS)$2TjUrR7-|@@7*ZG|8B&-e8EP487}6Ld8ETno7;Bi)n1UHJnX6nC
zJW_KCQi~MQO7oISGV}8kO7azoQcFsU@)SxkQWXj+<4Y<FQWX;OQWR1WOA-~5a`Kb2
z71HvH6cQCYT-+4$@>5b3Zn1zA6@i?g$#{!3v81G^Sd;k{TSh!c`W8FXlv^wyBUUmL
zu`@6*{L0fW&@U*;&q_@$(NE6GDo!>r(of3FOV`iM(Jf9bOD)RG)6d9E&dy2I&&f>E
zFQ_cZ$j>v@Gt)25EJ@WZNKDR7OiwL_D$zC8Gto2FEvPisGt)~=PlpQYXMlZItXEKZ
zOE5jP1nimk#JrSvkV{xV!Ndj%J0?Cx4%Q-01_lOrSc1Y4gu!7cgc+8g@Mo?vBQpHZ
z0;~vRGBmu3K(VIDR0MKO5i0`&BzVAKRK&r+z;KHL=5RicdqK(=nD`irxR6~7(g%_O
zyS9dbfuV#Ui?PTmouQVoLaT<sg&|g{mZ^lPhN*<PnW;#lgr$Zlg|V3lloqmBYnYlD
z85s%%!Wn89YZ)LZ5u~S?v6i`nt%jkQv8bSiIg34orJ0e3p{O8*HHEE(p_ZkDqlTe|
zC5640DVPDoZe|Q-NMQ+P(B!B(rVx^tovM(ikds+lqL81aP+U@!nU`J+PDsW1MI{QE
zB??KE3YmE&sp+Xj3T25orKt+#nI##Sc~G@nTnY*b3XVnT#a3Vzh?kvO84os514L?C
zDfmE*&rYo@hG+-^CtfUi;z8D?rYK<3s-pmvL^ug-=PlO6;^fRsltcjv>R?bzNH8!k
zq%+ho#0u0hf`U7Rp@y-UX(Cf0C<qvfLD3qMqOHkvi>WB}7F%&@Vo`Dih^J?8i#a>B
z5)#9ptN@PYUtBgIKPBhq7T9Tkj0YtVHijx$kjvsh{?oI`$xlwqDYnx?s6^y_Q04=r
z0I<`<z$u`FF^egMv4)|UDT_IUshJTJKbp*bFqc8nAImM~%)F8!9#F#L1tmAu^rHOI
z0!_wSj1_23yv1RYlbD;7l4z#`av&(<R7qoX9z>NUV-Y_C1A`{#E#~6Pid)P9k;b>U
zGK(|wic1pnl2dOn7nc;>Voj<nNiDv`mRL}bnwN5mIVUym7He{TX<o@KHjvq=DVnTN
z9I%Lw;)Jn`z-CAB=9T6qr52^8fP*NCtt3Al<Q;IZf(dXI;slulRv`>>IVk2Bc-R<O
z7`ecZkCBC$k6DC~gOP`+2vlyOL^wAC0|Pi;`GG?<i?K+cgsFr%jj5I~kE?{Kgr$Tv
zjk%T)lv6+nphyE$HdpeLu!C|ABSQ*PFarxi2}d?pkq09K2!iAz7*d#n88lg{J}5Y6
zq$X!8lxL)tWTb+MRY*yvke>u98Wl2&6*5awixQJ^QuQDy+%LZ*)k>ixBePhcq$oA9
zq*$ReFEbfb4ueW*NJdjgELQOI_fzl;aSd{G@^OXeM9oYEmGSvWSyl?D_Lbx-l%y7y
zKr~=vCXiF&^RpE+lJfI&G_4duib_)zGSfg#%tdypj)GfaPH`$!2rQMCUjoikZXte}
zoJFFbq{bW&X;cJ?j9ctbFBD0F#5qA}BC)6_v9d@U#Ag9ndy6xqvLMy9s3^Zklj#;4
z#Pzp$ApQjx?8WHmRWKP8wDFn6@gM_0L1+Yuei2a0<z{2#Vgf}!3nLRF2O|q3A7hak
za*+zkQ1GY+$yD-ys?%l$Mg~v>vM`h|W^)xOfFcwU$?z!M4UN*wG=)U8(9JAX$Ve>C
z067H`!bp(`uYxm+6_QJfic<4Raw-+_@=Fv-i&Il{6p~6yK;c<hoSIge1FnM8N=r(M
zQo*4LDjgLRQc}xObMgyvQ}aroiG`FX2bmG#=BLSYiz7ZB<jnZ^B4tnpW-L+x6%XM2
z1SY_l3K6;>wIDBffkGElFEMbz!xU7O!Lt*nOca1NREqP!*$I?MY8fj9N*GI+Y8XL<
zR3}3?L!J=}LkV*>SCI}hd>O%+h$Wk=$cvGomWh#}iUkx4DNMCYbxaWqc?uB>C9LfX
zX^bh%DJ(4<HH<C{AhU|nIv9e%@de6;nyiUkj0_A&(UX${PP)agswXiI#0Td#g~Xy%
za2u#NF*j8K)WE?Sq@a`q$wwd-C={SM7ZShNa}dZhE3lcMx-qjPHCG`yKd&S)GcPp-
z<`hu#3udIAf@d01G=m}vjFHuV5|%=4VqPWK++rPtL<LYLO-;;0<V8r!E;U5~R0gLe
zra*lZl$w@Wl$w{E3avIXN=gc>^!1BN5|gv@%TkNda`MadlJj%*3rkarL9N7MeM4g-
zQzK(zV`#ff7o5{{Gt+bvb8>W3b5cP`uvj-gO*c^&RITYI7NzQzWTfhX0#g^{eXw`j
z5cL<Rx&tK-V^A8B0=Ktfg+LW6C{c7UWOIZuFftS}r7#3DXfpa~GJ=~>noLEYMo|%{
zRM%v@#g&#>R19h4L5c!UI{n3Ilb)KFT2WABR|qm2WDyHPl{VPL@PYxv$OV@oNYyM%
zpC((85-0<JwBBOLORXrm#T#6bU*HKUm_gxji_^6tIkf=HWKPV<(PWO|0L2|BVMMWH
z7N-{8VoHTnr<#mKnjo`4PP-+6%MsBamx6K%11|>?6QcklA0rE+0HXj?kpU<}azN^9
zP$3F#wSpUi;F=3uh82M_9atqK7{GZJOn`0Wu*uC&Da}c>0~K+_paA4x<Y409;$Y>F
J=aAqK1OOHdKnef=
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/hickle.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/hickle.cpython-36.pyc
deleted file mode 100644
index ff8228b3857699bdd27288d585d63a1bcfa08c69..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 17742
zcmXr!<>i`^S{<7f#K7>F0SPEEFfceUFfbGsF)=WtFhnt=Fs3l&Fy=BvG37ExG3T;G
zu`q(fm~&X8SW_5Nm~+^o*isl%SaR5NIifgnIioni@~k;rx!h6QU^a6OPZTd$EngIW
z3R?<$jzF$plwht<lu)j4lyI&{ln5h33R?<Cj%cn}lvu8Klz6U0ltiv%lw__{loVKw
zGe<O6I!c<6p;9J=F@-BfCRa8}7Hm3qj$E#MlsqFtI76NY3qz_xHdoORMg|a!WXNNR
z0J)YYM=@6^N-0-4N;y|0N+nk{N)@b^IY%u@EmgVbL#kSqdXz>b57?EOQCi^)c}@_s
zCNLJ+fy`oLfXZrvW#!<q77$smPOT^{h@B7;ET$8sQ^}Xgk;2=|n57$~7tW9;#ln!v
znawqUv8Vv5moG{WqQjjbg+E20g&{>CRkoQqO5dF!MKDFEg&{=<%r|gnND)pEX<<ka
z0rL&r8B#=3#9A0q#K3$bcZL-46p0pw6p2*hX67gpcZL+n6sZ=56sc7HEYoJ@C^L74
z6zLS17KRj=RQ@dUX67i16x$T}6onSXD9aSP6vY&!7RD&66#Ep_6txz{DC-pU6pa>!
zD4P_A6wMT^7RD&s6vq_p6rC2vD7zG=6x|fP7RD(16z3HE6oVGVC<k|j6vGsw7KRj~
zRL5rKD5n(T6q6Q)DCZQ@6tfnFD3=tM6!R2|7RD&o6w4H=7KSLd6zde57KSMI6y{(C
zP1i&bW(Ec>Wrd8)<m{YOy@E<EF6W}u#FEq$g``Ram&ClhN`-);%;Z!BBLhP-T>~Rs
z17kfdE)TGBh0J1wL<J8QH&cbA#NyNxh2)(4yi|q!G=%`Lyq<z*UU5lkVhTv8IJGD<
zF(<PsGcR4CBwry>p#ZFgD=jl8RYw78U`lCjK`}%cY%EApPr<VUWK&9Nab|j6Y6?g?
zDODk{SfMyGH!~-(2qXY8Od&ImtF$;V9V}arUtFA-1Tws!D8DQ-1!QreLQZ~SiiV~_
zVqS^@$TAI0g|yPV<dV$%JUuQhWn~4w{F2mSxCb*5ixo=B^A&OvGxHP@Q_2$aN)ppk
zixu+AQj4ITwBj<<RPb>Oa(7h-@O1X}aaHhe^mFm?^mEr!2+ae#3u<#>QF<!W*C78C
z=j4~`fV@$XkqT3pTEtb9nwV0o01`{hE6FT^>nO<p8<Up;_Cm2jW{E;(UP-<}Zfb6R
zQ6(s-6_WFF3lfVmi}Ukzz_9`j+hT>pqErQV;HBjkDdd1%l9E`GSe#l?tfvqHG8&>T
zKQE^eY(Z*@LU~4No<d@Ba%ypLYKk5gmyxD|bC7><utI>3V~Cr7kgq~;Xh48}P>7y_
zV@XMBZb1pig(dk4r4VnY!Xp{#>lB6jJcaPgyp;U%Vy^r=1)t2k(h6`?lz~!MS!z*n
zA}C~XGP6^2DnZejmYA8NP@Y+mp^#cslwVY=0FqKr^2{qvO)f1;RVXeg%FIjW0*7*{
zk{;OC3ZO(&tWchrlLPjNLRw~CDmb8X^7GO`0iKj!lA(~Bn4FQBms+flkyw_>m0?;?
zsgRjhT#}fR1NQ7oP${j+e2YIZsW?BUv?Mh?Gq)hWsN@z$N@iJRab|wrExv-H%)F9#
zSo-nPWV$7go>~&0k_mEFQYk1=-4by@;`!y5xaF7TrD!tV;>pP@Ni9muiBBy{%z4Sk
zz`&r%a*G3`y`-`r^%iGweo;w$c4}quEy3iB)a2~=%;I>kj-;H_TM|$qkO@gSsUV5m
zc#u9#)>~X4x;Q>1Gr8mzUq*HgSRBe1gvjOPm&BLk$Ahf6C6b(<SCW{Sms*6ZPy!*E
zomz<`qse$n6l!^Kd|paoQBh)LJSb8$8E<jARwSnufPAgVc#Fd?KQA?;vLIEH<rZ^5
zq!EZPzQqFynY?s}fwy=w^GZ_FVK(1lgA_tl>IQlSdIk!*3I#c-iN&d~Xi+H31Qi+u
z`o*awr3HEgmC2wef?+0DO{&1az>vxi#hAhn#gxJr#hk*F!ra0T#gf93!rH<R#hStv
z%%I7Bi_<MLC)E{{sH(IS9P<=XVQyC_O3Vb6WtnN<ObAL$pfs75S)5U+kPM0_kd+`B
zgq=Z^Hwy!(@(yOuWb|9fpvicPr8qSwtw@@IfdNeX%GEE>FDS~-N=+`&PtM9JPBt;p
zPs+?o*U!z-Elw><Ey~Q(hxlDTCo@UEpt2+*KhIdtOusm@BvrQ{F*zGl=s=a|8ta+p
z8S55Q8ta+qrKYDtg`t#QLFFxu`1s7c%#!$cUy$QK0mHypB?R}c9!wgKXBdMSG@1OW
zLY))y^72a*@(WV)z}ZSqApjC51&PJQ3aObT8L34IiD0R`#9UZ@)PV{q<R^jrtfP<*
zQUuCkdLUPT14WauNCp-_V1Xi81_p*(?D6r%B}MV^0Z2|41Ze_Cl^#^qPm>MgOOTcE
z@wd3*<8$*<N^?@<<8SfA#}}3+=0IiG<Kt8EljGxy<Um%*g9v!6AS^ACVqjqK23g3)
zz`(%4$im3M3>Kp(f$=-%<QJ!=z}-&OWCnIT_8<iNH5}v{Xv|BXIfR7xPk~p$3ZR-W
zDK#|@RL6s=1M(9>9wa#tlN^dbQ5J>d7jaC#fRl@#CR-6G;}<D{;uzGbE>Z@u@W;Cf
z0|P@S$aJK52Z>P>?>v6_zKJEt8Q=t4<%6$mfn*N_P@xPeVn8{qG%qE!sJJ9QKSe<!
zzlcjgK|!Iov;b7`q^4-%$*>II(uYv8NCJ5lnk+<--AO{4Z~=SFN}(hpvl!y^{4@nf
zkp<1~3dxXCBPmrOIX|x~wFuOj1h-WZVeL$O4h6+|Dw0D*&>TuZgsU<zFvNq521hs(
zIHQBaD2i~-kbIYXaG+F~VU2B2lPfPjPd78SASX39H4hXVutq~-PG(6Z#qpX0@)I;(
zh2YL1AyUo2?uSMyirYXLKTo|x0i5@deNdTNqDLqi^O4*oi0m#3;!ur&fguZIE;tS$
zg*8YFB@RJVGq_(ZfYBNVX3%85B^V6tjijc8Cl=*p=A~DKf#ME1N0uiRE1+gcE^vMf
z$w*ZIwVfd?%%oHWSkERkMWM79)C7h!NA<v}{WMvymQ2{{A$5?qKp51Z;9y~4ECPw3
zc#Ij;_y8%<LGxHULmDHvslyS)0%n7nI~-B0?F=joQJ^Ld+bup1Q}?3$(t_}!#Dap<
zqN*bbV6H+rgs)JNkyxSt=_{pzI(8*RiOJdVC7HRY#R}QwiACvpkN`jy2loJz^7C_2
z6Y~_(auU-arX(lkDS$(!1k{RmOUx-w)ln$Uhe(!`=YxX=)U+yANQQPd6H63QGSkvh
zi&FDS6d;y@oA99Kxk6?Ns56wDn4{q9VhT!<8L7pP@P+3cP{@G7jUN;`pc<frp$0S<
z!PLxL!<52k!%)Il!c@c1%#_7k!&t-A2WsLl``zM3@=b^)%Pscw)RM%KlA>Fz#ia$Q
zMMWB*RtX=(k@28tFHSAdWCpigZm}d5r5E2~1BXp9r~w=wpPZ9eTpS+{&KTepO%W)f
zT0w3CIgWu*fU!yzH6HX(4TpOOq#lI99@1i9U?^cIVFY`C3F3iT<`Tvl=6L237Pv>5
z<5_D!{d#7<B25Mch7e7*B5hDT2=)aJ)S`4yA}G=aiGz#<tJGuxd#OklB&rAUGb>n`
zCL=^4IJtsaXhjAL3=Ewh|8RhkBBKx^8zUcMl{9ugLCj#sngc)q12*jzFEreeQ*&})
zVF$7hW;xhikQeGf_JZ?0D+?nJ3pjG9mOXeqOhJt`SnfEZ0A(qF%5P9>MFAuYNx~y6
zmw=N<h$c%BIK6;VlrbowfT9qZ84xK47HgmwL}Y<!pr`{SXK)sfK?!S0asYD;b3984
zD^d<%L(Ku~kQ@MZFDUJT6AdI$gM9<aCm5-}$Pm;a$C~zMfjoke_NB0ThNz@pWCRLE
za7cj&SkQqJJjlfpK$amTc#zmF#-fm9P-zFse;_$-1_lOjKXeDEAIezE3@Z6RrCk<t
z8dEb<kwpzd7E2977Aq4&ElVw94MP@NGXo<-340bt4NDDUGgB>VUJh6fXERe#NDV_R
z3yNF-SdOcNJ&QYwCyN7Y=Cn$#8ip+1T9$BzJP8(t625G%qAtkrY6*W6BV>5BhM|@<
zg{hV~kFP|ah9OI^nURs9M5u|egsX-%Ua*F#hBaQehP8${jR{mKC%$21U{G)bcP7DI
zgw!I1jMSWh)FSv09b_n|I6pTPG9ILmk(if~lbM&UrvMtnDh9PY!CH~(my(R4{L=Ic
z^e$<l0ys;9>L76cG^@0@1m58XbswRXonujYu@%@jP+L4rK?9-#q#4|q1#5<kYG_)4
zO9)*^WtEnhTBM_ps+X<@_7TWPP(K)?9okNYRA3-ubMsSD6*P)Vil9b-{h$C6)l={X
zjmQ<H7M5lfrKTukrh(-^t&F5fguxI8APhvR)D$#8)w8CRLI|ij2iL3$nR(!mGHZpB
z(xN<t{IoQXxs~~)3gwA;C6G2c$cDrcxW_@R1Y`6%x+EXeWJoDZPDQGuL7psD(13V8
zzX)DwD}aZNG@(Ho;-)Ebiz~AjJnWX2oO+7|r0*6}X8tYSkkpD2Pk(3$e2WDXdbe0W
z0dk8q8Pu=7#gt)siv?657m0(i5IAEn$7kf+Vg(see2W7_fwggkCFYbuhI)!Xy}Mhi
z;Bh-h5diA?gG)C}wp)y8MOGlQKqbO0F0eK6pe9Wbs3m!eIl~lbOb1-B-QosW5)T?3
zNiA9k%BrABih-AfS%8s`k%N(sk&BUqk&jV?k%^HD!ee4%;$mcBVq;`uEOKLDV1Q?2
zE(Qh$R@gX%3?l<WIzuf>g$k(7FJUTSPG>1$DPc`#c43I+t7WZW$zrQv1!bo~iExGz
z_8Nv}#v+#zjv59~%OH)phPjWCks<F&31<yM30E^yQ38hCjuLK=t_h4q93?y;RxMi%
zTMA<~N70rNUWhr(%(d(lbtQZtSr>*_xfrHej#|zV{u<62_B5sxrW%e+#v117Odz`o
zCxkQP1=X<Cuw=6pox`xr0>w-lBF%gf&X6a7ZWegnf+f)bR9Lu_<`#eknZW~Q&|C#-
za3(5%OJ8tt0UEdmr*?2|fh9dq=M~n^0rkC;!Q;LPMWA8j;><G8@N-fnxI6$4{bNoQ
zfHDJW*-%g!pP!Uvh1M4?$p_CfAhKpgJj87p8K&R@L=Vi>w1PSrsvgqA0Ck!XIzeuX
z&rDI!$jmFzv{J~-OG&K&`4(E<R6+wly;z|rH7Btovn*91CqFq6G$0Gg@t`RM*a$Mj
zP`6v`P&55B1#Yo`%86U-ph<+B)QVfINtGq3#kW|~a`F>PAel@8)GlO8ttd&&ODPfp
zafF~vO3*knv~gTy3X)&}jXxKGYNjF{5MKaEA=tyWI8*aVb5n~FOHyyK#)HCFlj{~6
zBn)q{Lj$SE9h4tI&Cw`+kdj2uV0Q{=yrlRRb53zd6lZ2hYA$#*`4(43d~s<K#B8R_
z%v+oVl^}uqq%25d98|x8b8S%|0|Uc3P;Lcv8yR?57$q3_m^m1Qn7P21gPDVohf#=8
zhK++!hzZPM0qeQNSpJJqKN*xuK~*ovU!aN_JaQ<<$iR@oSj#qnv5Kj3RtY#C)-X3S
z$1v5h*RrKB)v#wVrZC$ulrW{RG&5!~gZZpB3@J?Mj9@mK4MPb_7Hcg>3VRJl3Wp>^
z4Mz$mnB<aVNMjabsO1FZyDY{OZX1Rg_AIs<mS)CUt{Scso@|byNj2<Q><|&KiCkHX
zDIgO;B?4C(Qw=L9kA^ekC9yD+K;}YOAeD6udlpwMcMW?MH>gQaWmd^o!jr{Y!wo81
z!Wr_EpjG=9i0TslEP)i>8g4NLkUXT44`)c>DdB8lj9@_UYuF|*7D>QNjbKRO31-mb
zOZ>vbz@QLblv#q98A?$oN-Zc#El$lVfh5%YG*B`IC2LT61r@E}d<zmxO)demp?MgQ
z`|^{r6f_`RcTM#Cm#<J<l3xUxL;y`yLZ%!+c^6c2TY-&(8m$B7fNBirtP;3phu8B^
zRVWoA$P7@<D$mGF&Hy<XVinlo;E``V1xIi*1Dw%8wLj8OIjHBHQw-?|gB%3Hpvks^
z5>P6KWKd9r2WuDPgJ)I}a}*Mb(n~?3NIDAYMd}I~p!q@t(6pVVj)Ho*x`IYID1bFV
zwy7toD`<fF`gtju5CcJ$fUswJUOu!r0r3Q=zYk_Xf)Sz`R1_4Y7K0|caJoWI!Lgtq
zCo{DeJo^Y!4JyST(~Eiv{$Pg}TPc8Mgme_rt1=68P=Yb1DosH{TcH>v0cwbV+XJ9!
zI8CrJY;75gz|d3hOj9UMMILS_N-cm)WP_(xK<yductSx@KB%<<HX0NZAPgx%LBk86
zFo2FM6eN~pz`PA+gXSSK^As{t(oDf=1svmvWzf-}l6*Y{m(;Yx(wq|T40cg|eu;uU
z*c!KDP%|CujMQ7K&OVO8!9|jwQjtB(H7MBA-|rR+s7$NkP|ef>qax7wRS~F@@e(wu
z7sVG48RFsZ7Y|kLSET^1`9QOysVRC2&iSP|Dd4fHqSV~{vQ!1tVogDC^WqjOXe^_+
z2vla>Vh2sIf+jARiz|z7@j!X;nR#jXx0usYOTe9Ra7kAL8WaULdy1?<?qtd@zQqPH
z`xd)@FnF-=7DqW`<gEA>3y7u3QxwF&zz_v)c!6p{q{b8rDEx{*<#>?`s6=-K8(L5n
z1uF2%Qj2c!WW<BQ3sN=SVod{OL$qT177M5bU|?imC<b-o7=&4v*q8(uxflf)d6?K3
zr5L%GxR^K?xfpqvgct=F*%(0+Ied(Kj1o)&ENqNLpfOUEu@z7k3fyf1bpt_NoMy&a
zMo_mYld+bmhAE4=gauT})i7nT)-aZ^H8a&RgL{^YiMgP*M>uG-3>0mk#SM^L1D<Iw
zPgN)>sstrBP}3M(;MhZZC!j13Z61Kr1*ne9%+E_Ks)Vke0JZYV5_3Q!VEJhZi3-V}
z>CfVl#1inZv7Q2GB}7pwsL>8tA&{J3RFs+wX{Dogl|Z8wi3;hNWvO|P#stW%7)2*2
z^+PIM^n$Y_Um+(yKU=r7K%pGuRd{O&G{Ik-3QGK-2m#ZX#o3Ud!61m^5FMsm&{T+(
z0zxy$sNxK8f=x<QC<V>5f(i>zLk`q{02!C4ke6Bxc5`tm#BZAHMKPds;|EGN!qBD!
zDD+?>a<|w@3sMqGQj2&&%0SZ`w>a|h<H5FTvVarjEyj{49#9w*$AkP;T5yXMERCMb
zL}A85$}gy5b5Obiwb2>4IT$$@xtMsEii(ii@1Pt88r}f620$~L&=kc|!vbkd!&>Dj
zpn*rGsxJzVbv6o#FrUI&tDx!wQdS_Et&q?~)OdKK4HPYq0vx0a+LTfl;Pxgs!TV`4
zf<sM{6&xi+!JycJxADM3rD)Mn1e(M40R=uNa2dEbn870OZWBlpo{>QXObufeLkeRJ
z6R3#-8XyIa`7(l<T%ghMW>9mBIg2HQ#fAZ#nOUn&AtC~l?7$63EKvcf7_h~{U<%$U
zc2E|ISIxY|npm8inR$yX6I{=OGmj=KxO~6GTwGEF7P!ThnwOlPl3El88tZ~&5~LUb
zO^l*PhCc%XLkuV~K=l~|2OpyhV-ZLU9vz@63zQSUxnLCo0|QEQuz*TeP;U|?I#?jl
zQBegNqH1PvVTjF(VX9@VWh-H;VJ%^AW-1D+VaZ}hVXa{WM=E<BABbPV2<ibhGuE=F
zuz}mq;7A4Az}k!&;p|x~DeN{3B^)(u;GQc-3MW#})g+uDg)5jrlRHrt)INn|OHg|V
z6qwLn8K^K#O-#;E$W8?<6H)+`J&=+ck|w~b#t@}m2&j7rF73-wiz*c$D~CWsEAVm<
z(qt?xPK6ZY@P4NPxGMzCETGJrn_7~Q4_Wb6nwgWLP+Fjnk(pYQSd^TR32Lr@otT@S
zlA2QtY3HVZ+M}Qn1U_?*C<z=vc4ZcWySqi1C13+e^5JE3W?s4ucvV@kLUw8aXmuH)
z{|hOPK+6L`t^#3DegLn`fiCaM%mb?iEw4*0D%Mv>$pqC*pp|Km5F(-MvQ|J;5J&+H
zvK52}Nsa?K1)kxGia{k=35WnyR8>r>Ma7(;(b&x767cG*TO6=;RFFO`sDbCF$qCMq
zMWrx(MP(osC_mm}1@&x;!JTPH=Nr^()#SidZborufR}rLi>_PD*{PL9p!FX`*m7_#
z0|P@QC<lYO?+ko=j516@pf(f}A0rQw3{w$E9-f;)xg1o$f^)MIcm^s9Gy}z0%TmLb
z#RMs4Ygp15Q<!?0YFSg5(-~`6vzS4JE2uA<!dk)x%I#tdpj-@ER}5NtmB<HbJ$a@n
z<bjq|mLz9@(qnF7N~#WI<&{Dr(qb$!(oPDb4G!)|l%(c@Ob5;4g6n$F%05uMf$(5V
zMxeE0pezmUFF<CI!9{hIp(=E(1;{IiH6l9TwgY$tBV>XeF%|8n$#IJfV#F<0P>Y}#
zQoVu_S5X=$nWcjWP-+FwGHbG7Pg)QoQa}y*TZ}2_6#{62zYUa}z)d3#E=CbX9wt7<
zB9J)ti8;^+Df03>@LV-$i5>@Hi5}9NHFrRzGi-rkqAMunI6{hKUC^RKXr~;q5EfFT
zf}5?0Dd2zsuWp4TA<sO8{Gyap@LJ&{SW%j&P*Rkd3Nfj;q^LBxq_ikiN1;4bAukoN
z&NwM26<i*pn+&lv5wykyvVaoYY5*l)NS9kfQvp(ZLrUNP$RK4&X+aKXc0Dfz)LkuB
zNXZ9x+!Awg^2@<%YQW2Jixt2V$)KV2vdom!Vz2`e6>>6*OCZjLb~wOO_Sn|hfK=)z
zfGyJjSp*sm0F|W>x4`>BpkfHrd;;NO(BL>|ycsgb3>ptgVW?q<X9kljV3IWjvd{=N
zuUrR8L!d=gxA+h%MnR42B1@1=IjCg;S?C9w+CmJ&fwV0Kr5#YCfq{{a5i}@XC4v;;
zm<w1yN<k$OxZk!0)V^TIVklwEVrpi{W-C$v&5vg>NHWwi)_}{YV1^oyJh&&91=>Hs
z)XbF4RJ0_Wp_Zw_rv}v4Z~^t|ve>fNQ<$3>K_hOpEH%to94V}l47IE^tR<XKv#Uxt
zYd~UbpfT=dPzlJC#RVF1V`M0F2nX$_U~gdnnFwl;l(3X=*D%$vH#0Reg4__ykirqn
zpvhIWQNbCqB1$2xC_h&L$*-V}4|Ek(GHARCR1AQZS3$dMpdt&_yHWtHFV%rBzXY{a
zaK$$`;ehnQu||H8LOyuiCMcd1iVIScGt)9bQ%iZ^4j*`FEo9^cqz9>Hiv&d}C<eg8
z6Yw+y8leEi38*n&EW!Yqxu{`?<pIS(3WFrWM5aO((Bd?(CUE(o$#jcJ&)^mdsD=Xz
zKnT!S>Mt%EkWWAZPj-7jX#-?98$*>kQb_4R7nSMRfJV1-itY3edi-Ag|NsBLCQlJ)
zr9csA@y{(5&?0I`6$|P<LaJC$>b%8TT#%DlQUpq*MS`H7B^xB;-eRuGEYRctFNlcZ
zg4!0JmRHmQ3V=osQ3)bI8l$*M^1<WJAo&WAC^&(G36O_C<LTffa4SK+0VR0`cF?pC
z2csAx3#j4+4OlROx|DoOe2jcdRl-<&s)>C81<16b4v?KjAOdVBh=64^ko`rV96B3h
zH>fh_U}s_EVFQc7qYqXDfU^`R#7aQ@Ck9YcyNNLcJTqJ61R7mKudgBd13(QvaLW-^
zWP+DZgGxd8ifV8gtdfAPB?f6wP%Q=*(BK5_2U-muRGJ4`^8+4P(qt?GhcC!4h@^3g
z7uKPV2N`h)<R?(-U|=rl0(Hbe{s0ZjgTz3^8IpfMjg}IoW+qUQ1SQ490#HRl)Bp}_
zD+A4ja6p+596+$K8g-C}I;f*jo?irS&O;jD;9*B-fk0$~8Z_21IL2_8i}*mv6f`&k
zjxkLpNHiCMMA70}7!<{@!5!!j%n4A8g9;i39v-HmZcv<q;u#d@pm`?HfGRl7PeG<)
zm?tn+t%D50#4y#e)H2twWHHt-HZ#_;PGGF`N@1#D$zsZ422~uzCM7IQj42=$FcCM9
z2uMW=vm`?;TP05kYYiKyXb)$|Q($2zVaw(!s)0_5u-34put+k1WFb=^HS8&@*-S;-
zO4y6mlyGEmrm%qq5>r@H*g@O$YB?AgYB;iZY8Yymnwf$bN_bN^ni-oI^ETD6r*LL7
z6@4mUFM0vi#|72Lox_#OSIf!BP{Rq<$5_LZ!qdx)q_?bwJ%u-$X#!)B4XBd~VU@5?
zU@S5Mo5V-7NjhMYSSK(RWt8wYGt_dW@Yis~3zP`faMiFjGl8d^SR@%hJ`M(L0(4_x
zC}9C_uiF3(50I@(z_toNZN=+vNSK8)qzKk9Phc!MQp1uZ1ZsRlFr)|uGiVAWg0i}T
zPkv&G0&M*-I3IvF4i%*)=jTD%Zs-%ikcJ;>HUafo!IQKg2DJ64gP5WLPx*kPK;zHQ
zX&`Xs1l5EfT%4Ge3SUs+l35Izs!>QREy>SKEXhn(NJ#}ZmqCNSpz+(Hq|A~c(C99-
zVuv)^iXm+<kZ~aF;;-Q69|CTpL6%hKD}dKaDwO8sq!xp^?5Pm%W)?#l4WL$QX)$OK
zadCcWQF1D(aT>*`sR|h-B?VUc`sL;2ddbClSsA5zsVSxUb;bI*;03D1`e0|K>SdJV
z=75*Cf%{wqIf;4TNjObN(1BbB8n~y@BuEixA%9UcD7S;kYVh<%l_hlJA1KfjpfeJw
zDWFYSVTL*iiHL1wIjQN1$&jsSp!Gg_dYX*4cp;@?JVdT499qM`Ck>%ZX|UO#)x4lC
zLrx`p#VKe$M*-R_Hv*@o#5{%koD|3cm{gc^^on8`7#ONNkX;IDC?pnv=XsJ+!Fr%>
zSI{Uk(j@3`J`{%=X)+aMfZ87{pyc%uRH#+?5N(^Dp`HN&8^M!JRaVGGW<u6BLtUMk
zg5Sx0n!@1eq#|(P4emmM+CsPZ9E*!ni$Dtuz!mo`=FGg3B1cdRi-HJvH}e)lQ4iRI
zpkloUw0g9t7o-5xH~}}+Zn1(Zb?|JHCig9NP?UjY?m@v))CW=upEoMX0?jfphq(qt
zafZ1D#rryX`3HfG2b+0|r=T((-WP#1AHbChxHUAHfq~%$s2&LeRU3RPOe~B%;6*w@
zjG$o<CPp4cK1Lz%)DagW4<iqFN{IzD+|MY$#KkJY$OdCGae@0QT#N!tpvfkPY9UrO
z#v+g%@YWS5GC>1D;F<}vToO_v)iPEJlt5}DrW(+yWX53d7HUR@Cg?OebeiA{tl$In
z2Z~cmbU@ur(3T2N83S4#qL5hvntBD#13~AOz^y7s7Dg@oGUD^V3(Y{2+TivVXf3l9
zxcGx{p=BbdHv*bihL{5DcfgB6kYlVsi)CPCZ)yt6Zb%siDXBoQoeJrSgSGl;G8KUW
z16+A)GT&l@I4Fv}pc34Mj$#Mt0JWsSF$N|;(RPamEDo}@IJHEDiGiUQwD^L7lY@x~
z+;(PTESdpo4}jW!poj&HMu1d+L;V4C;0jdR*D{qbWHDxe7FsZuu+=b@u-7nw*E60f
z;b>;4WvP@X;jCfI;!0tZWJqBG)#)`X@!a7Id15RKB|O<&MPZOS4i-x_j9I)jEb+WG
zELq$s%-KvuWnkSbk_<I0@t|R9?h^iN_M*NLff~jv!Ddh!fi;f<!meelkgH*JVTcv0
zWvgMVVUuL2WiJt`VF7g;Ygr2CgfrBzWC=Gj)N<4?WQo*pfa=mbp%PJW8-^1s#t9Ou
z0d>h)6BR*iTFhD!7K6#9MWA&}pwZqu@OEcV6oY~lGM^0!TiAdGQV&D}S{0zCBjg;W
zpn;@Z(+XCe!gC<p8n_MMqyVa0kVfDjV>;kL3dppZCNx<>l9LXGybj3R;)2YSRLCYN
zka;i++S?8GSz=BOEL}q@UGQicxF?D_JPYmwYKj%vfQlMB5CK{Me2W#lK_Ut~tx}v?
za*N#s;yTE%ENIOkxGz=Y1=0a77D4^aTO8i0mEZyqF$z)yTE$-kN|%sc<P?w&aNn&B
z!~&&WO|Bx)B5CjxOcXbGFgG5&)+ve?6n4d+oyDNOK@>k^WfMGzqu4UQJB2|ssNqsH
z9aKVc6qhD}ThLML#idCg6VMAokQ>ZEnF&-bF>rG+axwEUu`x1%)<ZFJFmW(}=F+(s
z`M@j=5Y5QL0-F0Q0_oIbDl!3YF<=CTrzUez1c+M%8YR$#ieaUgz)C_iJ-}P^K)W~7
zN<jl}@$sNTFc^wJMP?DGonMpy3L4NbVia#dc6xkKDx|>!S<X-d>d@X|PAo1dVh7bN
zta+um1(mm$@(Q9@Km{B)E~D5X)8fUDat9p#pw2=OsC2l+2I?G@7C<JA!JQLuEP*<~
z-~j>18aFO@N4B^~7;F<L7K)2NLmjucJ;A%e!Lc31T%1`^1g`nSKrRO_Oa-l(D>4Uh
zEkJ}BC~8<zGIMf@su@80U6Ub4B|wH`L3`6~amU9)tL6B3@HSNNs#VC64A6?TTil_=
zsYTFjOW?8FqM0D0K_jb0wjdTbVS-kdLKbu6fW$zP)!?a9@Pr+BoF6<&dy54$-~vg;
zpnf#CYzEh1;35?q)kp~kQ6_WPKsMdlfxKA^s;F6*M8KsnXh8-CBL}ktqYM)dGpJ4l
zEr(@el3?UuW?^Jwfrvmz5hhR>&c`Ug$i*hWCBP-XEx;+jBP1v!$fe6=#AVJQ$|b<X
T%O%StDxl86#v#Dv!yyC!{UjO^
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/hickle_legacy.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/hickle_legacy.cpython-36.pyc
deleted file mode 100755
index e81a055331c0c861fbe8dbf300783bb85bcdd730..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 14836
zcmXr!<>hkDTM_fdmVx0h0}^0iU|?`yU|=Y2U}RuOVTfW#VN7AlVT@t~(@arJV469K
z1x&L>v86I)v1f6lFlRFrS*3EOFf}tqapm!(Fs88NaOZMGaWOKautxEuuto8@Go-Mm
zaI`R_aHR4zGe_~eGo)~)aJ4X`aDn*(?hGm1DLgF<DLi1lpgThfZwg-vLkeFiTb58W
zbCj?<LkfS2Knp{PKq_07NHcSkXo_cwaEeF^W0Y8mXo^@1LzH-mc#1>|LzF~{WQtS^
zLzHBSbc#$1LzGmCY>He9LzHxie2PK~LzGO4Vv14=LzHZaa*9d|LzG;KYKmG5LzH}q
zdWuF1LzF^_W{Oq|LzH5Qc8X35LzGgAZi-$DLzHrgSBgQ3VGCoFN{Ug6aSKC~YKlpU
zX$wP?T8ejyd5T2~W0ZP|Pl{!VRSRR3Mv8BWb&5?3W0Yo!ZHiqBLzGsEeTqX1LzH%k
zV~SG?LzGU6bBap~LzHfcYl>S7LzG^Mdx}R3LzI3BQ!s<3Ut%;P0|S?`LPlnCc1~)1
zPHK8$a;08DB^Q@-QEFmIYKlTqrGiUhUS6d_Kv8CLs)CV$p^>hEsjiWQ9v7DfSeHU(
zu|lGPhl`u3LQ-OJYKlT~PJUjhLVlV;09am6!85P8BsDPwBvhPQl$n^5S(TZWu27P%
zkjPa4R-=%XnUkuc05veBG`FA_EDbRhB&nz1Spu>tCABy+Jufu{B%PG1kXWoxoSB=M
zlUT%6lCJ<YEHh7`v^X&xEL)IYT%4Jdld8wXrL3&rmtT@v%*7Rwky#8fEKvdCV}*>w
zyp){OA{~Xa{32*zD3lq3T~(~d<(Z}sW~5M*nwSD|c3NU)PO*+c3CN7{%$yvBq*R6E
z#GIVe6p(8Y6+rU3`DLj^dR$yDK}p~5B{Ks9gC^rG4!``o)R4-8)GAg(J!3tCDsH5p
zO$I52VJ28MQD9(TNM(p(Okn_JjVR_6rWEEDhA5U4mK4?&hA7q)wqOQL_FJ58nK`Mh
zMMe2VRhkNpc?zi&$*Bb;nfZANMTwc9ILS;?D9K0#dkEz5w9Mj+%49~Ubs!pqok1BB
zltWV(f*CX!{Z@h;%~G72lUBsaz`y_|ezoWq=x5~Trs`*v7N-^!>t~b}7w0DC<r(T{
zl@_HYC+g?s=oY7zr50u8=|huMW?pegVor{JPG*vRL1jrsex9+OnSOC*Nvdu^Vsdt3
zdTKFLsjjh}iJq};L8Y;tnO<soI#d`+A=2k9j`;Y@yv&mLcy*AQLBYkqSS17x1wEKF
ziXTDBz<%UmU|=X=NMUSdWMlyQmC3Iv1{{1=3dxCidHE#@`30$Y;Na3z2*^oIEKXG@
zNGvW^NX;zCNG(!G1WV;5=B6qXmlS2@rRzY26!Md@Qj<$`6!JleGE569^+1l&WW2>%
zP?VWhqRCjq1`A4XcowlUFfiO=kB=`dDT<HRLh_9uNF&%$dQe$EO*W9nLDt5{-{Ojo
z&&^LM%}I@qzr_<DUs#%$1C?Qqk59=@j*l<m09nNaBH;0du(Svi(<&efK`!E8WMSf9
z28&UY5_$aceG^NPGr$h7^1zoEAwi{3Qdy7+N|T^aFU?CyEh;X_&rea%$S+bTE-fg?
zFDgk*(L{+yY-ttjFYKWU4hAETM?ndIfw4*y*^Oi*6&JAQtRO`I*!lTs3Xp^ijel4w
z0u>GUd1a{}KPY78fl54R@uH{6Sj5QyDi*LMh$2v=nIbt^1kK6d)Z(YfR>Td8T~K;1
z;svqzKm`71=VxGG&<7cf6zw1}ilUt}B;O?;94u94SmPX2u;%6G>1O5@<fP`N=7H*#
zw9>p}P$89=lUYKzG$ItkHXuJiV^|3895SNU4D1hR^rE;?AtSLkPrXDTDK#}u0of;&
zsU<{2s6CPk1(97yL2L>zFfdqx%teY#kQhpAg6bboSzoMyRGqXlq=8#qEgVtIU^a6K
zOAAL7OFIJ#LlkQ;gC^@OJ`Yp(qWsc=@S?<mg47~-CIp!WDm_5SrdSDFnAU(>a?Q*&
zOeu^u3?+;uOf?M6Oj*n|j5SPsjKK_=%zn4HONtVcv*SxLb5n~$G+A!3r>B-AmXs9T
zVl6H$NG&Q71QmsR$>5epd`e<TVsUDTCNnsH++s;AN-w^}mR$~_H5qSl#>Xe;Bo-IP
z$Ai-axEL)GWnf@%0R;xgaSV(Cj8(FzKF~uo9PS~IdJqPC2vi7{FqANYJ-`I<KrM3#
zV-0gWa|sLFqs;NFHOyd-6$vpgFobBb6^VdyAJ`W>P>a$*!B->!5*G#$AeEXdU@sMk
zfds@ser5$L(`1Aw1eevIYN1Gyfq}sT<R4IuW?&RzWMkxGtdhp=Cx{ssB?2f+zy{so
zg$7x2YEBL;*g$r{>;+p1ay__?1O*2NBM%EWR;ZGics)!(RT^~?6Ew_Oion4N4tN<*
zc!InOO*4o9hWQZWTSQ6;0EH~bo8XipgW@_$5(;w-b3984D^fyXLro~`kc0wuFDQz^
zK@5p|uy3#?k|HTcZ2?Y82#<jDhk!hSl0>AidWKL6DUt?-AlPgW0Sh&7x&S%WA7l?G
zz`*GOByx+fC?pwF8h}a{kRA{loP0{4r9uf~GgFaX4MP@F4MP?)6GJU?En^Ks7E3b&
zBSQ&m7F!K-4P!G?ElZwR2}2D?gguKRi>;Zd$fbs%mL*RCEXvu;RFqJ|P|J*>gQbM4
ziLr#ch9#b-hN*@np0|bt)NHNlQ*Z<~&%kZC)FOq9)SQCUA_Z7Uu27y?lA%zXpPQ<X
z3TiP!8vmJj>3UoW3S0^b;LbsEVjfsK*q~yCl8mDK()0}UMqZ)<sD%es18(tUl@^!4
z+u5MT7|2LXzFS<G#o#t@UUKR!7Eo{I77IxB7E5k^O6o1v<edEC)LTp$rngu?8RQmo
zd`8YKR*<^lTg=5JMc`Ih5h%PNi3FTJSZ^_=6)A!u48*;~0a6Y&lsUr`sl5(P?zgx>
zvhkocZE8_8C>BA9iGhuak&lssk&j7$k&Tgyk%f_uiG_)ck&UrPn}LBL8C;h#Fff4P
z6cp>;AdR3Jq=qqzA%!i4v4$a&v4&|nQy(KE16YhHg*lxug{7AX)GG&dOsZ5BAf36y
zJcYcJ#G<0aN>~f2Bp(rYw^%@>{Vf(yTj~}&RG}tgkq!d`gCCUEWCR!Qn#{MD^OLfQ
zK!tP>sPKdo|Dd!7@hT6<!g#0|<sc`6+|Ixx!dPU0(^sH64D18O8io`mko&<r<`kB6
zMhHIzR3oxgsUi8NKp_R>U~C=&yBzB9B3)1s&yiOU4^|HH544a1wZd?FgcsrwsJ2>!
zPxu&%K&@Hq@k6vvG;sR_ltD7{6cQD{1qQNTZn1*v<sw69#1Q2XArz0qr(`CVG^6;&
z1ZTY1F)%<<7CV;I#sLz`Vnj=WkT_!n#UZ$J%t>S#1bg!qOKxIuHYCQ_K+Mz>Xe@%F
z4U`-;8H+&ea!n?1YJ{+vi}1%JH#h(w#&v>x4r;4`QluDTkvX#8K`9l4!8y$jT!Mne
z!CDw<7+e^d8JihvnQNGf)Id#>8pbpxP|J#uAy1`-F^egMBZV0)KQe&^3DOx;*uY`I
z9?YQ0S)~Yzb%mVF;u3607v3fUWq*)kia{<0l@~P(u|l<sB@8u;C5%~2%}ikoj0}ZL
z;QlkCpC)sW94MVL<rNg^fodwITTFTex0q6M&@$*RPMh@9ywr+<BD*OdlOfhr$$-NM
z<S0Fxoc!d(oMJmYxI%E;Kw^ajqy#-qm~SyB=2hO}jfWc&pH~2mAWasuxVgoboS#>c
zn3-1`pI4e&P>JTcTbytsCV*TA>W49~2r;oR3V=!w(9k73-+^KgRPKOdD+rcSA+f}m
z#RwXzL6nBIEERg7Sd3MxWvyY(VoYJHVG#qjz-w4D5v3?g3cEQ&ElZ(FI712tIL6cz
zTuO5b6cV8kmZku9HZ<nIqhz46T>&YG*z*eFkwU2m)LpJ(QZ0T7V)$t?fm?%+&|yOl
z9Z)d^&h@vLGBa<1Ov)@t%|#2OB2c!R3kn`k$CH7Xi&21yg^`O<iV-XW4<JzL0by_e
z4awwOWC5yRA?djY)Xgfg0ttX(9TFzswvjah1H)QSa6r?tEI197losTq;!n)1U^N)2
zxyTljllUO18LR|naxMb(DIrb)I|}5UTU-eHSAv{{mWV+;3V0mRClRZ|5;52}pn+h5
zsTh<J!K$Ib15d_9PT<N1UotKNH9?C&J;5SZEb-;ez`(E#6fhXc7bF7@8kBYqsO<o1
zi#9XXf@&m2EbSk(91m{)D8S3x;#91~EmFQ^26cDQ8a+%!pkWP2W+hb3aDbz|IJM*u
z$Zw#CWnki96k#j^<v;Ac0=1&REgf)cE*-rt#ca<Ju0N1G#0D7~g+?G!4N6qA0a9y1
zj5rDM5vV9(U=m|2@<pyQQDO=1A9hf^iQymgwgE?#1SDaC2Ie6HktO-y<~+1K1(lf4
zdM7VGFO^K6K<X5b0q0Tt0qRe|;{_C7peziEtYRN<-Jiu+!j!_;%UH`?!d%19%vj4(
z!<faA!coJL##qCg!aRow()zAttzoEPO<}2FX=SnjrB0R<))cmM#uWBmP)i)tD<Y=m
z*JO-hDa%RC1IM5y(=FEGjKqS}Tda^~D<l}$OY<_5^HX3U0SypM)*@vF28Jl!;)2xV
z%)}f>OScGAC*EQPnFx+>jLL>7r6h_eCnt&}r8u<&y%2-6<Du4F1%(^97-QmL<Y1Iw
z<Y6odMNS2vat|~B3JyIEW(I}{%oX5P(*&m2Z?(*zoLj>z$<WLQ8ks6#E@3HQO=Hbw
zD_T;*m;wrArV{29<`R|^mJ-$!))KZ9woJBKmKv5Uc7)Dm##&ZTC}%Sj<$+sZVDoAi
zYM8-c&XK}i!wjy`IY45P3}~s3BZU(rU&ETtn8F2$mqHGN*(D4moLO8c-06&<Rvwtg
zjo_y-WwT9SEXph4ftZ}a)66u1vB;{1F$-#H3NOr5uo#+4vUrhN7$7}-AeW>ura)bq
z#fOm3GXk3cc54bhK2s5TP<+Cl%`t(oXbadBG<P6$fMp?e2*B(>F+~7u$}g}wVoVXl
zYKmYs#{|$w0?5yBQ%d+Dq1cRC+QC9n2+1TNh{_V-W~K>DMP)UNSt5{FZf2}yD-ngr
z*08}^;wU~~1KYg|)dwlU*i!{Kw_>D<EU^;KEb(lR-G30~Kx~DiA+RVU97T}aEz!hS
z$Po^i+7k6E0?m=#V$RGfxy726lb=`u$(Srj`T03Tehdr@pvECM(^oOas}@7^G`N}t
z_iu{KKx#PRRg2?6eHd`_9#Sp2fht!o5CJYILA^pvNpL{{wzDV<Boz)KKt(}O1c(JH
zuWvD@r<Q;#HE;v#7AtsU;1*ME;w|=!#A48xFxWj1>%lc3!YXh#g9X%!1UF=FaY5Qi
zU;$8t4w(`GjpS;w-(pHhxy6<NZqb5A3vMxIr&iu#1@+sDZ?VLKIGlN@<?#>|pz;?|
z4uQ&(qBsTyhI^or2V^+|BOjv_6Az;RqYyI(qX3fvqZpF`lK`U#qXMH6lL(^%lMtgE
zqZpi2ViaQ%VB}*IVG?4LVH9H)V-jN&V*#6a3o*6=s?$IiTyBB-m*Co%F@-6eF@+ga
zO4NXQ5-gyeA816qh8dRDpyLsU`WWQ@%>2B>qDs)zIDA}60TLatpe`AtZ2=BhP=5_n
zD1(C#>?n+$Qc(gZx=@1z+C2i<CC13WPz<WH7?=bYxfqMmkc(rGI4CasL7|tyz`#%m
zTC%~E!%)i-&X6a=!cfAP%~hnq$N++%syc_MmL-xQk12woh9wJ9*|CBvJESTb%x41^
zpy2L(32O>_2^*+l6k`C-<bswoa8zk3c%~`j<tyZZrXfKSi@Av@sXCB}aD~J?<hgNB
zs`Ud0&@DE|+5uSN2Sr7dRd`}iUS?jpm4a$9q;-HgN2#L#8n#l%NGwZ*ObNkOH>9R$
zvfpA!$;`>Q#qF63n!*H6DcxdEhAfG=#R_Vw6+?0WXciFc22JD&5<WRllm)6;xgpL=
zi3g2+p~VSDUVc2-v$CKl0ZsleFflRmF!C|-F>x^pFbXgg<$xjw<Xg}<EhxQ#{0C|z
z7f%DXG(dq^i&Qt3u$8cvaFlSCaFuYE@RabD@Rjgq36uzy@Mj5S31^9vh?ekYiIs?#
z@MlSYNl7p%1tz7zqzssp1(R}MQXWhyfJsF#sRSmKOT;t9YME-7vQ%mqOH`YgK%*Ss
zk=z{$;Tee~kf6^`gZ6n95{pt5%2O3eiYh_NG2qKS?70*W<G0{7DbOidP-=zd49EhN
z)WXutvc#O!yb@4PBvBz5G`(M3l2`(sm;{a4dX^{@rIwTy<$;ENQx%f)i;7Z{OBB#Y
zl0lQ2i3;hNWvO|PqyQ>cK>9V+z%dFbB|(D*MW7<22-Nc`0+*Md{9Xjgl;C^=DJ?<G
zT5!<}ZrT)q3d$l-1_GCnkWv@a3jq({-eO74FG`K#_Dw9#PEBzH4XlEPlA_qbCA^^-
zc!&_nGd6-U%uJ$KGV@9-qF6KYN({}wl(7+*GBb%{Ed|T5fha?>C@^ho1fk7LqBxWD
zL36hiW+qWwP^O`gMNuTElmMkKXaRDIv7{)8fq~%`D>zL;ic@fk1(m6{z@wU}Me#ZL
z`ProfI-nE_8q8#1<6vfCM24UdY0y|SHq6D$!&p=RvV^fb8I+ts$r>~P0xHr#L0h~C
zROYyW8d#uaTL~j*7O#f6nYoq?)TD+qr%PCB*qRyBm>{J`4X6<fs>)fw!`W<Ete`21
zaE81v7KReGY_6g?pt6dQp@uDsy_UTODqp2j$ydUW#aY7+DuKco@+2UtN|;iZYS<wr
zL&~*qh7{%!wkF011_(a|G<V37*v-Vipb%b^S(2)d2wlIFqEM7tP?TDnnpcttnjQxw
z()^?>P_YhL!vh+{hxAeM!D~&xY*3=(Qcws;EK1BxElDjZ=2B45g##`HP~Ncu5vj=~
zpnQV1Tp?efxFo+QRRKKe3t5H5r2sbA3QR(E>VP?*SqAvB7tnes*eVyODzN$BL8;6<
zh4PHd<P4D0Ar^o=0Cu*Xf+KhW0K7g1vRnqF2W&Vf_vRGmD<md^mN6xQmv1R3fF?_<
zpcaGe$OkWRO3YD6EJ`oUP0cIOQBW^ZSI_{h)KbXL%c<1VQBW^eSI{U2g|8;a67@uN
z1r5+-c3z66o`Pq3UVc$(ib7@@*kfQrL3V?AP~U(;7L?qJQj3c-^Yg$C04v0>T~EQW
zpdcqRwO9d3J;>{zTB10$L{GsV?BHT61yFxeM<Kl`vp`27rz%ZBLtCL3!~q+enWg}q
zEdnhm)8tZ6K$<nTQUEpDz^brV@0q3m8f7TSS8z+rDF)SfsRfW#eBgyxpwbe&V6C7i
zKczG|RUtDaHLoNyIWY(9-(rwYt3*LH6G$bf%7gZWRg0@M;Z-3>9=ygaHAPRsIlnX~
zMFCVR7NzFqm!&GG7W-*(fwLxf4hg)(1zZ%~VgV)UBGB|i5vXleqzTHtO!>vP*dSUp
zIUr+^@}TkrKCcCCv_-LjEKMx}&z0O_O#`J7^m2s-q$ChjoXCI*17Rj64kjK(2}V9}
zMaag;0h-=oWMdR#Dk=k2hD;z;3=9mgQUhEWdN49DOkk|if|eUhpv4T53@J>K44^6r
zya?5(hAD+blA)HRgs}$HEMv|4QNx_Vn$1+?R>D-n1fI-f&3jVAoWhn3nzV$eW`?Rh
zRl}UZp3PKL1viHUDzl-6IfWyesb~#ok~51PBA>#U!r96s$xy?T!X*h>&<f_IaQ8CR
zvgXy*FsJZjGZpPh;jCrNE2v>k5zJ;Qx&YM)ay4t7Nu>ys4>hTlHJl+&j)kFwBb#di
zV-XXiyn?8yWrdVY$Q6GI{~Wejwi>o9{u<^Ifo7&!_8RsYR#2%L4swq`iC{AWxO?hU
z!;~ciDl;P(YCw}^LWv2C3=9fB`H3kCi3;Ep2wi}fm<KB2lJoP5ONvU9OF#u5cvT=Y
z$3doHaTjHvVgjvL0yTHR1&IzQlY<n1(g>og;8IX1PE1Qx&;Vs0O)CYL%wkZfqL5fx
zlAoJcl9{ZKk_v8Nfl_{cnnGexQf5gJs4Wd`LqLk!Vm;6Vv5UWgpMMA>tAQ4Q!t$F!
zX<klhF{H%+assIBnO2matB|NrQdC+DTJcnzUs{x$3d(32#i^+Z86_nJR{Hwo<>h+G
z#d=v8rFy9;rTTTn`njdWnR)5O`d~Mu>SdJV=75)`f}5`eIf;4TiUCwOfHN0(Y!uR*
z05$B2(m{F15kz=`2vGmc58Ri6&R&6geMLSXIZ($;lMT{P0k=#*0uU+Cus3+751j8b
znINJdc}-z(?!LvDlb@IZp3u9+4Q4{zdy5Cmflfo*;)U_R!?}>w3}_O%C<;^raD^r2
zltR|7qPqgzZ-tr#nLI|A1s;?_G(K*zK_-pB1&k&~5vUZo#R7_Ru;YsYL4K725#Rs<
zH&B>!^HVU25s+J}LB$BD!N$PD!pOnM!^FYJ!N>(-F~V37z7`_~BO4QF2Azpfgi(Nz
zkBNy92055{7<rg@KwYCo<N^oOTmdcG0~a`;c2f;waR9t~#Dvs6(t%`6X!nQ-k~M1>
zQkX$SQVJ8OSg2*L0j;rOPGJO%ZGj4@U<S}A7N{{&!ji%YrP)9<c*Qq+Vmc^Ox}?Ip
zDhi<72=X;38-f>Or$F-|c%cNiMSzrfA(=DWG04x;&)o{T#2d6O4wN(Swm>pW)2xtd
zd~hoOVjL)*hGX6Y^|<`N{p~7yE_e;<5B52zbO4n!c?zX@*?IZpdEjo24mdh85{to=
zDtMhCC~i_ys_YbC0MzR!PE9T?$}Fh_tx7M>&ns5Y$ShGv2G_<(sS3G?Iho0s`K6$i
zNJ=WGp_5i+4%4nsnO~|<o>>fTm4PNhGxJJQbrf<F^GXwQaw@^)8Yn_+!HozmP1Yh%
z2fip1lzBjX(Oaw_7Zro6c1;#Y_Ns?Tf<23tsfsEX7#KE#@)W33%fQFN2<x4K#z$Dd
zxeKfYKJ^UBBOnaUBcLg`bcR~S3O!J@oW<D8(9Gz<5Gz;9RKrliBnet~$Oz2~j9E-I
z%%HShC=||616p;+R3)qca;`!mblwy+F{<FF$$X2g7`(Fh7E5+&W$`UGXq%7)yv(Wy
zTy`dcJf#C7z}*)x0qVmPRWdLzTmbn8G9t~wC<My;j9_uNA3$*r>cWBj02&K{E-7Se
z2Dyu|N)6^NaH*&OiipHa&}akHq2M90Vg*emNY@c$rzRuBbzql(39xgj7#J9Cf}8_x
z(=Zi*_$bZ*IR)$tP?*+$*Y;P5z$#|QL>$;Q1#rY^GTvfNEGh!CAq3c#8qi=E+!p2{
z5FfiOpuwLShAcexfR@9Tfr@=i#-cVX0SsE~@&w5)5Fc(AXuJdz@L(VLF)}b@GSo6o
zV61Qg+1bqC!Z3j`)})pRJVehV$xy<S#hk^G#oEkN<N!%9wag`KpkxB-6WTBo3e+%V
zv4i#k6bqy<)i8-M)Uwnt)UZfG3R;$IrlKN<zAO%ij=T~id4X)UqA3u0P{9mU5rm|I
zE1Rk4QVKWNWLD5He-<}XJx>Y`DB*%8ZzLH~czeNXO>G$RM9}0JYe3UFa5vSm!G>f?
zKtnP@DSUI-YuPJ}YS^=cK|R`9jvBTU{%p_`fm#X=)I|a{VAq4H35ZBIL!khqf{A1(
z6o_Cb6o6(K!74#W+CUmNxy2X{9vg-<G(h9LkRc0f1H7QtMiqRt7&Pk#UNFOsTL*Z6
z7c>wJZkK4X;#8?AQq%%UE}$V`@QhQDDu@fJ*Ke_cS1K2QR(##!DlSPZDk%o7V=8I~
zsQ{HB;Bgahn**o$kOl}S*MeIXnw*gQ77sE4TpFZ+SfItQQEb_%mGQZWkcI`QIZ@OA
zk^|>TFagdqAlJMCC1lVNCkAFvNg)DSaR+KmFbROh=D=kK2csI32s00Oc`p+q7YH(m
zfY;=K{R2<>pbQSe;G_>JQy?RE;IfXfYBD6^5*2b1i_%k3+T0q*<5{5QGOUsZb>2X|
za!@B#lM8PR4r<hr((nck$7w=V6KXPn+hmaH0vteK0u)NOxIiUzJlM^AObiUgpkj)F
zgM+cC8#HRoR3r;tUd>pP3gH$)xFMR1A({q7;2jU3HA=Ubiz|z7v4YDpa3k{;3&@|~
zbaIQQEVT&KeUHz~OUnlhO=aa5L7L9s*ak)5E$;Yum@4q@n#9tQjQpbb_*<N=u<cZk
zm6+iED`=dx2(+3DvU(S^kp{9v6SQ^`ywngp1Xl!#dGOF&5qKOBv~;EjJPrt2YXF{_
z2hV3iM*2W=n&8oR$gm)&c?juDgZivRpmrp<q6F78;Bp>Z3>SgQM{r63hZv;L1_v*g
z0EaP$4P>*B9mw6q0SpWbkcyXs5!6rSV&Y+9VPs(!;1=K#;1c2y;^C6vU;{&T4qgr(
eFlGnyIXNu2*g5z)RJpXcEVyJjI5}9s;wk{#qWoz9
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/hickle_legacy2.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/hickle_legacy2.cpython-36.pyc
deleted file mode 100755
index 477aa15d70a77296c4b1d4b98e55aa747dd6552f..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 20746
zcmXr!<>hkDTM^R{#=!8H0SWLhFfceUFfbG^Vq{=QVTfW#VN7AlVT@t~(@as!V45Y0
z6-={5u~o{XFs3l)u;+3_aWH~Zv*d8*az$}5GK4eaiLfxFa%XcDg)lOJU?f8xQv^s2
zQw~oQPYP=kZwgx!pF2YedkRMjLkdSKM>BI2zdJ(;X9`yfLkbs|FW}CQ!kxm?!jQrP
z<_o$rr0}NjwJ@acf%!u23@Q960xb+F0;#N7!p+Q4BJK<+f+<2R3@JjXtXZPX%u!+~
z&MBfPVl9kO;wj=O5-ki-5-E}?QY{Qok}1+DGA#^IQYkJevMF*cj8W1lt|^KsN-d00
zGAYU_DlH6AvMFvUswrwMj8Sqa?kVai8ZC@b@+lrEnkiZ>j8O_Ho+;WXIxUP*iYZ<x
zx+!`sj8RG{-YNPi1}%(H$|*i6hABoZj8Q5nzA45jCM}FnswsXcrYUAEj8SSS{wd}u
z7A=fX>h26FmMK;(3@KKr8qLg6nkfM()+shEj8R%CwkdWk3{l!C_9+f63{g5Mjwwzp
z3{ko%Ou-DAfmMsSloc{Ele2SD<8xBe6O${A^a?7uxSWep6H8K46p|_xToUv0Dis2X
zGLuskj0_CTbPbGj4UF};xI!{Aixo196%rL7Iu$Y!^HOqBi*yvy@{6E`DwG*1q-Ex$
z7VB|&rYVFOD-@+Brex-&E2Jf6<`nBFlz_}A&&<hDNJ>>mPRz+kO;Jb$83~ft%`ZzW
z(&OTK2})spnvAzNT`Q7P3raHc^E4T6arouurG``%q*k#S=^5ymRB<D@G#R7?hM8cg
zQh|YiA(bJDF@+(DDTOhLIfW^OxrHH$C50u0wS^&yHH9shL6iL!r(0%Ds%ueEeo>W{
zf@7XSD$G`eqQuPN)D(rxG=-9kRItZD0g#qioKdNe%m}p)M1!z1C?l{ifHFcbgC?Wj
zN(N2FTP($?IcY_#3=9ll;#aGFfqq7QZmND(X>n>%v3^EradB>9UY?<TR%uaca-x22
zj&5;kS!z*co<205GV_W{5_59&b25|k3o1)8^7D-K%=C*hOHy?U5|gtN(^HF~N_CC(
zO!SO(3o4EE%=A*z)1ksp3Ms;Gam2@G=4F<|$18yx4GJ&@#wsCrIOxHo@pzIkm_d`t
zuPW3zF)uH_L?OQ*H4hv@dI|wKsfop@3I&P9#R{pJB^jwj3W;E;yu{p8h2oN;%)E3R
zsE|T_QdVkmiH<@(NKuAqL8TtZ72v?qWGrHX1sqtQh@F9f;TC&*d~r!pe7q`>(*;4A
zz)`3NmG#qP1NjnUWqkZCuK4)e{FKt1)cE*YJn`{`rHMIE8TR=2l>FrQ_#zIFRh%FK
z9$N@Yi$L)TN>0Tf7jZDMFmf=1#VAU6{LVS~#i=Q9w^KFwfgO)M2*G~UK?*(zG>4E7
z|0y0WZl>VaRmeyzR!B-s%~MDQdy9gEkOxUl#3Y9zP?YH-`9&PlFW}_jr^!~t1&Us7
z5CN*Wig-aR{PE7mz`&pdG94-2L1GleJC9$!Z(>Pu1~|c1`QR&aAlXBqq_Q9tRP=yy
zT4`QNYEf}XetwFAMt%{Of`WoVacMz8eo;wkiYA^6%K$Ew2qg=1kY}L<f+(^(Nk|he
zV2@crN_()=^V1X{MHe)`D<nfo4NxtRpI4R&a=SuiUP-<}BD5ya!{<;?oLeC|R0PeT
z6ht^b0|SEz$Y^kcGl4TYNQ|Ng=M2er$p;5Yl^NF92GyW>`FXmTxdl0?xv6=ex;U*g
zFBw$#CFW$7R8kzTjvzll<5dXm91<ec4D5brq@uVDl=1V_OCWh4*$0)WB?P0<1<754
z$nK&b4h0w(80<mjg5wZUScAk+;*c3yeHViYLgZFLJ3|^HxTU}m#R6u7S_>RetnCae
z3{jvK1KTY=4^#J|{L+H(qQruN)S{{*3Sh26IfSoJl95=VP*9X#mYI?YYTc9+B_?Ob
zmt^Lq7As_zCl;maae?zKvN*V9m6V^KlbV>Pkd~8}4lyM;F%Q&MDo!m?D9KlFOUx-w
z)ln$Uhe(!`=Y!1wH@Os&p^do25`~n^w6xTs)VvY}h^62rS4Lu4szPQ;YF<fZa$=5x
zuZt-t<}yHyHBcb}Phy~u0o8Y)WM2$wAe1oFfZO5C%r#6Yj5Z7<j3rDp49!eg%r%TP
zOnso10JGmMZY1A?XtLa5PfsmLEGa3v#adiikXlqE2&%mKAdZYrNi0b$PA$=723PU7
zSQ3lUi*K=k!=@P2@{5m8&PgmTj*kZy1K^6kNR)wrAqW($AjdH<3NTj5qQ-+Bs^M@C
zfz*RA*h8Swyo8~I5$pjbhzDw!OBidI<C#lX;2vd;XRTolX3%8zD-vR0U<lD<D-r=!
z7hqrTKrKoKC4wRekhn020IAeu0eh)P3?v{9@-r(~nI<DdA-Ht`YRD8xGB7ZNf&2ri
zf*2Tu7}*&47^|eQ`w3zOJJu=;6fj`ZZt+6HEjcwO2Nrf93t^Un?FD(k2V^fed$6)F
z^00s-hich_*TWRlrh(;-GYU`^q@!S3P^kcth9uz;mP??CiKPgfUcf0z29!`hQ3%Zp
zh?E11HBby9vH&QWib2U4oCRc1!kUsCz+A%|&r-sQlmpmMa{xOe2Y}rRO1t1h14-0i
z-(XGsMN$k54457P=}!lxL6o#Fh1D}eCH*34P%wf+3Piwc1Sfcqi(^5SAtiW_*e%AQ
zkYrG42dWA{x<E}sa5ruj0|Nu7)GJ|ZW-8LFVaQ^tVaQ@;VyI=VWvpSyVrgbzWGG?H
zVyj`UVQglqWyz}nt7dOzDs}?1IGUM?LTVUlnNd`Qf#o=xnTqqkEY>WpEbc6}W~QQ9
zm0UFpSv<AO;S70FEDR;Q*<3}FK%*dx3?+Qc3{8xX(U2O3T9!P468;i_8ir=ZW=2Ma
z62T_M63!Zyc%d4m8kTtB8WvD9FYyf{1A~Giq+Jf`nxz&gWTfU4q!uZ_s&0ky%#sX+
z;{4oHg;Y@A7t)>2%uCl(Pyl!JK{XFZD^d+pl2Mdjnx28)cS}?Nwba3ufje?prNt%i
z9y6$)1+DHJi_(j&z{Y{<n=}OtP`#xC)(r0Tfi*+=hniO4GC~(pU8QA$h8a@z()GZW
zfQ$rn7(v>h4N5(@vAOvvsR|m!B}GspAYBQNsGfp<UQVS#QEFjnW)Y-Y5Aq*qcp|A1
zVKBr22m_I-H3bb&1+8hN5CW>u!F8)bW}X753uLWOQd*R!ke`+Y8ZM~JFI6Z{%qxL3
zW<fS2mcU&I@;n%$*U}~VphiPVX)<y(4f15Mf(FF%`9<(*TOm0kH91=o8nl`sx41Hk
z!9xXk$*H$kK)P<RfCB0k3&`)cSizmLTTB_Iw^%@>?=9xylA>GerFogj`6;PI;OxL0
zpOJHm6{NNJ76*s|>*oqf%qfM8j}(ErKDSs4iZb&`Zn1z`_TWx1qznLOf7V-!X+?^l
ztO(*3X@crM<_uG$ArNpuc8eQid^{-lQ;YIISyKg+2YI*{`4~AE`4~kQSs3{kr5KqQ
zK_WbiY)m|iEKF>SY>Y*qz8<`02aUVHGcBk`0LoU)jGzWe8dEk~kw^+-FheCz2{S0W
zFfxD!!B`kdShBf_Ody#HBp<<$0;=g*sy-+<gThpyJR=o4YJu8m0gZo^q!uM6<)lJ`
z-7mi+71|^&0kyu06`%pCkeHVO9qIrLaQXTBDR_ps201$UxI&tKsCl8FGCn^k%L=WN
zRFbbyl3H8>(GZjhN?=I&C$l8AC_X<MDNUB9g3?}QNop>ruLzO{IaLSJYJ>`brSkG2
zsnIROPm{F>)Br95jiwZ-fuf5Ar12IfXs8X`<kw`p#RhR7q%8(2C?Sz32u_0WnZ@xS
zEg-w5fT9wVNEo;U7+DyZ7(qiNe2hiL3=9nLNCWjkK=l`><SjOWmbA@`wcu#0VFZ<H
zoeU++;I=no3R4S54WkRVv@6o<U<igrR$>>ZgoDIaW|~4`PL2Y|<Hg`&Ei+FcF%QHC
z$CW~2Q7X7NDh7=|fl@6r_E6&h<OfJBfLI{kK+6_<;SVy+3er8#PXpPQtB{<ZSCW{S
zmzn}|3TW&QW~82iXBuWfpa6CrObw`2u8^CUR|z(^7&KG{Dt18q0&x2u6q7~Z)Bq~P
z@`_7R6H}l*3Q7ex`;$|l=_I41q`*pFzqlkZIU6*(nwFDau9uvjt6x}}S`6x37V8@t
z8<`pz8yiE%7IndKs+*ano0yZMn+oZT7VGAx=_cytWEPj`CKjdYmSm*rf&x<)<bAMr
zP$Cx;v7i_)2G#vF46#DBj3o@9i0xp=<_Ke8WGG}xfs96IGJ;c|Cetk@J%d|}6`G8<
zxY9C<ic8`_CPUI2s0#nZX_KCsms(L!WVaoZ#2`tiN*nB0c<KW&a=|+FY;y9G6LX5~
z^x*p3G+BxaKm`*hx!+>RORXrm#hjRvqsbJ-0Sb9g)JL&o7N-`1t<_{KG6Sgw*?vm`
zmpv;$jsdkQ7<f4tc^Hd8MG#6*fHDiXMWF{z=b+9<7E=miIyj|+dCUktsMxAytO50@
zn5sTQ3nWkiM=Kk^-Nn4p+=5Dl#G<0aN(GJF#NzDK6i~fbl$u_elUSq)Nw}z~9Fmi<
zlr&&3!i&Dl5>P5jR6w!;Vj8p<0!KQi63Z-(hp2&7jZg&&Aw{s_2*V^Dh18tlR9LYD
z$~Fp_#d+$GGU+9#OxI)qr;A%mc?Gvv!T!6&l$)r@R0Jw6qBvldf@2Cofa6I7><LhT
z1a?zASSP402Wc;JX)zUn#s%PQKF~N4r~&E?ipE#q7y?xtCCnu(C9EZECG6?!&5X4y
zB^;pYqsX9yvxWiGVN7GLVeVsOWXRiA!UdMOR>BRUYgubpQy8;3ikeDzAS#=gYuPI5
zN_auCE)219F-*1WwHzgUH5@f;X-p|hHSC#;HO$kQK&BPOg)<aBt6{BS$!05>gKk&h
zGZZszh%|FkI76NQx>=CwF3|y0g1VIE7APd5B`)wd9JtvHE5j1OjT~?z30l;E+lcVS
zWM;7fbULRfHMz8?II}D@rxG-s1S<S;6EpJ^Qc80RG&B`pI-sTynK(h35iPQec!=9H
zGEBkkMm;cB(@FtMgT_ff>J>m^4B$aUgieqf<1<qfG&1u_G_4de^HNePK<c5*%t~m9
zuU@QBl$w)Rl3A9j04hj8B_*h#2l5Q4tN>LG5JTN=u|v)D)8xO!T%1~Riv?6z-(m(e
zkyz7m@)JvLvE=0ErQc#t&IgU|RUoQbaALeA1RYGs$j*Tci-7A(aA9&w07(|?y<42A
zd8N6jMTsS;w^-vr9@XRoXNz0x&~PaNP01F4h908$LGmEi#;1S^)Z!>kP$da2B5!eJ
z#21$)LCj>z%)G@}Pze&qPs)PiC{WP_uIG!q7#J8%g0hGlsFB7k#>~Me#LU6S!3-+%
zgcxPmI2eVPco>D4WLUszZ!wntV$@Ft8357%%4HzefOFXu1_p*oQ16GahN*@bls{4!
zYuK_FQ<!WRQkcOk78?do=?G@C+A!3xWii#TG&9z+*RZFsWpfmT)v#qTLqx#p*s~Z@
zSZx?UgEuwoX-qY&pxji$mc<I{rG+!(C9p7*uw`=<or5&_YS^;aYdK)@Rc4iZB^+6t
zH5{PcR5(MP5(`5KS2kDCK8WfP?kt`Z_8JZ`29P|Y7Z%Qt!dAl8#2CQ<;e+Q9ITF7x
zF)%2E7iE^Df-)3%k~Kx4D7By{wKz4eBoUm5^3y;`43y+RX$jQ+1?MS{U}|y+n2nJ$
z@{_U@G$0cJn&<_0zCv+Heo-nYaV9Ewfb9eq&7kJB71%ha(K=ubs6h{HmV!H5pne-T
z3qe(Z^@56Z5C)k6GPpb=GdTm~WQbK@hl6MB^b{Nsoi|X$gEY+ts-JR-A;Xg(!$3GO
z88n>>N{*1s1L~N-x@!60dEUevg~X!t(%jU%5*-EgB6S4~&;+tVeqK(crjCMoxw?Wz
zIVgZNLAI$Usw-%K24M42G$96pECFH9^t^m%4-euA&>$?B1qnuoYEXVGN-ZwV%+JH=
z3OxnKf`XjP)M5o>)g}4h!lgL1L{GsV?C@eM1<?GojzW4>W`Pb$Fy>UHDQIXb6oVwd
z?Q_U{Zfc4qSQ)nNB}QO?nvDvmW57kJ1&}#=@H8=~iwPd&Ehx$dbs)h;gMtEtA*CB=
z%oh{}#Sq^YB$i~rybWf9#sf0*6f#oMp!0bp`3l9LA!BekT9U7);F6k_SejE(3`z?{
z`S~RZ`e18rv4NeCdW+TB$1yngC8%+Bi#^OWDA?2A4_uG?RVje$8PF_$YKoqMbAD+~
ziUMddt0*-$zbsWjwOErE+_SsI3K}vlzQvrLS^^%@1Q#eppi$@|(C{s!N6D05e2WdD
z^%lE-Fu0p?i=#ZTC@(WFz4#Ujhy^Vbqre?mP~)x$-Yvbw0`f4pge(HJ%Wm;x#Dn|;
zsa9^Wrh&2vS|NCg1yq!N1(kxJUJ`>K6B7?37b6!V4-*#?2bkt#<YDAvWP|p3_!uRa
zisF&`J)k}hs7(oO53T~YHbCtHNWTZ#=3psdD`6>NFX1TREa52OD&a2SDd8^RE#WTV
zE8#BTFX1i`C=o0X$`URSDG|yN%@WHJ2eTwfBuj*{q`;&!n3MsNvS3mUOv-~v1u&@y
zCY8XXGMH2Wld2_>nbNgPHB4D*HH;<d%}lk-;6Xsf#9UBoAsjTZ0jfz9AO#ty%mVeU
z%2O3eiYh^g5HyAXE{5!(!$_cl0NT6&r<RmV$gEaTr9x67Xmtf>IHoKy2Q*ocpQezg
zkPKQpP+XE&0-gxdQ&8|MQGj%JioxBa<ou$d)MQ9w5q(SxG>?|3ke*qVng`CI;2so4
zfeOkKINQNF`T5zpr3DJ*Ag{t&E1(o#TAT`^K*<11XBKCpG>H)-Te+agT`PD~2QtuI
zoB>YsNvR5@AYX%ukkZ@&s4GG9o_VR|U|w-5#BZ7!;JkB-1r%bpgrV&K5C_&wxWxkM
zk=+u9N`W{znc1muY1Wd`f}GSM(9|Ee(Q!)>suD89g{&6b^)CXY%v;h>MUb%%NW%eU
zWD#h(uE-VCF>(SCZXg2G-GjR?FFz0NfFc{vcmjJ~3b-8(cV}J!q+t#>)DEPKB{{z+
zHHzC8+>3Dpb!np5!ObT_Gi06Mb_`4>RE4n-lwoEP#gds<ViCofnO9<H2BwUSz?7Lu
z6l*D1jtxW^nni(WV<QM{W)j5-D&ccdE6hxyxS&i!Ba0%?49zW$y!?1@tZK4A+FFby
zw^%{>p!gOmSRB2y5QPRJsO&6GEr|!Es?q`lMh1prP@phyb1<_oVux&uT#TTRE^L^K
znTN5c092@e0v*)20yVNg1q--w1scUGVE`8@EHx~UE(ol%0vhSkWUBh209os$kO)g`
zuqHmJUWXLqh_*f?fe+<II=Ia7(}aYYCM!7hia>#dXeEOu7SSTG$en?K!5<V|pulC|
z;$Q}gz<Z7$QBV|sa!fI(La1TPVn|`EVFI-iLDPxg8B)e(#&pIS<}7AV>lR#&GFP2K
zL;xtqgWKO&q5xD=V+(y~8((37`##`ctzriy?|9Wra6zTX0;#aUBZ;6vmZDJ5)DI-X
zAO$&S>JdG(Js21m;z6Me9!uc>4FDH`#Ngo#>fVFG929fKpz;DVZURc^;Ql#iHkz@P
zwT3Z^84?Ge8T=YnSl_yqt%fBX)XQK=VYXpN0nL!GR(%6ENk9`K`9%t#qEiRdz|v6w
zmzmf?9%qdKsu8Re5am0_FkMjL3JxStWiv=34dfJfoZVspMIJcjK!d=DSW5zx>7e;Y
zNc^#5%QClE!R7ER=KQ3rTZ}2_(FvNO%mGCwXmEvrgNKoWQHrq$BnppElyVC+Y=+1{
z;K*ZuM;=QGb2?)UOBQ1aiw#2wQw>WNb2EewicHpE22Hj^Nl@bi6c5Fqf#19W1+ep=
zof=SW4eIECdq5P$Y$9|#tT;I_C$UIJ0X!*_SD**2r!)*9O?pr`55_1jDg;%u9C-zh
zN*x;aERgaKlv%;8grXu)w#8mjmVnX*DC!_F4$759*$fN}RiMa*<XtJyI4DRAC8|Ms
zz<C$cY(S3c8pu)%#uTPGOpuYVT3CcLfg>EWGy^>PU&E5c1nTF5d)w@Za!64QZC}zX
z)*($HaAOiYbflvIj&U6Y&`^&;YDx0oj(1HaaLj`v_7)GsGe}9`7C%&|xHJizZ~Qc&
zX#!q~gA)dh+7nxvNMm4NXa}VUNCCjdC<ICsAW?X#0Lj2pMFIl@10q#`%JdW_Xr^bj
zVW@Bb&A>M^xG=<;#4y#eg39wOh8kATm{1C+WCx7^G21YdFxRkwMu1XSQ`lg|2O~qF
zN;pFbdoY70M<O>U%|O~epb-I30SFy41a%xz6O%I(vQsNTV;-Ppc4mHFVo@bU1qPBa
zph5wJ2W7zk3V(2M5XA{{GI$~d(m(=@#)HxWXu$@!Lb$~W8bT}v_scaouvHFG+!^s8
zQ@~A|Tg=(1l|_D_asyk%pUA+#Fcp+gctDL=4p0|b2-KNoECR{E6AP$lfhU$p(5^FR
zs6LB1iv=_Y2%0o(W-7|4Va#GpVFES7(il^iYmjOYunY@ACJUZKYFSfQ!HJDMh0TT`
zg*~0IhBXV!&Qk;H!fIX$2a+DnU<OUDL}^g+LexT#ZV$~WA#kS(G!B+mpqHDd2jN0z
zfd*xHRAddRDd25JP-Rrb1ImN(s+sYi*{fT^(5@nwmzo0Z=0TDrx(rBpQ3eA8gP$fx
z5olN%tPI@YECMwTi)=wD4%CXd#R66fs@^o&uqD?bPyu#}F&9#$fl?V{)H{=bfng~q
zX@cjoIV2dRm}EdBJWNuIMIbqN;sn)#AZ_6K4P+Ro3kF^}WKzQjn%RRUI;I+ibVgXY
z1h3_oLA6|-O$kdDYYAHlYcHf7%LH%7a@Vk?u*2&{wi<>SwiFH<h7?Y4ix#Xqg)5jr
zlP6ISDfxk0v{X-ekZ}WW+g3pXw$dTL2s)xXC=(sLy$h}aZ?QqV08TBMj8QCQIjMQz
zbg9X7i?ui-5wuVhlyX4D9(YUy)O7@<TTOOIG6T1hLFo!y7#1lrFfc^%78j%@XC~%A
z>r_y8P!nsj8IshPQc9v&Qi@Yc&{JG00|UbjM2h2Llwss!;$f6w<X|iU$-z?`D7C?R
zL!e3;QFnqG&!7e~QaS_Gkf7!>s|^FB?Ns#<-1h;GSVI~$CHdqwj&nh4y^>NPX61vY
z1qOAu2b>IwK+$`P1vFL#P6>XROo&D>3C$eP3dU2Q2m?h11BVP_5l93cO`w7SRIh;x
zhAhy;7Y}$NBttDrI76Nc3quKGHdm1bWTFWy!&J)xnNkAvazNc%aHp0fg&9<Y)Uc*A
zrm*xf)v|%>E4D0FP<>Uxmc^dJ4w<3`4L(3tIcahv@_~v2&oqU+d<F0j6=)hhH!&qu
z2ePdV)KP-(bR(mYlLDCx0MBSZR~&%`^ubo8ra<Nx23dOt9PNJa5<!y@oEUDg6@Uk<
z!6~82P_-D6B|x4*Y_-t=kCrKbm(YXvI2ETBWhUliR;8wBaz?SFWai}D;`Yo1?ZSb~
zMzALbfK9!{3T_Y}l2=g_sC<bA5ululqo8O8l`Gs3hoyjLFF+LwdV!I{z`$?~l>R`A
z?HD+iK-Ci?9}^d&2qO;@8)Fej9@OgwwXMJyG-L&06l<VujRY^Y0&S1vKx~gh+VH>~
zQ0WX^)>M?}3QAaxkcKm)E`a2F@LanBe6}zrKQRRyYQ?D~#n3T$&pd_vqLkDkP?tQZ
zG&3hfp%k?Gprj}@6=G6xNl|HXNoi54jzW2=LSAZW3Rr2X0%+C{T(FenE1;MRu{9CC
z?5YSnm<CGSkf}ipO@-Xll8k(4k_-UvpoEO_fCdXvK%+jz3Mu*E>8Qk<ocwa|z9l^c
z@B#<$E(Fl3Oz>6?umcjobt}ZVP>+N6dtlqk1R4s04k>|$mq5)mP}3NaJ>V;aK*bR#
zYCyOc)S7{gZZTysf#$GN7-|^enZYCrY~d6uWHELtC>?>iQn&bE>&}Y6)B8oB-HOod
zf3U^SplSoswgKt*4oWznhBX6t)~ZSbDULDs1%aFfD&@cfo}dwe8V1mSXEsxj5O^;I
zV=x0)gejY;NCh(Z2^C?^W-2m+tiXhduw*k8Wq~?rEetgbprw7aj4A9XY;YbEX!#9j
zp-nSWGuYf<h7^uq22IYYN(E=gb}@ytqWoM1Bo~8{38WjV0GdbuCp>UQhNe7FQ3PF@
zWTgPwl&b^Zn+oa*;)*D6jDz%6@!EjK`xVqdD%I_(_@G?yaG|<g6%UkOoLZu8R|M)$
z!DAOR9|{WkV9?|tC|c4PY8YZAAdBL%KrKXuX2ujINrq;|iA;r{(Mjf_08qSw+8r<d
z|NsAAlNsCr2gm6x*5ZPk%o0teBG5LqTcFjN=uOvOTs9zAf+jKSIG7k1ia{=5W2jO`
z3NO86NI)X3l!NO*aXBc`!7c}FRsg%42edvdg+Y=5>RLu{xe6|upsp(F1UVn%XRs(F
z&2=#_FmNJV1KN+-22utdZ%hHPz!O!Vpu5Ffm01AY4snYmJGHX7r~;$_)kM&W8t~r1
zKcJKh+Hb?a%Ebtpi-lo6Mm|P9rYd19{?^34l?IeCih4nI;n`0B%5?8Rc7k#b2RjQR
z4{H%f45i`+?Xm%9mN}qaPZOw5(98&$P0eBiErqOQsdP&L%?&VRF*h@0vlVNUurx8I
zFx4<;!9=V;A|<TN47IE&%r&g>Y$fcV)nUy{;B{s!k_;&<k_^EN&;`wP&}IZn3TqBW
zEn5v+7HG8^Q!`UALj*%O11P(daDi+pnpVS{#SJP~A{bIY!>8<thd?E|52(C`ZNCIJ
z2f>S&i&B&G^B~1CXju!S_=J>u*jf#sc?<ATW)K6~0YP8l50U~+u|xX}@H_^}XT^zW
zsqiJuE}6xka$g~_v?M<_u_QBDAte=DCxT|0L6gNrNtq=@iA9yL0YHemper0e#(}Vl
zzk;8C2&kL~m+YXD0<w;_G%qK$7&560auTS41TIz*6-tUqi$QxZi}Op1l2ak2HOMFs
z)+kO*Re&z6EiW(EOD@*S$|%)KO)1r{E7s2~Ee5SY&<8s+RWG9?HwV1!DG_w8KtWDo
z9(WxPbZI8YIM7lA>a4>k0&R{f3IQeFKo9|{Rs6sU95i{riz~o$nUH=<4+8_kEk4KM
z;?yG0x+HL$1)PS9KzZ^OYdolqy~UdWZo@!N4gjrVOH8@N23ir9Qw&~{p~-%W9Taq+
zMPo4_BUwQ42ez{)7R2W%sEmhKOhpqw&O*(^MH3kq82CZO0;t|&-~lZe=VD@F6aw#u
z1I>kk7FB>2m~k<J)xrxW(6Ta6>l&QN7BDhE>K{f)22fE7-loHt#*ocaqyQQ`tYymM
zsbNTAPGN3kl4Pi5s$on4)j~EU%r%fYhB1XTo2kgBgayuL%VsKyDPe{4*|V978o+fN
zyCi614_g*HXbD>iM=w(?Qw>uvLk(jJXEsyOBv6YSu8S+1X#!)BNTpg0Ll#F0w>U#B
zOASL7CnyVNae)@G!QI3b&X8xr!cYP_0_6cDGuN;})YP&?GUO?+FqH6Qb4_3<`iDi9
zDT1MdEsGcICbkqFBscMbUEl?FIUm^Ne4q_Ppk*W}{Bu}A%j&YYYM7fDYdLB-;O-Ux
z>q;mQu3-X&C%7JFOc4Z&mX(O0hzfy4TR_W7OGH8btO0G3V#pF}W@u(+WB`Q?NKR9@
z>Wl)klmQL<6sMNxfSPTf?cbmb4c`BqSpr&W1zzC+%haHH7h6%25uXR%>jR$Kff}f3
z1zwQ`uKq#mr62_YD9eKul0i&?H3Q(~4oI(+0#ZSjngX*Msvle|fU-m?q)7|b3hpp~
z>pnlyS_h?|2^L21a6EVntEdXp+GhoidTBBiRfBR#8HgwcDP)DWEi&^;kXjz#DF}5t
zgw^?Z=}27A)P%ZSl>oF|0%{E8q*kcg71e@F28~e^HGqs4hZG2Ppxnw@2HNUVgjOTl
zfHZ)|(?E@vTWkf1pq<6w!sHg0Cuj*Rr1${stc47|fwCmlr6S<b6R_RdAVWc=kS6CX
zHb_iHu@_WA+OD9X!CUObrAeTukKzOk;KhT=$tX6+LM?C*MR9^;!6VI4tR=~b#pvTH
zw|GD?1|CENt<(dRF%h874=1RH#KFYF#KFYDh=8D-Ml9gw0~dJT5gVfzsQJJs0ER4}
z#a>KIj1b7d1Zil1mVkjU3zG<A(QMGV3{as9vINxR0yUjL?Pl;6b5Q@jg`tz7gaO>w
zVoqUc;efP$L0ex^SW{S97$6&6Kr@<E#}q;mvq1|<L9H^-YAeXV6|~EeUsR%ySpr&L
zotamXnx0yu0FF6u*C#U%R<D9nDeB^l?9|G5u!$NVQqxMoC$qQ&WIU*mg(G{w^u!l~
z&zMlard3A)DwzbDJ^?!kY&*QP289bKJitv!P+JYu#Q_hFf;W<<Fo8z!!TYVNm}643
z!P!ca1+pTEsVMapTXAY)QE~=sJE<us`JmP$w>WHa5_6MM675PsQ4NZODru0b;z52x
zZs$T&!JP(+D$pj@V$f!8a4T1!mazl8X{ClSjS14yg*0YCw!+%Em7pBNc#EwhKRz=L
zn*KneK1HC$ZW+iKAj8=hs${V`B(t~#lrQZtoB|FdJPmG8&j9Qim_vLq!vpMwa)cW|
z`?x@^NKFB4d4WZ06gMbz;*%=j9?)bgnh!D!<d0jNARXWU%mJAMYVa~}axjW97A*v&
zc98Et0S!7>1r)oWj9<I~JU;{K*km!+GL<l7F=nx3vDPw|u+=b@fVRkLfLJw5plz=u
z9L=B_kui%i1zL00u*7rKFlKSru*7rMuw-$iFlRFt#g*{XFlO;KGc+^SvgUC>*tM(`
zay6_j46$OhY&DEEY?2JM>?M3PEH&()ExCm`;S4n_S^UimwH!4JSptv+-g!bLg5c&i
zCs>RVw5YoVv=@mbQ4xD17?vrZyBa{N`tral7(tav4yerxUNH-*LtsM@NZkz$XiE<@
z7a})`6f}^OYg#El2Poi;Cb%_l8^9S0)O19eN`#EYfQJ{LCo4cJLP!OsgCVa2GPk%O
zGbI&rgbc_$7*5PffsC<$cGZEbg^b{Uq!D!<xR0wT1Ren|QU@g%&~C3=tl*O{qQJ8}
z#i=Dx>@E=ZKt|7HK<YqUB}k<MYHNZ!wcylL1gZ{-`aw!SJ+NDBkO3TUxvI%kG!-NP
z8cL1g2G1-)M%$x!p$>#{`60U);K2u~GC(ODM1%SqMKeKlEk|)_61ZZB0@X4g6TrC%
zLVyxJ*cnxzBoA78%)rgX$i*ze$i^tZ2rBydm_Q^4BOen7BM%D)BNsDRy(UwUEcmPl
zNRL>PxyT>H1$X8(p<)zLOkl%8H1!~d7lngsZSZC%@Z8re7Eogd5+|U3=PfSy5N~l2
zD6@h$+ZKT;rJ^QKc!CBaZ*j-Rm!%eg*1<!L%1SIP$;dB?kB4mT1x<fJ)?S0oP$^mj
z(gy0>7J;_xf;SFBIv1cFf8aeUMW8)a;Q3+j`sX6hdPVT+xgyY#F7PrS@S>3-(0Tyy
zJTG`Y4LrpG9zX|=TS8hP;IsnjB7jo?B$<Fy2&fc>L>HI^inCiBHjqQU>_Ej)F=z;d
z1y;i|fyPD!Km%P&pe?B^>;hZ@Tmswzyh4IPf?R4GA{;^-{9HO*+8p{^0$jXYd|)iV
UCCnwn#ls~jpui>1VaLG-04tAcY5)KL
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/lookup.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/__pycache__/lookup.cpython-36.pyc
deleted file mode 100644
index 6cd877598197f890a4cef46feab4f938a2529c61..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 6661
zcmXr!<>i`^S{*AT%fRrM0SQPmFfceUFfbH<U}RuOVTfW#VN7AlVT@u-VN7ApVajEW
zVrFDWVM$?aVTfW$VM}3eVTfYQWQbzRWQbx<;Yi_3;Y#67XNlrS<xJ&D<%W_hsVu21
zDLivHqj*x8Q+ZSQQu$K_Qkjw2DZF!dq6AYJvxJH|Qkhc)Q}}EcQiW3mQ}}Hdz@jTs
z1X2WZL~_}pL>U=U1X6@@L~_}q#26X!B)~G_NHXFOnN+qciB!>MMn;BI_AH51F%S!^
zRx+0@N(yR<WG;J@G?*=%Bbh4`CBw)7)+Ys04b~?OX33;8M#-j#q=>dKM9HNnriiCV
zv@k}=r%0wqwJ<~}q)4a8v@k>|rpTtqwJ<~}RjQ;2q{!ze=c+`h<f=xgGBTtpq^f{T
zQc6_?(cuhvS}Y8yYS~;97>mjn89*?SAy15jAypkL)&&w_WQb(QV~Sv?1o=`SM+0OY
z!mMzJdWadju&CF>qy81L`V^*M22G{J4V(-NT*^85`ProfdIgnSTp<~m#R_SeIjIWC
z`FSOYnR&$ui8(n6B^jv-xrqe@nR)5O3Q4IY<*BK83K^Nn**U5D9xiUC3c0Bzi7AOC
zi3*8%DGCLZB^mj73MG{Vsl|F+AsMMfsS1fjsS0WNr9}!UnaL%Y`FV*&nW@EK)%kg;
z3dN}<3MCndC17O*`30ppi6yBi3Lu?UTwI2l5VPVzdbz;Hf|*t@H^H5RY%|z!kefh;
zCl{qAf~-|YE6oGjU#z2$s+X?ErJ$gIWVnK@LN!<jB%G63Tw(=cfH2sA)Oe6sPG)v$
zJk($vxa!P2gklA#YKU{<Q$S9})R9-9mzR=QRFqh0r2x~BR}c>tfvJWYU(3a1q^XdR
zof8i-tQhQnE>tdNa6n=RHFBU1SIEgvOv%hk#~L80=As0Ok`2gy1$9tBsM{$iC|H3_
z0tXDt(TG5SsV*rk$Vmk$hpGniVak#HXsiiKc&SCmK82=66hVZ4i&6_qGmBDF6iV_z
z+RIXlK*=dpA)wM3l8RG{6!Md@Qj<#*GGXZ!l-xlPi;0sGld}~v^Gfo;$sTTYUSdgR
zSt>ZO=%JXLS*!q8lvoV+4XVM#`MIeInI)-3iAgySLyEzPJu^4Av;-`sk&>^FmtUfg
zkyw_hkeH{CSWu9fm!hEwk0DJ+YDF~{#Pa}UuuM>HQb?^x%q_@CwF0{WTcF{KOe+OY
ztm2M6D+O@8fz1ZlTg$~|f*fDjsg>B{3{?UVZ^Z?v$(d=H$%#2Rl?rM3MX<;NrAtWK
zO#zn|3ZO)*0FM{A13->I#ff0`;PRzXp**uB1009NiMgo?*{PM#$P3Q^6>yNuiL5?5
zwGvnKfow@m%u`4LMNOqbMq+UWIMN`gN(Ypzz~x|~LTO%Ra(+sxesM_=$QwwB1CO&o
zeoqAZ5>!~_CFZ8;DY#;fO4P8#8lA-@Me24+Rtm)>MVOHZbsty(lnr-SEf<$5tgOw;
zFNrV7k1x(IDghU{@kjy+pu`4FEDHH)(0TwIwhD<QB}JJ@r6sAwkn%h?u@V$!8JVd?
ziABj7ps3d2N-8Z;NXgGrFM;J71&}SE&`PQVWxUk%)FOq<yp+@mJuWU~Wd+xYlGHqK
zVFgJPplqI)lA=&tT2PQ*R02*p`6U^tMGDC|iN(dxdIeNY6@zNDRE2_~{N&W)Vwk?X
z)N*j4rJquoTcD7fSdf^MnUh%ps*pe)D^^I$&&kOzFNRb-&Y-FzRUr{tEhgsVq^5wi
z$4BR-mKT?lrlrN|6@VNL7ELWG)&~{Uo_XM;2P$UK6oRT(9fg$Cw9LFzg-TGJ2(2za
zsT>slpgI(+4<>-9v_Mgnn3JEE4oTdgx*1;fKr3iOip~U8zs8yhjwvai0wpsS99p2<
z2Ci9kz%?YSwuQPD)Dl230b)^UK}upts)nWlNOv)$cu@u$4032{5h%=x6c7Oda;HK`
zQ6)qrDEFil<>xBsK~;lkM6g6FK-~t8GjQFekXWpcSBzYofD8d)q6|W-!61gA3WA*o
zG7w>@9@LvC7Hh&4qUuLdj4FZ9pr@y&paBYftR-=3NwPv_8Y~Z_K(wV+B&QaXD0qV7
z%C)E{zX%b8MTwcksa!8X?RP&-##=l&;QAydKD8_{Cz%l>%)r0^qL~;N7@Qdx7>YR<
z7#M08QWzx}f*CZKsuUE0QcFsU@)XiCi;7DWQgc#sQ}aqdc{mZ0n~OlYS2Aca-eRmM
zVrF1q029A*^$YY1it<4fsD5%*R&lb4k$zHUUb=p6j&5;kS!z*co<1aN>gQx8=|f6Z
zV?8td;>?m%-GapA?8NlcVyF^bV?7f+W8H#EV?8sy)bw<yus*yMdW)wB?8<nsD_ubT
z0C|Xksfdk%fdR#%Am@TT%EAEZzXUUAGWvm?tjQR~mJy$kpOOmZLI_p{28LTapd<(?
z%rY|b(mmjYGZwLf*kE%qlS_)ZKvoqO#pfoL6lGRGxyi`fB)F_5XE84W0|QFSsF)8S
z0<JaH?TUF2LdB^i>UIz{;4%x=pep7EseuZ?^nneBia|7z<T@S{i$Qe*#9~moS5QyO
z$xj5i1#AgKC@DWb2i3Wmc_1YMAh$u3B<F*g85J-;Km@V67i=H0uV5B}0|@FXsEuGZ
zK!qSy3WBVJiXiMnHV$SflH;J&nz|iIwB+ZdtJ}Q<r5H`dTi_}U(u&q(yv2u>ff;X!
zBxj^1XUAt2$LE#i7F0qS<?y5jN+qD&0Zw{4prprG%UGdQ!cfB4%;>@pD;C34%T&u;
z!(7AA%vj4<!&t-@z);8+&QQY$%7=b#noPGiQ}ar5Q;QNyQg4Yty8-dwo(06_w7g<X
zmRoEEmGSvWSyB8Thd}xYU`4l>GBcw%U@9T08|0uOa1s}VNTBKI2BmLM5@cXxVH9HI
zU}RxrV=Mv%TSzh}V}Z&K5DhX1>@iTzD`CiDtYN5O%47ufVM~~@7;6}7m?69xmK4Tp
zrXsx(<}8*PuzU?`3R5;`kwXb<7E28)NC&u|$(#rpUQh^1P0uVYNi9-HQ~=e$kOBiz
zW$ST)3t7jabZFfIDwlIB!4VIx1Hi2d4KQER3d{kOT3|j{CrCZ0Hjjr@l?obgR!VA`
zrWL#u4Qd3G7iE^Df}2|)9;o34Hw0Wdz_sEt1f&vT2DnB-F(tD&9$Yqq+WH!x#I0$i
z;0bCHf%^ct@Q#2z++I)$EG{Wh&?qh`(zH?thO{0_@)aO%Ey-5^SpaE+l;kU<q!uR^
zWhSLUf)8#Aw85qTw=)e~t!I{`=7QR6FgNKaq~xa-D`a9cz!2_%B}qt9O3T9=o~6a9
zpnwFcRme<(M5ddcCR-6FsK{l5RI5dxQtOr=sH`Xk6~M6O<t;WySEmS+{WaO5c#tFh
z7CSU#qIgl_H;NY&{_yCIVuyM?N&wyE=(&Th2$~M#!O3S9JX`SbG4e5TFmf?+Fcyh|
zihod90m>Di+6SI1q!<_&Dnx1+To__`Y8guyY8d+%7#RxL!WmM)nLAOAk%0k`)^kg9
zN-_&T<0;UlX=a`Rq_TyC5^9nMy9nHuR?q<T#WbxH9KkiSjzT$TFhn6WF*yU2dce^M
z>LeDWmxAl?l6-g~0ObWV3>v>d^)^z!5}Nt^@=H>StrWl`Kv|{5B??8Qd5|Ov>dzor
zAt0kc{S`=}%2UX#1i2Jd9;61%J}r=XNE^pb6B59hOi^42AH&!nb?9*>jA|G-gw}&1
z4pb5|aC0#hi6KWADB|D|1}gGEC36jf4K#|Xwkd?9=H!5z>!4l-D7$BtfJUtH6*ANF
z@<GE@d6ke12Wx_Z`;ze9je<HT728HcMyX?qMre`+C2**D&}@w`0IVO{M=U|IHbo(|
zA~~ltB~<}js(=!CW*&5$%1@IC+)4pwBu&OBK@5A)!%Yk=+@NOf1qB@_RWq;^i6aLb
zNDLlum?<4pq9amz)l~&gNN9nJCTK_%Lpls-K2-+=6^fS-;pPKRxlreUqZgh5d@_r{
zO;ku=6@zj@W^#r?a(+=!YH>k+9(dFzBPGog+K_~s21<k&7#!}%-Bi$!OG!RBEkMRs
zK&cZHD4I-BJjf0z5`ndcA)ySZYT*Hltn?@-e8EYRor|$Zf`NfSlj#;8B&owR+!BE4
zfQe`_7D<AYhWKe}-C~cAPsvY?kH5uSoLO;;IkC8;2oyWuHZUX=fXY^IQYaDvm9i|L
zY<P<m+#b5cT%1~RO9WfD;}$Ep0d|W8R4v_N&de*h#SU#*-I73PKpked#T*c6bc+Sl
z@r)7%rODKyVn`nj($<gSEkGXni4rWRj0Yu7P-sJ1Sy3Wbv>6wH+ILYxC{o}CR}^nv
z0l4Ca>K4o^Kr^t28x$R&Dg)ex0hid|QUIKy!GVruaB*^GK_xhdi$F>H7B^BS>lUY9
zVs0w9zlG*fP{8LGRHA!3F)t;t7?Kb`83w`vCkYOl-29Z%oK!n|P%3C+U|?Wj;$Y-q
z;$Y!mW?^Ds;$dcDM1>rzsC*cYgPDzygAqjXF!C|+F!C^QF>*2UFoGZ#BNsaxGlu||
g2nQbr2L~S$qW~iZBM+khBMYMdBO4nNSQRo001P04@Bjb+
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/helpers.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/helpers.py
deleted file mode 100755
index 6c3d7f9f..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/helpers.py
+++ /dev/null
@@ -1,113 +0,0 @@
-import re
-import six
-
-def get_type_and_data(h_node):
- """ Helper function to return the py_type and data block for a HDF node """
- py_type = h_node.attrs["type"][0]
- data = h_node[()]
-# if h_node.shape == ():
-# data = h_node.value
-# else:
-# data = h_node[:]
- return py_type, data
-
-def get_type(h_node):
- """ Helper function to return the py_type for a HDF node """
- py_type = h_node.attrs["type"][0]
- return py_type
-
-def sort_keys(key_list):
- """ Take a list of strings and sort it by integer value within string
-
- Args:
- key_list (list): List of keys
-
- Returns:
- key_list_sorted (list): List of keys, sorted by integer
- """
-
- # Py3 h5py returns an irritating KeysView object
- # Py3 also complains about bytes and strings, convert all keys to bytes
- if six.PY3:
- key_list2 = []
- for key in key_list:
- if isinstance(key, str):
- key = bytes(key, 'ascii')
- key_list2.append(key)
- key_list = key_list2
-
- # Check which keys contain a number
- numbered_keys = [re.search(b'\d+', key) for key in key_list]
-
- # Sort the keys on number if they have it, or normally if not
- if(len(key_list) and not numbered_keys.count(None)):
- to_int = lambda x: int(re.search(b'\d+', x).group(0))
- return(sorted(key_list, key=to_int))
- else:
- return(sorted(key_list))
-
-
-def check_is_iterable(py_obj):
- """ Check whether a python object is iterable.
-
- Note: this treats unicode and string as NON ITERABLE
-
- Args:
- py_obj: python object to test
-
- Returns:
- iter_ok (bool): True if item is iterable, False is item is not
- """
- if six.PY2:
- string_types = (str, unicode)
- else:
- string_types = (str, bytes, bytearray)
- if isinstance(py_obj, string_types):
- return False
- try:
- iter(py_obj)
- return True
- except TypeError:
- return False
-
-
-def check_is_hashable(py_obj):
- """ Check if a python object is hashable
-
- Note: this function is currently not used, but is useful for future
- development.
-
- Args:
- py_obj: python object to test
- """
-
- try:
- py_obj.__hash__()
- return True
- except TypeError:
- return False
-
-
-def check_iterable_item_type(iter_obj):
- """ Check if all items within an iterable are the same type.
-
- Args:
- iter_obj: iterable object
-
- Returns:
- iter_type: type of item contained within the iterable. If
- the iterable has many types, a boolean False is returned instead.
-
- References:
- http://stackoverflow.com/questions/13252333/python-check-if-all-elements-of-a-list-are-the-same-type
- """
- iseq = iter(iter_obj)
-
- try:
- first_type = type(next(iseq))
- except StopIteration:
- return False
- except Exception as ex:
- return False
- else:
- return first_type if all((type(x) is first_type) for x in iseq) else False
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle.py
deleted file mode 100755
index 24b38c3e..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle.py
+++ /dev/null
@@ -1,611 +0,0 @@
-# encoding: utf-8
-"""
-# hickle.py
-
-Created by Danny Price 2016-02-03.
-
-Hickle is a HDF5 based clone of Pickle. Instead of serializing to a pickle
-file, Hickle dumps to a HDF5 file. It is designed to be as similar to pickle in
-usage as possible, providing a load() and dump() function.
-
-## Notes
-
-Hickle has two main advantages over Pickle:
-1) LARGE PICKLE HANDLING. Unpickling a large pickle is slow, as the Unpickler
-reads the entire pickle thing and loads it into memory. In comparison, HDF5
-files are designed for large datasets. Things are only loaded when accessed.
-
-2) CROSS PLATFORM SUPPORT. Attempting to unpickle a pickle pickled on Windows
-on Linux and vice versa is likely to fail with errors like "Insecure string
-pickle". HDF5 files will load fine, as long as both machines have
-h5py installed.
-
-"""
-
-from __future__ import absolute_import, division, print_function
-import sys
-import os
-from pkg_resources import get_distribution, DistributionNotFound
-from ast import literal_eval
-
-import numpy as np
-import h5py as h5
-
-
-from .helpers import get_type, sort_keys, check_is_iterable, check_iterable_item_type
-from .lookup import types_dict, hkl_types_dict, types_not_to_sort, \
- container_types_dict, container_key_types_dict
-from .lookup import check_is_ndarray_like
-
-
-try:
- from exceptions import Exception
- from types import NoneType
-except ImportError:
- pass # above imports will fail in python3
-
-from six import PY2, PY3, string_types, integer_types
-import io
-
-# Make several aliases for Python2/Python3 compatibility
-if PY3:
- file = io.TextIOWrapper
-
-# Import a default 'pickler'
-# Not the nicest import code, but should work on Py2/Py3
-try:
- import dill as pickle
-except ImportError:
- try:
- import cPickle as pickle
- except ImportError:
- import pickle
-
-import warnings
-
-try:
- __version__ = get_distribution('hickle').version
-except DistributionNotFound:
- __version__ = '0.0.0 - please install via pip/setup.py'
-
-##################
-# Error handling #
-##################
-
-class FileError(Exception):
- """ An exception raised if the file is fishy """
- def __init__(self):
- return
-
- def __str__(self):
- return ("Cannot open file. Please pass either a filename "
- "string, a file object, or a h5py.File")
-
-
-class ClosedFileError(Exception):
- """ An exception raised if the file is fishy """
- def __init__(self):
- return
-
- def __str__(self):
- return ("HDF5 file has been closed. Please pass either "
- "a filename string, a file object, or an open h5py.File")
-
-
-class NoMatchError(Exception):
- """ An exception raised if the object type is not understood (or
- supported)"""
- def __init__(self):
- return
-
- def __str__(self):
- return ("Error: this type of python object cannot be converted into a "
- "hickle.")
-
-
-class ToDoError(Exception):
- """ An exception raised for non-implemented functionality"""
- def __init__(self):
- return
-
- def __str__(self):
- return "Error: this functionality hasn't been implemented yet."
-
-
-class SerializedWarning(UserWarning):
- """ An object type was not understood
-
- The data will be serialized using pickle.
- """
- pass
-
-
-######################
-# H5PY file wrappers #
-######################
-
-class H5GroupWrapper(h5.Group):
- """ Group wrapper that provides a track_times kwarg.
-
- track_times is a boolean flag that can be set to False, so that two
- files created at different times will have identical MD5 hashes.
- """
- def create_dataset(self, *args, **kwargs):
- kwargs['track_times'] = getattr(self, 'track_times', True)
- return super(H5GroupWrapper, self).create_dataset(*args, **kwargs)
-
- def create_group(self, *args, **kwargs):
- group = super(H5GroupWrapper, self).create_group(*args, **kwargs)
- group.__class__ = H5GroupWrapper
- group.track_times = getattr(self, 'track_times', True)
- return group
-
-
-class H5FileWrapper(h5.File):
- """ Wrapper for h5py File that provides a track_times kwarg.
-
- track_times is a boolean flag that can be set to False, so that two
- files created at different times will have identical MD5 hashes.
- """
- def create_dataset(self, *args, **kwargs):
- kwargs['track_times'] = getattr(self, 'track_times', True)
- return super(H5FileWrapper, self).create_dataset(*args, **kwargs)
-
- def create_group(self, *args, **kwargs):
- group = super(H5FileWrapper, self).create_group(*args, **kwargs)
- group.__class__ = H5GroupWrapper
- group.track_times = getattr(self, 'track_times', True)
- return group
-
-
-def file_opener(f, mode='r', track_times=True):
- """ A file opener helper function with some error handling. This can open
- files through a file object, a h5py file, or just the filename.
-
- Args:
- f (file, h5py.File, or string): File-identifier, e.g. filename or file object.
- mode (str): File open mode. Only required if opening by filename string.
- track_times (bool): Track time in HDF5; turn off if you want hickling at
- different times to produce identical files (e.g. for MD5 hash check).
-
- """
-
- # Assume that we will have to close the file after dump or load
- close_flag = True
-
- # Were we handed a file object or just a file name string?
- if isinstance(f, (file, io.TextIOWrapper)):
- filename, mode = f.name, f.mode
- f.close()
- h5f = h5.File(filename, mode)
- elif isinstance(f, string_types):
- filename = f
- h5f = h5.File(filename, mode)
- elif isinstance(f, (H5FileWrapper, h5._hl.files.File)):
- try:
- filename = f.filename
- except ValueError:
- raise ClosedFileError
- h5f = f
- # Since this file was already open, do not close the file afterward
- close_flag = False
- else:
- print(f.__class__)
- raise FileError
-
- h5f.__class__ = H5FileWrapper
- h5f.track_times = track_times
- return(h5f, close_flag)
-
-
-###########
-# DUMPERS #
-###########
-
-
-def _dump(py_obj, h_group, call_id=0, **kwargs):
- """ Dump a python object to a group within a HDF5 file.
-
- This function is called recursively by the main dump() function.
-
- Args:
- py_obj: python object to dump.
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
-
- # Get list of dumpable dtypes
- dumpable_dtypes = []
- for lst in [[bool, complex, bytes, float], string_types, integer_types]:
- dumpable_dtypes.extend(lst)
-
- # Firstly, check if item is a numpy array. If so, just dump it.
- if check_is_ndarray_like(py_obj):
- create_hkl_dataset(py_obj, h_group, call_id, **kwargs)
-
- # Next, check if item is a dict
- elif isinstance(py_obj, dict):
- create_hkl_dataset(py_obj, h_group, call_id, **kwargs)
-
- # If not, check if item is iterable
- elif check_is_iterable(py_obj):
- item_type = check_iterable_item_type(py_obj)
-
- # item_type == False implies multiple types. Create a dataset
- if item_type is False:
- h_subgroup = create_hkl_group(py_obj, h_group, call_id)
- for ii, py_subobj in enumerate(py_obj):
- _dump(py_subobj, h_subgroup, call_id=ii, **kwargs)
-
- # otherwise, subitems have same type. Check if subtype is an iterable
- # (e.g. list of lists), or not (e.g. list of ints, which should be treated
- # as a single dataset).
- else:
- if item_type in dumpable_dtypes:
- create_hkl_dataset(py_obj, h_group, call_id, **kwargs)
- else:
- h_subgroup = create_hkl_group(py_obj, h_group, call_id)
- for ii, py_subobj in enumerate(py_obj):
- _dump(py_subobj, h_subgroup, call_id=ii, **kwargs)
-
- # item is not iterable, so create a dataset for it
- else:
- create_hkl_dataset(py_obj, h_group, call_id, **kwargs)
-
-
-def dump(py_obj, file_obj, mode='w', track_times=True, path='/', **kwargs):
- """ Write a pickled representation of obj to the open file object file.
-
- Args:
- obj (object): python object o store in a Hickle
- file: file object, filename string, or h5py.File object
- file in which to store the object. A h5py.File or a filename is also
- acceptable.
- mode (str): optional argument, 'r' (read only), 'w' (write) or 'a' (append).
- Ignored if file is a file object.
- compression (str): optional argument. Applies compression to dataset. Options: None, gzip,
- lzf (+ szip, if installed)
- track_times (bool): optional argument. If set to False, repeated hickling will produce
- identical files.
- path (str): path within hdf5 file to save data to. Defaults to root /
- """
-
- # Make sure that file is not closed unless modified
- # This is to avoid trying to close a file that was never opened
- close_flag = False
-
- try:
- # Open the file
- h5f, close_flag = file_opener(file_obj, mode, track_times)
- h5f.attrs["CLASS"] = b'hickle'
- h5f.attrs["VERSION"] = get_distribution('hickle').version
- h5f.attrs["type"] = [b'hickle']
- # Log which version of python was used to generate the hickle file
- pv = sys.version_info
- py_ver = "%i.%i.%i" % (pv[0], pv[1], pv[2])
- h5f.attrs["PYTHON_VERSION"] = py_ver
-
- h_root_group = h5f.get(path)
-
- if h_root_group is None:
- h_root_group = h5f.create_group(path)
- h_root_group.attrs["type"] = [b'hickle']
-
- _dump(py_obj, h_root_group, **kwargs)
- except NoMatchError:
- fname = h5f.filename
- h5f.close()
- try:
- os.remove(fname)
- except OSError:
- warnings.warn("Dump failed. Could not remove %s" % fname)
- finally:
- raise NoMatchError
- finally:
- # Close the file if requested.
- # Closing a file twice will not cause any problems
- if close_flag:
- h5f.close()
-
-
-def create_dataset_lookup(py_obj):
- """ What type of object are we trying to pickle? This is a python
- dictionary based equivalent of a case statement. It returns the correct
- helper function for a given data type.
-
- Args:
- py_obj: python object to look-up what function to use to dump to disk
-
- Returns:
- match: function that should be used to dump data to a new dataset
- """
- t = type(py_obj)
- types_lookup = {dict: create_dict_dataset}
- types_lookup.update(types_dict)
-
- match = types_lookup.get(t, no_match)
-
- return match
-
-
-
-def create_hkl_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Create a dataset within the hickle HDF5 file
-
- Args:
- py_obj: python object to dump.
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
-
- """
- #lookup dataset creator type based on python object type
- create_dataset = create_dataset_lookup(py_obj)
-
- # do the creation
- create_dataset(py_obj, h_group, call_id, **kwargs)
-
-
-def create_hkl_group(py_obj, h_group, call_id=0):
- """ Create a new group within the hickle file
-
- Args:
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
-
- """
- h_subgroup = h_group.create_group('data_%i' % call_id)
- h_subgroup.attrs['type'] = [str(type(py_obj)).encode('ascii', 'ignore')]
- return h_subgroup
-
-
-def create_dict_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Creates a data group for each key in dictionary
-
- Notes:
- This is a very important function which uses the recursive _dump
- method to build up hierarchical data models stored in the HDF5 file.
- As this is critical to functioning, it is kept in the main hickle.py
- file instead of in the loaders/ directory.
-
- Args:
- py_obj: python object to dump; should be dictionary
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- h_dictgroup = h_group.create_group('data_%i' % call_id)
- h_dictgroup.attrs['type'] = [str(type(py_obj)).encode('ascii', 'ignore')]
-
- for key, py_subobj in py_obj.items():
- if isinstance(key, string_types):
- h_subgroup = h_dictgroup.create_group("%r" % (key))
- else:
- h_subgroup = h_dictgroup.create_group(str(key))
- h_subgroup.attrs["type"] = [b'dict_item']
-
- h_subgroup.attrs["key_type"] = [str(type(key)).encode('ascii', 'ignore')]
-
- _dump(py_subobj, h_subgroup, call_id=0, **kwargs)
-
-
-def no_match(py_obj, h_group, call_id=0, **kwargs):
- """ If no match is made, raise an exception
-
- Args:
- py_obj: python object to dump; default if item is not matched.
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- pickled_obj = pickle.dumps(py_obj)
- d = h_group.create_dataset('data_%i' % call_id, data=[pickled_obj])
- d.attrs["type"] = [b'pickle']
-
- warnings.warn("%s type not understood, data have been serialized" % type(py_obj),
- SerializedWarning)
-
-
-
-#############
-## LOADERS ##
-#############
-
-class PyContainer(list):
- """ A group-like object into which to load datasets.
-
- In order to build up a tree-like structure, we need to be able
- to load datasets into a container with an append() method.
- Python tuples and sets do not allow this. This class provides
- a list-like object that be converted into a list, tuple, set or dict.
- """
- def __init__(self):
- super(PyContainer, self).__init__()
- self.container_type = None
- self.name = None
- self.key_type = None
-
- def convert(self):
- """ Convert from PyContainer to python core data type.
-
- Returns: self, either as a list, tuple, set or dict
- (or other type specified in lookup.py)
- """
-
- if self.container_type in container_types_dict.keys():
- convert_fn = container_types_dict[self.container_type]
- return convert_fn(self)
- if self.container_type == str(dict).encode('ascii', 'ignore'):
- keys = []
- for item in self:
- key = item.name.split('/')[-1]
- key_type = item.key_type[0]
- if key_type in container_key_types_dict.keys():
- to_type_fn = container_key_types_dict[key_type]
- key = to_type_fn(key)
- keys.append(key)
-
- items = [item[0] for item in self]
- return dict(zip(keys, items))
- else:
- return self
-
-def no_match_load(key):
- """ If no match is made when loading, need to raise an exception
- """
- raise RuntimeError("Cannot load %s data type" % key)
- #pass
-
-def load_dataset_lookup(key):
- """ What type of object are we trying to unpickle? This is a python
- dictionary based equivalent of a case statement. It returns the type
- a given 'type' keyword in the hickle file.
-
- Args:
- py_obj: python object to look-up what function to use to dump to disk
-
- Returns:
- match: function that should be used to dump data to a new dataset
- """
-
- match = hkl_types_dict.get(key, no_match_load)
-
- return match
-
-def load(fileobj, path='/', safe=True):
- """ Load a hickle file and reconstruct a python object
-
- Args:
- fileobj: file object, h5py.File, or filename string
- safe (bool): Disable automatic depickling of arbitrary python objects.
- DO NOT set this to False unless the file is from a trusted source.
- (see http://www.cs.jhu.edu/~s/musings/pickle.html for an explanation)
-
- path (str): path within hdf5 file to save data to. Defaults to root /
- """
-
- # Make sure that the file is not closed unless modified
- # This is to avoid trying to close a file that was never opened
- close_flag = False
-
- try:
- h5f, close_flag = file_opener(fileobj)
- h_root_group = h5f.get(path)
- try:
- assert 'CLASS' in h5f.attrs.keys()
- assert 'VERSION' in h5f.attrs.keys()
- VER = h5f.attrs['VERSION']
- try:
- VER_MAJOR = int(VER)
- except ValueError:
- VER_MAJOR = int(VER[0])
- if VER_MAJOR == 1:
- if PY2:
- warnings.warn("Hickle file versioned as V1, attempting legacy loading...")
- from . import hickle_legacy
- return hickle_legacy.load(fileobj, safe)
- else:
- raise RuntimeError("Cannot open file. This file was likely"
- " created with Python 2 and an old hickle version.")
- elif VER_MAJOR == 2:
- if PY2:
- warnings.warn("Hickle file appears to be old version (v2), attempting "
- "legacy loading...")
- from . import hickle_legacy2
- return hickle_legacy2.load(fileobj, path=path, safe=safe)
- else:
- raise RuntimeError("Cannot open file. This file was likely"
- " created with Python 2 and an old hickle version.")
- # There is an unfortunate period of time where hickle 2.1.0 claims VERSION = int(3)
- # For backward compatibility we really need to catch this.
- # Actual hickle v3 files are versioned as A.B.C (e.g. 3.1.0)
- elif VER_MAJOR == 3 and VER == VER_MAJOR:
- if PY2:
- warnings.warn("Hickle file appears to be old version (v2.1.0), attempting "
- "legacy loading...")
- from . import hickle_legacy2
- return hickle_legacy2.load(fileobj, path=path, safe=safe)
- else:
- raise RuntimeError("Cannot open file. This file was likely"
- " created with Python 2 and an old hickle version.")
- elif VER_MAJOR >= 3:
- py_container = PyContainer()
- py_container.container_type = 'hickle'
- py_container = _load(py_container, h_root_group)
- return py_container[0][0]
-
- except AssertionError:
- if PY2:
- warnings.warn("Hickle file is not versioned, attempting legacy loading...")
- from . import hickle_legacy
- return hickle_legacy.load(fileobj, safe)
- else:
- raise RuntimeError("Cannot open file. This file was likely"
- " created with Python 2 and an old hickle version.")
- finally:
- # Close the file if requested.
- # Closing a file twice will not cause any problems
- if close_flag:
- h5f.close()
-
-def load_dataset(h_node):
- """ Load a dataset, converting into its correct python type
-
- Args:
- h_node (h5py dataset): h5py dataset object to read
-
- Returns:
- data: reconstructed python object from loaded data
- """
- py_type = get_type(h_node)
-
- try:
- load_fn = load_dataset_lookup(py_type)
- return load_fn(h_node)
- except:
- raise
- #raise RuntimeError("Hickle type %s not understood." % py_type)
-
-def _load(py_container, h_group):
- """ Load a hickle file
-
- Recursive funnction to load hdf5 data into a PyContainer()
-
- Args:
- py_container (PyContainer): Python container to load data into
- h_group (h5 group or dataset): h5py object, group or dataset, to spider
- and load all datasets.
- """
-
- group_dtype = h5._hl.group.Group
- dataset_dtype = h5._hl.dataset.Dataset
-
- #either a file, group, or dataset
- if isinstance(h_group, (H5FileWrapper, group_dtype)):
-
- py_subcontainer = PyContainer()
- try:
- py_subcontainer.container_type = bytes(h_group.attrs['type'][0])
- except KeyError:
- raise
- #py_subcontainer.container_type = ''
- py_subcontainer.name = h_group.name
-
- if py_subcontainer.container_type == b'dict_item':
- py_subcontainer.key_type = h_group.attrs['key_type']
-
- if py_subcontainer.container_type not in types_not_to_sort:
- h_keys = sort_keys(h_group.keys())
- else:
- h_keys = h_group.keys()
-
- for h_name in h_keys:
- h_node = h_group[h_name]
- py_subcontainer = _load(py_subcontainer, h_node)
-
- sub_data = py_subcontainer.convert()
- py_container.append(sub_data)
-
- else:
- # must be a dataset
- subdata = load_dataset(h_group)
- py_container.append(subdata)
-
- return py_container
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle_legacy.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle_legacy.py
deleted file mode 100755
index 61a171fd..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle_legacy.py
+++ /dev/null
@@ -1,535 +0,0 @@
-# encoding: utf-8
-"""
-# hickle_legacy.py
-
-Created by Danny Price 2012-05-28.
-
-Hickle is a HDF5 based clone of Pickle. Instead of serializing to a
-pickle file, Hickle dumps to a HDF5 file. It is designed to be as similar
-to pickle in usage as possible.
-
-## Notes
-
-This is a legacy handler, for hickle v1 files.
-If V2 reading fails, this will be called as a fail-over.
-
-"""
-
-import os
-import sys
-import numpy as np
-import h5py as h5
-
-if sys.version_info.major == 3:
- NoneType = type(None)
-else:
- from types import NoneType
-
-__version__ = "1.3.0"
-__author__ = "Danny Price"
-
-####################
-## Error handling ##
-####################
-
-
-class FileError(Exception):
- """ An exception raised if the file is fishy"""
-
- def __init__(self):
- return
-
- def __str__(self):
- print("Error: cannot open file. Please pass either a filename string, a file object, "
- "or a h5py.File")
-
-
-class NoMatchError(Exception):
- """ An exception raised if the object type is not understood (or supported)"""
-
- def __init__(self):
- return
-
- def __str__(self):
- print("Error: this type of python object cannot be converted into a hickle.")
-
-
-class ToDoError(Exception):
- """ An exception raised for non-implemented functionality"""
-
- def __init__(self):
- return
-
- def __str__(self):
- print("Error: this functionality hasn't been implemented yet.")
-
-
-class H5GroupWrapper(h5.Group):
- def create_dataset(self, *args, **kwargs):
- kwargs['track_times'] = getattr(self, 'track_times', True)
- return super(H5GroupWrapper, self).create_dataset(*args, **kwargs)
-
- def create_group(self, *args, **kwargs):
- group = super(H5GroupWrapper, self).create_group(*args, **kwargs)
- group.__class__ = H5GroupWrapper
- group.track_times = getattr(self, 'track_times', True)
- return group
-
-
-class H5FileWrapper(h5.File):
- def create_dataset(self, *args, **kwargs):
- kwargs['track_times'] = getattr(self, 'track_times', True)
- return super(H5FileWrapper, self).create_dataset(*args, **kwargs)
-
- def create_group(self, *args, **kwargs):
- group = super(H5FileWrapper, self).create_group(*args, **kwargs)
- group.__class__ = H5GroupWrapper
- group.track_times = getattr(self, 'track_times', True)
- return group
-
-
-def file_opener(f, mode='r', track_times=True):
- """ A file opener helper function with some error handling.
-
- This can open files through a file object, a h5py file, or just the filename.
- """
- # Were we handed a file object or just a file name string?
- if isinstance(f, file):
- filename, mode = f.name, f.mode
- f.close()
- h5f = h5.File(filename, mode)
-
- elif isinstance(f, h5._hl.files.File):
- h5f = f
- elif isinstance(f, str):
- filename = f
- h5f = h5.File(filename, mode)
- else:
- raise FileError
-
- h5f.__class__ = H5FileWrapper
- h5f.track_times = track_times
- return h5f
-
-
-#############
-## dumpers ##
-#############
-
-def dump_ndarray(obj, h5f, **kwargs):
- """ dumps an ndarray object to h5py file"""
- h5f.create_dataset('data', data=obj, **kwargs)
- h5f.create_dataset('type', data=['ndarray'])
-
-
-def dump_np_dtype(obj, h5f, **kwargs):
- """ dumps an np dtype object to h5py file"""
- h5f.create_dataset('data', data=obj)
- h5f.create_dataset('type', data=['np_dtype'])
-
-
-def dump_np_dtype_dict(obj, h5f, **kwargs):
- """ dumps an np dtype object within a group"""
- h5f.create_dataset('data', data=obj)
- h5f.create_dataset('_data', data=['np_dtype'])
-
-
-def dump_masked(obj, h5f, **kwargs):
- """ dumps an ndarray object to h5py file"""
- h5f.create_dataset('data', data=obj, **kwargs)
- h5f.create_dataset('mask', data=obj.mask, **kwargs)
- h5f.create_dataset('type', data=['masked'])
-
-
-def dump_list(obj, h5f, **kwargs):
- """ dumps a list object to h5py file"""
-
- # Check if there are any numpy arrays in the list
- contains_numpy = any(isinstance(el, np.ndarray) for el in obj)
-
- if contains_numpy:
- _dump_list_np(obj, h5f, **kwargs)
- else:
- h5f.create_dataset('data', data=obj, **kwargs)
- h5f.create_dataset('type', data=['list'])
-
-
-def _dump_list_np(obj, h5f, **kwargs):
- """ Dump a list of numpy objects to file """
-
- np_group = h5f.create_group('data')
- h5f.create_dataset('type', data=['np_list'])
-
- ii = 0
- for np_item in obj:
- np_group.create_dataset("%s" % ii, data=np_item, **kwargs)
- ii += 1
-
-
-def dump_tuple(obj, h5f, **kwargs):
- """ dumps a list object to h5py file"""
-
- # Check if there are any numpy arrays in the list
- contains_numpy = any(isinstance(el, np.ndarray) for el in obj)
-
- if contains_numpy:
- _dump_tuple_np(obj, h5f, **kwargs)
- else:
- h5f.create_dataset('data', data=obj, **kwargs)
- h5f.create_dataset('type', data=['tuple'])
-
-
-def _dump_tuple_np(obj, h5f, **kwargs):
- """ Dump a tuple of numpy objects to file """
-
- np_group = h5f.create_group('data')
- h5f.create_dataset('type', data=['np_tuple'])
-
- ii = 0
- for np_item in obj:
- np_group.create_dataset("%s" % ii, data=np_item, **kwargs)
- ii += 1
-
-
-def dump_set(obj, h5f, **kwargs):
- """ dumps a set object to h5py file"""
- obj = list(obj)
- h5f.create_dataset('data', data=obj, **kwargs)
- h5f.create_dataset('type', data=['set'])
-
-
-def dump_string(obj, h5f, **kwargs):
- """ dumps a list object to h5py file"""
- h5f.create_dataset('data', data=[obj], **kwargs)
- h5f.create_dataset('type', data=['string'])
-
-
-def dump_none(obj, h5f, **kwargs):
- """ Dump None type to file """
- h5f.create_dataset('data', data=[0], **kwargs)
- h5f.create_dataset('type', data=['none'])
-
-
-def dump_unicode(obj, h5f, **kwargs):
- """ dumps a list object to h5py file"""
- dt = h5.special_dtype(vlen=unicode)
- ll = len(obj)
- dset = h5f.create_dataset('data', shape=(ll, ), dtype=dt, **kwargs)
- dset[:ll] = obj
- h5f.create_dataset('type', data=['unicode'])
-
-
-def _dump_dict(dd, hgroup, **kwargs):
- for key in dd:
- if type(dd[key]) in (str, int, float, unicode, bool):
- # Figure out type to be stored
- types = {str: 'str', int: 'int', float: 'float',
- unicode: 'unicode', bool: 'bool', NoneType: 'none'}
- _key = types.get(type(dd[key]))
-
- # Store along with dtype info
- if _key == 'unicode':
- dd[key] = str(dd[key])
-
- hgroup.create_dataset("%s" % key, data=[dd[key]], **kwargs)
- hgroup.create_dataset("_%s" % key, data=[_key])
-
- elif type(dd[key]) in (type(np.array([1])), type(np.ma.array([1]))):
-
- if hasattr(dd[key], 'mask'):
- hgroup.create_dataset("_%s" % key, data=["masked"])
- hgroup.create_dataset("%s" % key, data=dd[key].data, **kwargs)
- hgroup.create_dataset("_%s_mask" % key, data=dd[key].mask, **kwargs)
- else:
- hgroup.create_dataset("_%s" % key, data=["ndarray"])
- hgroup.create_dataset("%s" % key, data=dd[key], **kwargs)
-
- elif type(dd[key]) is list:
- hgroup.create_dataset("%s" % key, data=dd[key], **kwargs)
- hgroup.create_dataset("_%s" % key, data=["list"])
-
- elif type(dd[key]) is tuple:
- hgroup.create_dataset("%s" % key, data=dd[key], **kwargs)
- hgroup.create_dataset("_%s" % key, data=["tuple"])
-
- elif type(dd[key]) is set:
- hgroup.create_dataset("%s" % key, data=list(dd[key]), **kwargs)
- hgroup.create_dataset("_%s" % key, data=["set"])
-
- elif isinstance(dd[key], dict):
- new_group = hgroup.create_group("%s" % key)
- _dump_dict(dd[key], new_group, **kwargs)
-
- elif type(dd[key]) is NoneType:
- hgroup.create_dataset("%s" % key, data=[0], **kwargs)
- hgroup.create_dataset("_%s" % key, data=["none"])
-
- else:
- if type(dd[key]).__module__ == np.__name__:
- #print type(dd[key])
- hgroup.create_dataset("%s" % key, data=dd[key])
- hgroup.create_dataset("_%s" % key, data=["np_dtype"])
- #new_group = hgroup.create_group("%s" % key)
- #dump_np_dtype_dict(dd[key], new_group)
- else:
- raise NoMatchError
-
-
-def dump_dict(obj, h5f='', **kwargs):
- """ dumps a dictionary to h5py file """
- h5f.create_dataset('type', data=['dict'])
- hgroup = h5f.create_group('data')
- _dump_dict(obj, hgroup, **kwargs)
-
-
-def no_match(obj, h5f, *args, **kwargs):
- """ If no match is made, raise an exception """
- try:
- import dill as cPickle
- except ImportError:
- import cPickle
-
- pickled_obj = cPickle.dumps(obj)
- h5f.create_dataset('type', data=['pickle'])
- h5f.create_dataset('data', data=[pickled_obj])
-
- print("Warning: %s type not understood, data have been serialized" % type(obj))
- #raise NoMatchError
-
-
-def dumper_lookup(obj):
- """ What type of object are we trying to pickle?
-
- This is a python dictionary based equivalent of a case statement.
- It returns the correct helper function for a given data type.
- """
- t = type(obj)
-
- types = {
- list: dump_list,
- tuple: dump_tuple,
- set: dump_set,
- dict: dump_dict,
- str: dump_string,
- unicode: dump_unicode,
- NoneType: dump_none,
- np.ndarray: dump_ndarray,
- np.ma.core.MaskedArray: dump_masked,
- np.float16: dump_np_dtype,
- np.float32: dump_np_dtype,
- np.float64: dump_np_dtype,
- np.int8: dump_np_dtype,
- np.int16: dump_np_dtype,
- np.int32: dump_np_dtype,
- np.int64: dump_np_dtype,
- np.uint8: dump_np_dtype,
- np.uint16: dump_np_dtype,
- np.uint32: dump_np_dtype,
- np.uint64: dump_np_dtype,
- np.complex64: dump_np_dtype,
- np.complex128: dump_np_dtype,
- }
-
- match = types.get(t, no_match)
- return match
-
-
-def dump(obj, file, mode='w', track_times=True, **kwargs):
- """ Write a pickled representation of obj to the open file object file.
-
- Parameters
- ----------
- obj: object
- python object o store in a Hickle
- file: file object, filename string, or h5py.File object
- file in which to store the object. A h5py.File or a filename is also acceptable.
- mode: string
- optional argument, 'r' (read only), 'w' (write) or 'a' (append). Ignored if file
- is a file object.
- compression: str
- optional argument. Applies compression to dataset. Options: None, gzip, lzf (+ szip,
- if installed)
- track_times: bool
- optional argument. If set to False, repeated hickling will produce identical files.
- """
-
- try:
- # See what kind of object to dump
- dumper = dumper_lookup(obj)
- # Open the file
- h5f = file_opener(file, mode, track_times)
- print("dumping %s to file %s" % (type(obj), repr(h5f)))
- dumper(obj, h5f, **kwargs)
- h5f.close()
- except NoMatchError:
- fname = h5f.filename
- h5f.close()
- try:
- os.remove(fname)
- except:
- print("Warning: dump failed. Could not remove %s" % fname)
- finally:
- raise NoMatchError
-
-
-#############
-## loaders ##
-#############
-
-def load(file, safe=True):
- """ Load a hickle file and reconstruct a python object
-
- Parameters
- ----------
- file: file object, h5py.File, or filename string
-
- safe (bool): Disable automatic depickling of arbitrary python objects.
- DO NOT set this to False unless the file is from a trusted source.
- (see http://www.cs.jhu.edu/~s/musings/pickle.html for an explanation)
- """
-
- try:
- h5f = file_opener(file)
- dtype = h5f["type"][0]
-
- if dtype == 'dict':
- group = h5f["data"]
- data = load_dict(group)
- elif dtype == 'pickle':
- data = load_pickle(h5f, safe)
- elif dtype == 'np_list':
- group = h5f["data"]
- data = load_np_list(group)
- elif dtype == 'np_tuple':
- group = h5f["data"]
- data = load_np_tuple(group)
- elif dtype == 'masked':
- data = np.ma.array(h5f["data"][:], mask=h5f["mask"][:])
- elif dtype == 'none':
- data = None
- else:
- if dtype in ('string', 'unicode'):
- data = h5f["data"][0]
- else:
- try:
- data = h5f["data"][:]
- except ValueError:
- data = h5f["data"]
- types = {
- 'list': list,
- 'set': set,
- 'unicode': unicode,
- 'string': str,
- 'ndarray': load_ndarray,
- 'np_dtype': load_np_dtype
- }
-
- mod = types.get(dtype, no_match)
- data = mod(data)
- finally:
- if 'h5f' in locals():
- h5f.close()
- return data
-
-
-def load_pickle(h5f, safe=True):
- """ Deserialize and load a pickled object within a hickle file
-
- WARNING: Pickle has
-
- Parameters
- ----------
- h5f: h5py.File object
-
- safe (bool): Disable automatic depickling of arbitrary python objects.
- DO NOT set this to False unless the file is from a trusted source.
- (see http://www.cs.jhu.edu/~s/musings/pickle.html for an explanation)
- """
-
- if not safe:
- try:
- import dill as cPickle
- except ImportError:
- import cPickle
-
- data = h5f["data"][:]
- data = cPickle.loads(data[0])
- return data
- else:
- print("\nWarning: Object is of an unknown type, and has not been loaded")
- print(" for security reasons (it could be malicious code). If")
- print(" you wish to continue, manually set safe=False\n")
-
-
-def load_np_list(group):
- """ load a numpy list """
- np_list = []
- for key in sorted(group.keys()):
- data = group[key][:]
- np_list.append(data)
- return np_list
-
-
-def load_np_tuple(group):
- """ load a tuple containing numpy arrays """
- return tuple(load_np_list(group))
-
-
-def load_ndarray(arr):
- """ Load a numpy array """
- # Nothing to be done!
- return arr
-
-
-def load_np_dtype(arr):
- """ Load a numpy array """
- # Just return first value
- return arr.value
-
-
-def load_dict(group):
- """ Load dictionary """
-
- dd = {}
- for key in group.keys():
- if isinstance(group[key], h5._hl.group.Group):
- new_group = group[key]
- dd[key] = load_dict(new_group)
- elif not key.startswith("_"):
- _key = "_%s" % key
-
- if group[_key][0] == 'np_dtype':
- dd[key] = group[key].value
- elif group[_key][0] in ('str', 'int', 'float', 'unicode', 'bool'):
- dd[key] = group[key][0]
- elif group[_key][0] == 'masked':
- key_ma = "_%s_mask" % key
- dd[key] = np.ma.array(group[key][:], mask=group[key_ma])
- else:
- dd[key] = group[key][:]
-
- # Convert numpy constructs back to string
- dtype = group[_key][0]
- types = {'str': str, 'int': int, 'float': float,
- 'unicode': unicode, 'bool': bool, 'list': list, 'none' : NoneType}
- try:
- mod = types.get(dtype)
- if dtype == 'none':
- dd[key] = None
- else:
- dd[key] = mod(dd[key])
- except:
- pass
- return dd
-
-
-def load_large(file):
- """ Load a large hickle file (returns the h5py object not the data)
-
- Parameters
- ----------
- file: file object, h5py.File, or filename string
- """
-
- h5f = file_opener(file)
- return h5f
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle_legacy2.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle_legacy2.py
deleted file mode 100755
index 4d018fde..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/hickle_legacy2.py
+++ /dev/null
@@ -1,672 +0,0 @@
-# encoding: utf-8
-"""
-# hickle_legacy2.py
-
-Created by Danny Price 2016-02-03.
-
-This is a legacy handler, for hickle v2 files.
-If V3 reading fails, this will be called as a fail-over.
-
-"""
-
-import os
-import numpy as np
-import h5py as h5
-import re
-
-try:
- from exceptions import Exception
- from types import NoneType
-except ImportError:
- pass # above imports will fail in python3
-
-import warnings
-__version__ = "2.0.4"
-__author__ = "Danny Price"
-
-
-##################
-# Error handling #
-##################
-
-class FileError(Exception):
- """ An exception raised if the file is fishy """
- def __init__(self):
- return
-
- def __str__(self):
- return ("Cannot open file. Please pass either a filename "
- "string, a file object, or a h5py.File")
-
-
-class ClosedFileError(Exception):
- """ An exception raised if the file is fishy """
- def __init__(self):
- return
-
- def __str__(self):
- return ("HDF5 file has been closed. Please pass either "
- "a filename string, a file object, or an open h5py.File")
-
-
-class NoMatchError(Exception):
- """ An exception raised if the object type is not understood (or
- supported)"""
- def __init__(self):
- return
-
- def __str__(self):
- return ("Error: this type of python object cannot be converted into a "
- "hickle.")
-
-
-class ToDoError(Exception):
- """ An exception raised for non-implemented functionality"""
- def __init__(self):
- return
-
- def __str__(self):
- return "Error: this functionality hasn't been implemented yet."
-
-
-######################
-# H5PY file wrappers #
-######################
-
-class H5GroupWrapper(h5.Group):
- """ Group wrapper that provides a track_times kwarg.
-
- track_times is a boolean flag that can be set to False, so that two
- files created at different times will have identical MD5 hashes.
- """
- def create_dataset(self, *args, **kwargs):
- kwargs['track_times'] = getattr(self, 'track_times', True)
- return super(H5GroupWrapper, self).create_dataset(*args, **kwargs)
-
- def create_group(self, *args, **kwargs):
- group = super(H5GroupWrapper, self).create_group(*args, **kwargs)
- group.__class__ = H5GroupWrapper
- group.track_times = getattr(self, 'track_times', True)
- return group
-
-
-class H5FileWrapper(h5.File):
- """ Wrapper for h5py File that provides a track_times kwarg.
-
- track_times is a boolean flag that can be set to False, so that two
- files created at different times will have identical MD5 hashes.
- """
- def create_dataset(self, *args, **kwargs):
- kwargs['track_times'] = getattr(self, 'track_times', True)
- return super(H5FileWrapper, self).create_dataset(*args, **kwargs)
-
- def create_group(self, *args, **kwargs):
- group = super(H5FileWrapper, self).create_group(*args, **kwargs)
- group.__class__ = H5GroupWrapper
- group.track_times = getattr(self, 'track_times', True)
- return group
-
-
-def file_opener(f, mode='r', track_times=True):
- """ A file opener helper function with some error handling. This can open
- files through a file object, a h5py file, or just the filename.
-
- Args:
- f (file, h5py.File, or string): File-identifier, e.g. filename or file object.
- mode (str): File open mode. Only required if opening by filename string.
- track_times (bool): Track time in HDF5; turn off if you want hickling at
- different times to produce identical files (e.g. for MD5 hash check).
-
- """
- # Were we handed a file object or just a file name string?
- if isinstance(f, file):
- filename, mode = f.name, f.mode
- f.close()
- h5f = h5.File(filename, mode)
- elif isinstance(f, str) or isinstance(f, unicode):
- filename = f
- h5f = h5.File(filename, mode)
- elif isinstance(f, H5FileWrapper) or isinstance(f, h5._hl.files.File):
- try:
- filename = f.filename
- except ValueError:
- raise ClosedFileError()
- h5f = f
- else:
- print(type(f))
- raise FileError
-
- h5f.__class__ = H5FileWrapper
- h5f.track_times = track_times
- return h5f
-
-
-###########
-# DUMPERS #
-###########
-
-def check_is_iterable(py_obj):
- """ Check whether a python object is iterable.
-
- Note: this treats unicode and string as NON ITERABLE
-
- Args:
- py_obj: python object to test
-
- Returns:
- iter_ok (bool): True if item is iterable, False is item is not
- """
- if type(py_obj) in (str, unicode):
- return False
- try:
- iter(py_obj)
- return True
- except TypeError:
- return False
-
-
-def check_iterable_item_type(iter_obj):
- """ Check if all items within an iterable are the same type.
-
- Args:
- iter_obj: iterable object
-
- Returns:
- iter_type: type of item contained within the iterable. If
- the iterable has many types, a boolean False is returned instead.
-
- References:
- http://stackoverflow.com/questions/13252333/python-check-if-all-elements-of-a-list-are-the-same-type
- """
- iseq = iter(iter_obj)
- first_type = type(next(iseq))
- return first_type if all((type(x) is first_type) for x in iseq) else False
-
-
-def check_is_numpy_array(py_obj):
- """ Check if a python object is a numpy array (masked or regular)
-
- Args:
- py_obj: python object to check whether it is a numpy array
-
- Returns
- is_numpy (bool): Returns True if it is a numpy array, else False if it isn't
- """
-
- is_numpy = type(py_obj) in (type(np.array([1])), type(np.ma.array([1])))
-
- return is_numpy
-
-
-def _dump(py_obj, h_group, call_id=0, **kwargs):
- """ Dump a python object to a group within a HDF5 file.
-
- This function is called recursively by the main dump() function.
-
- Args:
- py_obj: python object to dump.
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
-
- dumpable_dtypes = set([bool, int, float, long, complex, str, unicode])
-
- # Firstly, check if item is a numpy array. If so, just dump it.
- if check_is_numpy_array(py_obj):
- create_hkl_dataset(py_obj, h_group, call_id, **kwargs)
-
- # next, check if item is iterable
- elif check_is_iterable(py_obj):
- item_type = check_iterable_item_type(py_obj)
-
- # item_type == False implies multiple types. Create a dataset
- if item_type is False:
- h_subgroup = create_hkl_group(py_obj, h_group, call_id)
- for ii, py_subobj in enumerate(py_obj):
- _dump(py_subobj, h_subgroup, call_id=ii, **kwargs)
-
- # otherwise, subitems have same type. Check if subtype is an iterable
- # (e.g. list of lists), or not (e.g. list of ints, which should be treated
- # as a single dataset).
- else:
- if item_type in dumpable_dtypes:
- create_hkl_dataset(py_obj, h_group, call_id, **kwargs)
- else:
- h_subgroup = create_hkl_group(py_obj, h_group, call_id)
- for ii, py_subobj in enumerate(py_obj):
- #print py_subobj, h_subgroup, ii
- _dump(py_subobj, h_subgroup, call_id=ii, **kwargs)
-
- # item is not iterable, so create a dataset for it
- else:
- create_hkl_dataset(py_obj, h_group, call_id, **kwargs)
-
-
-def dump(py_obj, file_obj, mode='w', track_times=True, path='/', **kwargs):
- """ Write a pickled representation of obj to the open file object file.
-
- Args:
- obj (object): python object o store in a Hickle
- file: file object, filename string, or h5py.File object
- file in which to store the object. A h5py.File or a filename is also
- acceptable.
- mode (str): optional argument, 'r' (read only), 'w' (write) or 'a' (append).
- Ignored if file is a file object.
- compression (str): optional argument. Applies compression to dataset. Options: None, gzip,
- lzf (+ szip, if installed)
- track_times (bool): optional argument. If set to False, repeated hickling will produce
- identical files.
- path (str): path within hdf5 file to save data to. Defaults to root /
- """
-
- try:
- # Open the file
- h5f = file_opener(file_obj, mode, track_times)
- h5f.attrs["CLASS"] = 'hickle'
- h5f.attrs["VERSION"] = 2
- h5f.attrs["type"] = ['hickle']
-
- h_root_group = h5f.get(path)
-
- if h_root_group is None:
- h_root_group = h5f.create_group(path)
- h_root_group.attrs["type"] = ['hickle']
-
- _dump(py_obj, h_root_group, **kwargs)
- h5f.close()
- except NoMatchError:
- fname = h5f.filename
- h5f.close()
- try:
- os.remove(fname)
- except OSError:
- warnings.warn("Dump failed. Could not remove %s" % fname)
- finally:
- raise NoMatchError
-
-
-def create_dataset_lookup(py_obj):
- """ What type of object are we trying to pickle? This is a python
- dictionary based equivalent of a case statement. It returns the correct
- helper function for a given data type.
-
- Args:
- py_obj: python object to look-up what function to use to dump to disk
-
- Returns:
- match: function that should be used to dump data to a new dataset
- """
- t = type(py_obj)
-
- types = {
- dict: create_dict_dataset,
- list: create_listlike_dataset,
- tuple: create_listlike_dataset,
- set: create_listlike_dataset,
- str: create_stringlike_dataset,
- unicode: create_stringlike_dataset,
- int: create_python_dtype_dataset,
- float: create_python_dtype_dataset,
- long: create_python_dtype_dataset,
- bool: create_python_dtype_dataset,
- complex: create_python_dtype_dataset,
- NoneType: create_none_dataset,
- np.ndarray: create_np_array_dataset,
- np.ma.core.MaskedArray: create_np_array_dataset,
- np.float16: create_np_dtype_dataset,
- np.float32: create_np_dtype_dataset,
- np.float64: create_np_dtype_dataset,
- np.int8: create_np_dtype_dataset,
- np.int16: create_np_dtype_dataset,
- np.int32: create_np_dtype_dataset,
- np.int64: create_np_dtype_dataset,
- np.uint8: create_np_dtype_dataset,
- np.uint16: create_np_dtype_dataset,
- np.uint32: create_np_dtype_dataset,
- np.uint64: create_np_dtype_dataset,
- np.complex64: create_np_dtype_dataset,
- np.complex128: create_np_dtype_dataset
- }
-
- match = types.get(t, no_match)
- return match
-
-
-def create_hkl_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Create a dataset within the hickle HDF5 file
-
- Args:
- py_obj: python object to dump.
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
-
- """
- #lookup dataset creator type based on python object type
- create_dataset = create_dataset_lookup(py_obj)
-
- # do the creation
- create_dataset(py_obj, h_group, call_id, **kwargs)
-
-
-def create_hkl_group(py_obj, h_group, call_id=0):
- """ Create a new group within the hickle file
-
- Args:
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
-
- """
- h_subgroup = h_group.create_group('data_%i' % call_id)
- h_subgroup.attrs["type"] = [str(type(py_obj))]
- return h_subgroup
-
-
-def create_listlike_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Dumper for list, set, tuple
-
- Args:
- py_obj: python object to dump; should be list-like
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- dtype = str(type(py_obj))
- obj = list(py_obj)
- d = h_group.create_dataset('data_%i' % call_id, data=obj, **kwargs)
- d.attrs["type"] = [dtype]
-
-
-def create_np_dtype_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an np dtype object to h5py file
-
- Args:
- py_obj: python object to dump; should be a numpy scalar, e.g. np.float16(1)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj, **kwargs)
- d.attrs["type"] = ['np_dtype']
- d.attrs["np_dtype"] = str(d.dtype)
-
-
-def create_python_dtype_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps a python dtype object to h5py file
-
- Args:
- py_obj: python object to dump; should be a python type (int, float, bool etc)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj,
- dtype=type(py_obj), **kwargs)
- d.attrs["type"] = ['python_dtype']
- d.attrs['python_subdtype'] = str(type(py_obj))
-
-
-def create_dict_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Creates a data group for each key in dictionary
-
- Args:
- py_obj: python object to dump; should be dictionary
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- h_dictgroup = h_group.create_group('data_%i' % call_id)
- h_dictgroup.attrs["type"] = ['dict']
- for key, py_subobj in py_obj.items():
- h_subgroup = h_dictgroup.create_group(key)
- h_subgroup.attrs["type"] = ['dict_item']
- _dump(py_subobj, h_subgroup, call_id=0, **kwargs)
-
-
-def create_np_array_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an ndarray object to h5py file
-
- Args:
- py_obj: python object to dump; should be a numpy array or np.ma.array (masked)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- if isinstance(py_obj, type(np.ma.array([1]))):
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj, **kwargs)
- #m = h_group.create_dataset('mask_%i' % call_id, data=py_obj.mask, **kwargs)
- m = h_group.create_dataset('data_%i_mask' % call_id, data=py_obj.mask, **kwargs)
- d.attrs["type"] = ['ndarray_masked_data']
- m.attrs["type"] = ['ndarray_masked_mask']
- else:
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj, **kwargs)
- d.attrs["type"] = ['ndarray']
-
-
-def create_stringlike_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps a list object to h5py file
-
- Args:
- py_obj: python object to dump; should be string-like (unicode or string)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- if isinstance(py_obj, str):
- d = h_group.create_dataset('data_%i' % call_id, data=[py_obj], **kwargs)
- d.attrs["type"] = ['string']
- else:
- dt = h5.special_dtype(vlen=unicode)
- dset = h_group.create_dataset('data_%i' % call_id, shape=(1, ), dtype=dt, **kwargs)
- dset[0] = py_obj
- dset.attrs['type'] = ['unicode']
-
-
-def create_none_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Dump None type to file
-
- Args:
- py_obj: python object to dump; must be None object
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- d = h_group.create_dataset('data_%i' % call_id, data=[0], **kwargs)
- d.attrs["type"] = ['none']
-
-
-def no_match(py_obj, h_group, call_id=0, **kwargs):
- """ If no match is made, raise an exception
-
- Args:
- py_obj: python object to dump; default if item is not matched.
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- try:
- import dill as cPickle
- except ImportError:
- import cPickle
-
- pickled_obj = cPickle.dumps(py_obj)
- d = h_group.create_dataset('data_%i' % call_id, data=[pickled_obj])
- d.attrs["type"] = ['pickle']
-
- warnings.warn("%s type not understood, data have been "
- "serialized" % type(py_obj))
-
-
-#############
-## LOADERS ##
-#############
-
-class PyContainer(list):
- """ A group-like object into which to load datasets.
-
- In order to build up a tree-like structure, we need to be able
- to load datasets into a container with an append() method.
- Python tuples and sets do not allow this. This class provides
- a list-like object that be converted into a list, tuple, set or dict.
- """
- def __init__(self):
- super(PyContainer, self).__init__()
- self.container_type = None
- self.name = None
-
- def convert(self):
- """ Convert from PyContainer to python core data type.
-
- Returns: self, either as a list, tuple, set or dict
- """
- if self.container_type == "<type 'list'>":
- return list(self)
- if self.container_type == "<type 'tuple'>":
- return tuple(self)
- if self.container_type == "<type 'set'>":
- return set(self)
- if self.container_type == "dict":
- keys = [str(item.name.split('/')[-1]) for item in self]
- items = [item[0] for item in self]
- return dict(zip(keys, items))
- else:
- return self
-
-
-def load(fileobj, path='/', safe=True):
- """ Load a hickle file and reconstruct a python object
-
- Args:
- fileobj: file object, h5py.File, or filename string
- safe (bool): Disable automatic depickling of arbitrary python objects.
- DO NOT set this to False unless the file is from a trusted source.
- (see http://www.cs.jhu.edu/~s/musings/pickle.html for an explanation)
-
- path (str): path within hdf5 file to save data to. Defaults to root /
- """
-
- try:
- h5f = file_opener(fileobj)
- h_root_group = h5f.get(path)
-
- try:
- assert 'CLASS' in h5f.attrs.keys()
- assert 'VERSION' in h5f.attrs.keys()
- py_container = PyContainer()
- py_container.container_type = 'hickle'
- py_container = _load(py_container, h_root_group)
- return py_container[0][0]
- except AssertionError:
- import hickle_legacy
- return hickle_legacy.load(fileobj, safe)
- finally:
- if 'h5f' in locals():
- h5f.close()
-
-
-def load_dataset(h_node):
- """ Load a dataset, converting into its correct python type
-
- Args:
- h_node (h5py dataset): h5py dataset object to read
-
- Returns:
- data: reconstructed python object from loaded data
- """
- py_type = h_node.attrs["type"][0]
-
- if h_node.shape == ():
- data = h_node.value
- else:
- data = h_node[:]
-
- if py_type == "<type 'list'>":
- #print self.name
- return list(data)
- elif py_type == "<type 'tuple'>":
- return tuple(data)
- elif py_type == "<type 'set'>":
- return set(data)
- elif py_type == "np_dtype":
- subtype = h_node.attrs["np_dtype"]
- data = np.array(data, dtype=subtype)
- return data
- elif py_type == 'ndarray':
- return np.array(data)
- elif py_type == 'ndarray_masked_data':
- try:
- mask_path = h_node.name + "_mask"
- h_root = h_node.parent
- mask = h_root.get(mask_path)[:]
- except IndexError:
- mask = h_root.get(mask_path)
- except ValueError:
- mask = h_root.get(mask_path)
- data = np.ma.array(data, mask=mask)
- return data
- elif py_type == 'python_dtype':
- subtype = h_node.attrs["python_subdtype"]
- type_dict = {
- "<type 'int'>": int,
- "<type 'float'>": float,
- "<type 'long'>": long,
- "<type 'bool'>": bool,
- "<type 'complex'>": complex
- }
- tcast = type_dict.get(subtype)
- return tcast(data)
- elif py_type == 'string':
- return str(data[0])
- elif py_type == 'unicode':
- return unicode(data[0])
- elif py_type == 'none':
- return None
- else:
- print(h_node.name, py_type, h_node.attrs.keys())
- return data
-
-
-def sort_keys(key_list):
- """ Take a list of strings and sort it by integer value within string
-
- Args:
- key_list (list): List of keys
-
- Returns:
- key_list_sorted (list): List of keys, sorted by integer
- """
- to_int = lambda x: int(re.search('\d+', x).group(0))
- keys_by_int = sorted([(to_int(key), key) for key in key_list])
- return [ii[1] for ii in keys_by_int]
-
-
-def _load(py_container, h_group):
- """ Load a hickle file
-
- Recursive funnction to load hdf5 data into a PyContainer()
-
- Args:
- py_container (PyContainer): Python container to load data into
- h_group (h5 group or dataset): h5py object, group or dataset, to spider
- and load all datasets.
- """
-
- group_dtype = h5._hl.group.Group
- dataset_dtype = h5._hl.dataset.Dataset
-
- #either a file, group, or dataset
- if isinstance(h_group, H5FileWrapper) or isinstance(h_group, group_dtype):
- py_subcontainer = PyContainer()
- py_subcontainer.container_type = h_group.attrs['type'][0]
- py_subcontainer.name = h_group.name
-
- if py_subcontainer.container_type != 'dict':
- h_keys = sort_keys(h_group.keys())
- else:
- h_keys = h_group.keys()
-
- for h_name in h_keys:
- h_node = h_group[h_name]
- py_subcontainer = _load(py_subcontainer, h_node)
-
- sub_data = py_subcontainer.convert()
- py_container.append(sub_data)
-
- else:
- # must be a dataset
- subdata = load_dataset(h_group)
- py_container.append(subdata)
-
- #print h_group.name, py_container
- return py_container
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__init__.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__init__.py
deleted file mode 100755
index 3be6bd29..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__init__.py
+++ /dev/null
@@ -1 +0,0 @@
-from __future__ import absolute_import
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/__init__.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/__init__.cpython-36.pyc
deleted file mode 100644
index 059bbcb18d24b4ed243c011342d5220fc0ca9b4b..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 252
zcmXr!<>i`^S{<v#z`*dB0SPcMFfceUFfbGgFfcHrFr+Z%Fyt~uF)}iwFa<McGQR{V
z)nvTIpO{pfpHo_r8lRb4kY7~dr^$4SD?UE0w4}5sH9o!wWXwv2A{GV)2=S|2zd*mB
zC_gJTxkNuXE2}u!#7I9WGcR2~H%GTPwJfzLGfzJwGdVjaRX-;)Nxz`7BqKl1SkFwq
zII|>Gw;(Y&J25@A7^+0qSkFYySht|kSkFu^H9Z|Fte=ygn37sltREkrnU`4-AFo$X
Ud5gm)H$SB`C)JJ-<SUT#0mmCgmH+?%
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_astropy.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_astropy.cpython-36.pyc
deleted file mode 100644
index 3856511eb477a1bd3a48b33dd2325efe9afb4735..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 7453
zcmXr!<>i`^S{-{?fq~&M0}|k7U|?`yU|=Y=U}RuOVMt-jVTfW(VMt-hVajEWVrB%1
zG3T)4vPQAyvPH3h`7Al?xg1d(xtvj)U_NUOS1xxHH<-<q!;{M!#S3P$=kVq7NAZK%
zj5z{P0x8-l94VYFj8TFqIw@Q!+%1ezLMgf_JSn^_j8VcVdMSJ<{4I=8A}RVQ0x5zm
zj8UQ~1}Q=*!YzzZVksgiqAd(j;wfS&;w=nO5-AcXk}V8Tk||Ot(k%>8QYkVivMmfz
z(kXH&@+}NeGARlviY*LLvMEX_$}J30a;eNIDyf30;wh@Ba_MZTyeVp_BB>Hk9&3tv
zs!*yVl*g8$kt&=j1?6$1Xr_v$N~fr%Fs8Gma;Ipeilxe=sHV!Mv!$~|$)_*|GiVyV
z1O=&|CgUxRz|zFLlFX7yO{QC{j(O=hskb<Svn!qR^NUh6nQn17=jRodB<7Xe;&Szc
zu{9ZQv4$ij<)msd-eL*K%mvYc>8U00C6xuK@rijU@hOQViOI|$I~W)k*cccXSQ!`?
zyg|_x!oa{#!<fa8!dSy3#!$nM#hAjB!raST%aX#9&RD~e#gxKo!%)JU#ZqKd!dk*s
z!;r<^%$UN~%v8&o#|7ipvevLdRkPbLq;Lc?XmTd<GcqtJq?G0s6e}d=DI^w`6y+CG
zDilKf$Hk?fprGJblwNEFW`TGGmGSvWSyl=Kl_eSZc?$VSS*gh-3MKhqQ>+z=GxAGw
zQWTO>6%t`eKt5E^$jmFzQAo?lPb|?<NXpO8QAjOG)`Xjy5uaX^Us|A`kzuOmmYI{P
z2j*&8DS&C1@d}_YQpn6J$=8GHOis+niO);{TcByBkeQc~S^-j@nUb1Ul9^Tsb)kB(
zLQ!f?Vo7FMszOeFa$-qlex5>Ro<d1RszPQ-Y7r=EAcj`4gKUXc&Ai19cCML;Cetkz
zkk~EO6p&wv1sE6@pg|Q63#wZzAgNodiN(p8nYUO<^D;~PG&yeZB^RY8mZX9sw>Y)r
z7He5zPHE~b*2I#MqT*Z3#hDeim;)k>Z?Ps-mZX9NN{Wj385kHe*>15xLi!dvG;D6M
zL%n*7ExSCiD82X=V+vU9N`@jC1_p*-)%pec1x5Lw_|Q+z$|_DaG15=U%uCnL&CxAR
zElVxR%+t@vOwP_p)z8UH(uYK#v7VWJab`)XZb4#lc4B&JF;t1Jv7U*Zv2H=7v7VV;
zYI-_USU)E}F(tLA7)-}Q13|B#@|HN%qfjAuxPZ!@Vm?r+=Hz4KVB}$BV-#ZIVl0wl
zU|^s|YOi6)Vryp1VozafW(22pR*1-GY9}$Z7jc1-9a9k~!iso6#SSl0GEW4Raz!AK
zB2Z+4)199tM-eD16bUjgFcb-a2w@N*0wTmf1SlonVoS|S&QD3zWGfN}iAjJ6Nf03g
zBBVhCNF6v0LkL+028LTAXbB4J0+0;0)TDr%nm`_bmz}FXWhX-xLl$EeQ!Pskh{c@6
zRLfezlEqrXTEmjY)XZ4RR>PRZ1}$CLpk*sNxNK$5Vh5M499f)2MI~HN{oKurSv=qr
z#{m^%%tDB9=E*=snL(npC~9gsYdF#wYe43*+c4B{LfITP3@Mz!44Pb{B{H5888eN7
z(jTnIDbB7;235g`WO|DuCqFM9R2JXj$Vn^#v-~u9!D;*!b53FjIO*PE&dJZa#gtcY
zizhihr!+S&zPKbYIUAC;aV6^_c~ExY!kVU)L1|hAl!%#9Qi{MSToELs1R_91HBvH1
zFE?SH&<Eu}P!3_><YMGv<YVH2mYyPvVoZFDe2hGdMH)C${W)l=uVpD=1gBY0DrZe$
ztYOV&D+;S&&0<brOko1G4@y{SSeqFc8PXYQSt^u3YF!v&C2HAfSh8447;8XfD_aV4
zGb2b%p;b7j9m1Z%0&a(})Uc+o_A=M9r?7!jF-Ho!4MPcM7FW?Eu&vz9jI|tK7Ed!{
zEoYtrm<`G&U=tB?Sr9%fXK<u&z;gyi3a1T23Rf_LCU>F`*0u_$Ed{OPM`m+{$tJXl
zsU{d)F2ma{n#ct?s5FOVl9J5aRB#5l#g>*|l$%&`i?uj8F((z2Ex_eCxZnoqxW!(S
zT9A{NoO+A3peQr1<Q7|EK|yL>3OMDXWe!jy#;8aWl)kla=L>k%a*HLiBsCYDGj1`b
z<(3qIG(j>2ID3GK`CGzh6#&TP9-v$S%0>)aY>Yh2p!Ng@BMT!BqXd%(xNX73$ipPS
zC;-YGhRC@CR3CsaxIH}qlskxMPuDPHu{1M6+stgB7IqCo7Asr?l-ijgtqrIe4yYP-
zxEf9yP<tIz;)7c0Twoq&QBMs+7O3P0iE-O7r0@hYX!7DH`IDiYiIJJOVQq2zB|j{T
zzzG}NHo3)?n3PnMT6T*iFEKau7H3gvT53^hUUKR!?$W&E)S{Ba%)FAyTWrOZ#U-h^
zewtj61P#ss;36NK4Z!UZBT%X~1`#G8!W2Z9ff6|@*1{iL+=Ejwm;fhb^g<ry-V9K3
z1~r!$pp9c8MhN6%Vq+{aM^3<;$Q{B4Py%MiVrXWlWvOAzVuUnZ(im%4z@0&+IZQRo
z)0z4hYgtp6(;30VJ&O$ksL500mBO0B)&feDY%UC-rcWNIyU13{R>KBX$8N(=p#bVY
zW^pt#xG=;D#W2;f*K*WwfMjdfZ5RsE!eO09Zcu^jQkq+U+;;?baz}PzMJaqiZDx4k
z3+iQp6KOIds7u7az`z8J*I-am042I~h8l)gzFNi-h8o5chGwRTOoc4L44RBZpmw(=
zW05E*=`r16(lfZll$?iFX#V1|$;m7(NzTtLu<Hhy2r{0Hp-Kb2ECo44&j!>H&MCIj
zLuk6ik({3c$}+`%nw+;d5{u&#i;5B}A-MxoE1<XYtwGHkmfX~m#9OSOLcLg%6Kl2r
zm;K<p0oK8somvTQe4%*&quE~qa%mmNE1+!2z`?;N$H>D3>O`_Jaxsc97J&w~;GqdB
z;z2_w&L9n-=1~bl4MQ_y3{x#*EfaX)gi(?qg-Md3mbrwfhN*_R8Pq8Ng)g%oxZEfL
z#fv6$5h%Msyb20IP1aj%8S#1fDXF*E3o5}wHsCfSSj#QOLUd2af_llY-Z8u@3>yTP
z400R?DE=9F82K2v7>is$IUQs(NDwyk1NK7}0|NuHA4(Xrn2Ky_7(fj;h)=RuK+UsS
zP({R?!h#~kk|%?#5>-ttO9?9|1Zr5ojX2gC)?fxrHa|!u4sr`P!GUXLO(uvasFv1b
zgM<yDrw59cTWl$*piW*9$Xl9hMUJ3C(g{R><J|zn0_iRSc?vzWBvC>ORL#ReY9=V8
zKuvrGaODnb`tmUrxgm!b4PqUXh%{Nz{Q<5Iuz5iW#S7rBD$EZ{Kz;xP3&;=P*mg&D
zJ4$Sap~W_8iUsqSAu(Lb0!qIr3@Hq)OesuhOp**WOi&&Rn8yt2+M>oRxDNu3Sx^h=
zC5WZT2u{K%37!R1z+glnSTjf;)+j_x@UT7%EC@h-@M2J}91;Y4j9QFEp2&d!k^==1
zC}@kjNC*UQkU$fCEokJ2F@+TxK%i=kF@+7vV^3kI<w#)%sn23+X2@nMDgv=n*rD<q
zCCoJ}HINL!Uc*+y(ac`USp({kb08HAML636ERb9Q&I>Hrsg=b=phT(3g*64iO}WKf
zl$sXBT$-0$1Zq*-VoXNQ8>j&bYd63Gb|WZlfhrAHz=Ar4d`v8ie2jcdB8)}8$bpQK
z($dfZ8Qe8v%3^K?wF#JOn6p??z+FBtiwV^ED`CxIEAmQVPGJGJ9+*I-D7Xc|k_Q?V
z0rP8FO4w_dYFKKRQBxkM4lDwd`|yGqRH}j!8b-AT9_MCgW<;;{TtTUlsmKCUl30QW
zaO(<8fSOoERtyXb$3ejXGMSB`N)07Pf@=|M^&MygA6(rP*@7(K07v63mXyro5=ilH
z2ddwg3-Sw!5G5WvNGW;_xy1ph1XB`A62U&b#aMzKmqM6MIskGpDA7P7kc*LviH%W=
ziHnJku?Q42@bCg<R#2G)&LN=e0;&{B7@L_uO#?qordwQ@#hKt)hvd{Eu=^Mx)i5}O
zzyv6mZiyvlq$X#_XBNY2Cveb!#tVu;A;-X26pEZ^Q0xU&wJ>{&HNY&UW+o_$8O~y9
zW-77(^^}T?Koch^Ou-C!EMOKxFoPzupIe9~TM=kZqX=XpxXLI3`K<_)AZ~GF7K1vZ
ziA9>&eISqF1K1Eeq^WWd6r`Y*3In4EqX;7#BMTE7Q&Bib6jYPInqn|bA;loIFi}wD
zQw&P2FadB;jxyj6E8I{7VHp!eFF0g186iyyA}EDhta+um1(ijr3=9lWyih0TfrhV&
z!9f)z3=;#7b7kfwmZTOVf-XuBAq`FnC6H362$b^?=7F6BP9sq~FbyDogCin}4>Cib
zmyw!NkXlp>&JU2J4;ryA0u8i+$DN8m-D_|c1>6DyH(|iFauKL{0~hb$;s>0J!D$K{
z2e<f=a}tY-<BL+$GmA@7i;(gNB9U^~<mRW8=A_zzQeQEs83F3(L%Q`mpn)(ZE+H-<
iE=ev0E-@}q4mJ*M4mJ*U4t@?k4rUGk4n{UcObh^G(c6#!
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_numpy.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_numpy.cpython-36.pyc
deleted file mode 100644
index 0b431b401afea544b1894a3b08b282d070c988e1..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 4199
zcmXr!<>i`^S{=(K#K7>F0SPEDFfceUFfbIqVPIfLVTfW#VN7AlVT@t~(@as!DU2!1
zIV`!XQLKy%DJ&_hEeug?DdH(?DeNtbQS2!aDI6)BEsRkdDUvB%DcmiLQJksFSzM_g
zk~@nhix<q|OJz@G&f*7?0$@@QObUTXVK6BICPl%d7?>0XlM-N35==^gN$FIMOo1qw
z6rL2`7KSL<6uuPx7KSLf6oC}M7KSMK6rmL17KSK=6p<9s7KSLrRM}LyRQXi-RE1Q<
z6tOvMQA#OH!3>&GRZ(2Z3OV_SDe-xwxdoMa1(jS}p(U9)nI)O2#R`deDGDj2xdjUP
zU^NOEiFqkGsYS&KY57GEPG(-Z0$3+lZE<pDL8U@sQBh)LF&EcMP!RiRGTstQPc4Zr
zsVqp1Ps~e+Pf09EOlAbhF)%PNF)%Q&GB7YWgMwd=fq|ifp@yNEv4o+7F^egMF`coQ
z5z1pm@Y9&G*=iYU7=sx!nW{c3IA^3LXDei;DI_WsRF-7q=PBeTWu+#UC}b8ZB*I(=
za+88aZenqEYKlUBkwQ^wdTCB#ktP?Hf`WpAV^Mms6_^F$6;#IOCuLconO2gokPNn|
zJR`LvBeh5&6VnEWX+fzarA2wgaPu;YA)%|Fk(8gGqiLl8RiF@3RGJEMI2Mz16jF1F
zQx)72L1bnc#Iii~60o5!LBXfVa*G8NsJEE%3U0B21N|0LZlWgBEjEY;qBvj{uVg6V
zWME+URjFT~Ur>|}3LO39tgPZ>6C?ek%)E5{+#KEF)Uwo~%sl;!%;fBxRQ;UHBz;I|
z8ta+q7iX5F>J}s>XD6no7DJWj8ta+p8S55Q8ta+qrKYDth4n$HDYd8=Ohf#wS5SFN
z1RQqpFc-yx-OCS(brw)k;L>6$;$~oANM;7{posz`;|)?0z`(#z!<fa8!dSy3#!$nM
z!qm%D%aX#J&RD~e#hAik!%)JM#av`o!cxPM#o7!JWwl|*6DwhZiKVbLGl5mJ+AyTB
z2Qz4LBuX+eFo4rHD5Wdp6)2>D!V{KyO7aylObaR%(lT>WA@NFNDgosfYlY&B{L-8h
zg``w?8Z1su%t<WLQApKG*Hg$V&`Zn7Pb@Js(=gNoM=B`gWyGf!<(C#HXk?h`xn<_0
z>VdhMRtjKRp(G#dTm?`TQpn6J$=8GHOis+niO)<?(8$ay(X>*?%u7kF0IAPRNzE(C
zOsj<YO1)U2C^aXsB(p44AtygMu_QA;Pa!i;p(G<!A+sd4C^0D~6=G-=JII!J)l5ys
zTPz@EF(?V;6~sflQ3Q&dTO4@>@!&wa#hO^0oSEsT$$pD3xhOTUBo&;`i&IN(u_l(3
z6cyiMF3zmD#T*c6bc?yTr05naR9jMINouhsYY_(n1H&zLsQ+)VL%nm0ExSCiD82X=
zV+uG0LkKRAS_!B%a2LTWlmn$0kP-$iAx0i1Hbx<)A}O3nO9Lfol`w)zpL9l)1jYnP
zU@5G@44Q1CNruQIQ^W&`1yH2sK~o4g!T4!17lAT;kuU=TLy-Uj0|PYbi^M<z;vhl-
zL`Z@NP=Z5>aehRMgA<P-$O=$0WZ;lsERsf!Vm4?A0xmr!L93M#Mo=k{#R98Vnwg3+
zY8bOvQ<y-dN*ZGda}7f#V-53kras167O)HpLMDq1F2|a}3NCrrQ`kUdNjhT<YZjQD
zrv}!A)w~oABt4wL44Pbt(x~N63b;&!W>iq#rEc*9HN6N_^yDV$LAcNwaZr{(MWDI>
zUJQX#R22`XB#Kwfj0e@^#h_G_2MIQ?Kq{nl0ZwYzB*B{485kI#n*205ZgFK6XXX``
zB<3Zjg7p=Fnj=LZMMa=g1WvBESU@@<>6Hy!RAEc6w-|FFsTG`VK_$>FaYT{@hc~Q@
zvIQkwK2TN1A;BocB*VzV$j2nbSR@Zh#GuL@Bn9$52!q>3pi&LeHi}`YWvpcawU26;
zQWzx}ni*@EYCw%SCch$(R!!z2P^${!4-rrnW4^_f5ucZzl6s51pc32=0|zF$V?{wV
zcRVb=!kn53aw;e!7(k9>EK)*tA}IHPFxZKR<{QL`H4IsdAP3emgYsPsQ#xY}GpJxq
zVU}coI+?|<h!^BS#v+hIA<k!kI3H9vK>Pr*T$7~;)JiW>1eFV*vgH<gacL5$x<>bf
z7|a)tdJ5)^Jdig)<q^mmT#O=&MJmW10fjT%BcR#>TUdh}AIzZ1?B@om6O;1`DpA53
z>`rDhccME_1mZksq6MYq;?$B7kmEqsgB(|+hU_qqp&<W)L%0cAk~K3zLby^562{^T
zwahgPSxmK{_78Inb2DQKLkdGHlO#heYdAxm6$?WNOEy<g03!nk*05xO)zq>^GUUmz
zFqE)na~0))WEdGx<d`BDO4zd4vp8y)YFI&@hsGMS-z`>fe$!+uk_AN{CpgM(vE(J@
zrru&JNGwXtE4js-o?3E?%M;WPaV;v!FS^AQmY7qT3T8th7M{s8+0a5QiW8LK;tLW>
zGNRZr;*0Y0OTadvho&@2XoB=Wi-6+Pk_u3uf~plzpfWLXF|sj=G4g?-0HXk7kuGvD
zgE9^XgM%3q6DbVA44RC7U<)-FK^Z6`KP45+g%IHU%>%JNza%3wFTE9HAt>uHFc#^9
zf`zRZl;B_)3bp-I4Dvj*7%B$k0dOx$4<>@6WW!acX);3c4l1R5i#<L*B|kYnzDN+%
zWMT!|0S@~pK1e%XFC#UlAhoC%9D(2#IJm6@u1&$E(JgkUlfXsdEtcf`qSRa5zTke8
zBdAk#iyhjIg))qdpbRsUTP&G*B^I|>GxJIe&A^ng5tuSFxy4!vmSY1^hGw_Gw6PI{
zHZ!@!nVg?nkds<rW^#)Q$}}{x0GATCxIo#yI6fsaxddDqfpaoA*MQ>`9QL>PGO}~P
zMIBTjxJ*V6;Mn7^$<0qG%}KQb#YM3T0|NsKBL^b~GY>NdKbJ5U50?}hIN7i<VuxJJ
S9IPB{94s8X94t(XFbDu|4<v8^
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_pandas.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_pandas.cpython-36.pyc
deleted file mode 100644
index 2df2075b66902547e8435e4a81eba6a175408411..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 249
zcmXr!<>i`^S{>`hz`*dB0SPcMFfceUFfbGgFfcHrFr+Z%Fhnt?Gej|^Fa|SdGQ9*T
z_0wdIVk=0@OGzxg#Z-`Ti!V7RvA8(CC^bE^xFoe`B|{Ml0|SKkRi$5`Ur>~vm6}|l
zpPZFdoNQvGpOl%GuAiHuTbx>!T9lcmpOKlIos+7clbNJnP+5|ZpJ%LRreB;{lB!#f
zn4F!Mo>~l5qHC;YqGzmIP-(1Zrk9$Y4i(nV$xlp4Eh+}n@eqgU6;$5hu*uC&Da}c>
MV+45!<TDOt0Q#v$)c^nh
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_python.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_python.cpython-36.pyc
deleted file mode 100755
index ada924cd472a1993704ccdda86ab632d87e62aa2..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 4349
zcmXr!<>hkDTM?5g$iVQJ0SSmPFfceUFfbG|F)}cuFhnt=Fs3l&Fy=BvF)@PJ%sI?a
z%&E*-ELp56EZIy&ZmDdk?9Gf(9I0%noXw0;?5S+2T+NJ8+^K9S%*~8ZJb64Rj47-+
zytzD4JWvz(qWDu}Q`l12TNtARQsh!NQaD=}qXbjrQ@B#NTNtB+QWR2nQg~Y!ql8oV
zQutdKqC`>zQUqHVqC``KQiNL=qQp`}Qbb!AqQq0gQp8&rq9jrzQY2d#q9jwKQlwiL
zqNGx#Qw36GQUy|FQw37lQ-xAFQiW2vQUz1xQUz1FQw3AyQw38MQUz0aQiU_QqZCs`
zQbkk6Ql(PGQzcR*Q)K3_M=7N+2Qz3YRxRXGR>;XuOo=b3EXl~v(<`Xt;_^t$OUX$s
zDpp9#FH%S;%`M2xOIJwDO93kcv6D(Ob4qkG^AsSO6iO-!Qj7Jt{G6;59E%l7GE$3E
z6%vb5!TRB9!HRT1vXx-T<f6=y%;dxzg_3-bph8+{UUErhejdnz{4|A(%;fBxR6Q=P
zm!SC5WV|Jqo>~$QG9*4RFC{)Du_V#&B{Ks9gC^rG4!``o)R4-8)MStVNIg3P0|N^K
z1A{jxvh5fc7)lsQ7;6}s8JihvSxT5dtXkF@#w_L(#u_FuhBU?$rW)2v#v117Onr>C
zY&9(Dj5TaoEGf)33@I$Z44SM}-xOR*a|=?7z@Ey<EH2SeC{8WWQ79=b$VuhmQczG(
za4bqMwgR(2yn@R3{G==^SQzCeWu+#UfIJHdQEP?bjQrA^6osTz1(0F7Ihon1aIG2f
z=|%aa1qvD&rh0CfIjMSJuBMd&n1-35015+z%)F9(J-E)~#GIV?%oGKU%)Am!D}~Iw
zl++54`plHnypqhcN(G2B)Qc60QgaeZGRsmGa`KZCLE)~DnFo%C%#zfi#3WGAgM3lN
z4zeX)HS-n=h;fSr6a#*mthbnpONxrv7#J9Cv4A{qi!Zq-HL)ZW95KbICAU};OG=80
zHQ8^mK?33yJJeUV*rATR#g<*3Sd?CTi!}vg=q={_q^w(vDJvO@gcukYes$>==x5~T
zrs`*v7N-^!>t~b}7w0DC<r(T{l@_HYC+g?s=oY7zr50u8=|d7rd}dy8Nn%cpeokhR
zJ|t-x>zU~nXO^Vu79=KTC#I(sLzU_p>zU{o>lRcR>zV1Lrl&)N^+8zyWEqG?WQkki
zP|tyU3W_P1FU3L0ivyJ4ICvO27^N7CL>L$t;Hi+6fq?;p!Ku)Sfq|ihF^d6`;%cDj
zF@>p@xt1k`Ii0bFC5tJA#fG7T8J1GPBCIwHDQv+En(T>kj0_Cm>{zUj2uoGqkb_1l
zD4sJ+3n~@TGAT@vaN|Mo0!~Xh3TZj{i6uG;N%{FX3aKT@n(!nCjn+YxST&i7K;d4*
z4@#drkfaa~4l78K<%bFtmnMNl{WMvMI2jliz_LYLAQq?u(PS;+2C;cS1TTo-0}+BC
z0+dF<3AIQVWQjC1g+L7kn+i)Ms_<08$-@YOMWA#8PbDCmK&}O+k`i!P1xg*wOhqO&
zj9E;O^a3uS7}6QxWfUyUFoV)eo=pi$7HbJx3TrPTxLiZ%;;vy$VV}bUFWDGs*itxb
z7*aUFB^+3H3Rf_LCQqUua{2<LbgHMT;*z4wymW9`r=U@qmzkWOlB$qjqyP~g+(`<Y
zpl-22oC;1Snv78_WjU#NFBw6#5YsKz;*7+CRB)Or0tww>hkDUZld(t=q=x+#S7tG|
zkjhI=Eds?mB-Mcu+%2XI(<t8Jg4E>9#2jc^l?H3VnFen$rIbXmq!g!?KoTN2NrF?O
zBs4WbT#H;{nSjzFC>1ksa52g-@-gu+$}n;;7RjTRTkzUb2c_Hsr9)6@g_I6KbtOv{
zV+y>as`?17D-}SsGPrF3Dw)VElyXapOF)$**epopHK<D-a58~JKTBSIUMe`H_-QgB
zBASFkMHm{5AcJ6qi5n;aL6O72A;Vat2#PjPVguDdpvDgq0|SFINC~JSDq#RMbYhrl
z8EctJ7;BiC8G{)#89}ZADJue1T$;?c*fQet@>5c8u@_W=+gA`TfD;p#0H-KHaE%Bm
z6<`ht0@(s~2qO<;kqQM4xy1@@L}@Y?DS=8>We@=>5Ybb$5ZDc1wJ=A7p*TW~5=Sr>
zr<M?Of&kbF#i=DQ2SlMbKm*+YpeO-HhXk}-Z)QY9M+&1P12{^U{J<$jlc@+CFQ9^f
zpi6`yE&-RQFvrB9I7X8yjscfJgknYn>=<Zq2XjplifgoxT?0z7APf!~P#Q{M0H;iF
zjB7GN(g8T}feEA(h@2JDKo)>PjDfL88`*eJvlN8E##^9;d<{buBPir+nM;^Tm`hkn
zK=mfm9M)Qv8kQ{f8s=tD4ay45EX;mIpgyA}YZZ?TxG|)jnOCB2SH%bAfm=oDc2&Gk
zeolT~IuaMuT2i;G5`fAk=jRsWq*kcgX>vef8eH%*XXcgMVg;LYiv?uBEf$czTkKHf
zx0usYOElRafdbDQx7dqIlR!?2;slZLDVfP7QLH7&iN%<;0XTNBwM;TWAqWa$21Xu6
zE+!^M2;^YmVk|NSS3U@zWfmiQ9GnZ)?cl*uoLZu8S0oC`x}c<91WGnV3LqmGA!!>c
zrF@G$K0YNsIX*s$57LI$%Sg>BNG&P`htDnM;>zM%JY}gx#hLke@tJvP`L|ee6SMLm
zJr76?3bN@IYf@!NYH<;$t}4<4IRccZZn1(wruY^Ms8w7f3lavEis054xWWS$pWvda
z2vl?wfwFHADBBc)QY$!3fr9~DQWk-laYgDNQ*=RuK8P@2U|@*i0$T%#0ltjv9B?UG
z4CaFqG!g+$$s9Jh`6;D2sdk_OxY!5ORb}E~;$Q@IAvhR07`Z?^CN4fMel8&{9u5{b
RWai*wVZ?+SERbk}kN}LkDqR2o
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_python3.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_python3.cpython-36.pyc
deleted file mode 100644
index 53f514453416469ddfd6d8843477414904a3276f..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 5988
zcmXr!<>i`^S{=Jef`Q>N0}@bSU|?`yU|=ZLVq{=QVTfW#VN7AlVT@u-VN7ApVajEW
zVrFEh6i8uAVaZ|1WsPEGWC&--6JcRUWy|I&vSDNZ!AOQYrU-^gj#PHAtON@~Dn~X~
zQ5;B)ks*~WiXEZ?Y&K^UR|;ziTMI)JcZyO9dkRMjV-!z{atdb(R|{hlZ;DC^cM4An
zV-#PCY6@=(UkhUte+qw!Knp{ZK#E|BPzytpV2W^xNDD)hP>N`ZSPMgxaEf?}L<>Wd
zNQz{NR0~6tXo_@-ObbJlSc+_lTnj^#c#3?ALJLEbM5<&ePpVWZPpWh(PpV8RU#e^>
zU#eUxZ>oGMZ>mBnZ>nM{Z>mxXQ!`^Ke<oLya;iY8V5(56aH>SANUC_MSc>8tjwqEB
zreFq5wW@_&$_hF8i7D{~l_eSZd3ptvTwETBc_}%mMa2qf`9%sTrMU%}dFcv?c`0Cp
zAa+t|W=@H2W}X5>lR`;lL29ucm!FfBf@85lNk(dMszPE>Dp)^UEm)BbNVXCznOu}v
zl9`;Cqfn9$5>!Yl%}XxH%+CW^ke{ZIk(r#Gld8wX^%4}%ewvK81k+PX;z6dwC+4NZ
zrzDmnYBJv9bgf8EEdc3DW&|l@U|?WkU|`^6U|?_tB?u)328K$Z8ip)}6h=vg6edZA
zTE-g2U~q&pGNdpEgRv${q9!8)gMvqDPC;rB+$kWZ6{VJx7Uh9m2=#4oNfF46AeVrB
zm|2pVTda_or;w<SlUZD%$Hk?fprGKFUy@pE1!jV{0hPuIP>mU&;02kGS*(z#kd#>h
zvM;+59KtF2sl|EfB?<*OiIoZ^`S}WYnaQa+l?vsVB^e4HE^elJa1+B*74lM3Q&Lkv
zu1rixQOHd#Nd$!oh+mvosgRilb~VISBnK)Kmn0UIq^2n3r==+*7Q+opsw_z@)`7ZG
zM<KN&Sx+G(BePf`0}?8rWKs<A+%5isO5=El$>1m{0tG%Ol2{W<N{Wg#nQpOV#OLLw
zq~2mLs06E6$xy_>z`*dUTE9TQpeR2pHMvATIV-C;*~CacDKjr!KQ~9WIJGRbC^JtV
z68HK!nMwMPFgMmS(=W~}N!2Y#OwLYBPc4Qj(KXgH(KFU9s5I6y(@RZHhYIV1oSIrx
z45pze+gPul@|G|t386V%3Y1K^Kxv1QiII<yg^`W12*gch28l5+Fo2xR%D}+j4N|y-
zfq|ifp@gx9p_#Fnv6iKT3B;;pO<}BMD`Bi*1*NcN#uA1S<}{{kwj!qzX0U3gcr#-P
zb2AgF%509JgcOz())s~uRu_h5##+`I#w?Z;wi+fehBU?$_8Qhq#v117Onr>C>?O=K
zEFcro8Ee?HSW`G`7;4y37_&Kx7Symo#W-ykQW%37G`Xt2DY%s8f|GN85jbV)C={ob
z=qQww7UZNt;?c1v9Vv|#RL18gWm&;;b$${kl0c~mR0dcp6ldg@=A<YjrGgF9&B@G8
zg{Pg2`1GRu(gFpI3{yR~%$!s`Fjvz`0ZhZpPyi>-%)F9(L^4iJ%*lz*Oi|Ft%q!8f
zQpn6pNv!~>&rC_pE6Gf&RDd``y;z|rH7Btovn*91CqFq6l;c5pt|TK>A+sd4C@~3C
zSb%bxA2==EVofYg&dfw96hM&}42ncC1_p+7h8l)gzFNi-h8o5chGwRTOoc4L44RC$
zSV~LMEHoK!v4Zo1Cetk@J%cEw{Csfog%F_9<rkL?D7Po)=N8!6feZx|J8TS9+Q~(!
zi6yD=AW2XVfeWhQ)Dk@#kh61&?eq}(s@OpR60e$hiv`5E#R4i*idYyJ7&O^#F&CE<
z6$vshFx+ARnN`FKV)8*v2PrR3Ehz$dNt5Fi8zky(u|wnG7CSUpZ?R>UCl;j_-(pPx
znSF~nKPl@L4>SjW^W`nZ6o{ul$-D^Uvs>a&OF-5ldqETAO;Fe|aPTm4u(B|UFp4n=
zF!QlUF$ytrFc!%&FfhP#4aj9649+#6y19lCR0e}mbqzHCq%id|*RrHAr!&^DWHF_%
z*f5kZm$1O{Rtjq~6Ih(hh9QMLm_d^xQ4Ul_gAyaC;DnS63SfUh6A38YWSACIDx_sn
zn8FibnE(`O;Iyfukd~95SfZnll%JmiDjYQjW8#OWC{3m!0R{$!B1ur3@h~tjK!PhC
z99-aJRm2TSQ2bEw;?g9rl%FPRkqAg5DCL2Z7dYV+34szNYmq2O3X}wj#6c_x5FrC1
zz^MmJfRm6kGzmdX2OAB`3E&g~Qp&)|!w7;!^2iB?oq>S?o^a~G#VshoG&2>M)G%f-
zr7(h$P#R+jQw>8pBfMyYB_n1~GRku+VW?p!VQFS6sw!d4Vk==uVeN$!tO)(wHLNM@
zbC{q-D_aeyDCMwWNa0LptYOOno0Gy7%%I7WD2SZwK$(K-$qiDIf{RoIjncf#<ouLW
zh5RCDT|Br`8aSO5gJKL^V1m;Ys2DBFNzHo+Y8GfR-C`}yNGwPNCq{7TT+9Yamf&`j
zpC)6GA}DdO-{Q(F&de(=Nz6-5g(Ny~LIjx%PJXwTGE8sr78j%@XC~%A60#;c&J=lz
zDW&8VOA4rv1f?}d83#_AlF+mX33TKFE(+u-P_|*<;9`_v<YVMw<YAOy<X|k)Ku)0`
z7l8_Ca6;B$U|>KLZJ=}sD$tP9B@4KCV@zSSVMt*MX3%7>`UtL=75wt^Qo&7UP_ad3
ziIrPgTmq_*!Dc~f%0XQ$ffEZP34lT{FF!98oM8MknIOduC~P%}FK2|IaS1XAR=#9|
zVi9CI1BVP_5vbCFw*o+-pn4kIHSmF!EzOKEOtp-)OpuDFmbr#Gg)y6{NGpXYg}H^H
zh6&Ut3udTcPGQMrDso6+MHVe#s$pto3}#4S2xib^^DE|OU|_IG&PgmTR!|3L2X#BR
z<sgee=^oq!0Tr6yCWr{A2~xrcX@np*M2bMAwk8uKCxDAzrXoI2k<9|?a6r;CIAwzp
zbrHzoT9C&fwp6Ktn!w;j2rO8zH9(3%jS851i%W{oTnmZ}uxmjjKG?Nb8cYy(GJ><H
zCd7$FT9{P=*oE~77iuyWft(6%Y8MHE3ILWOQ0Nqa%6~{Qk;XLTmLP_|%0Xs=0)~N+
zhmnJcgNcukgOQJ^NE?)8LFEsq84t1wguxLH%Hg02qM5Ol5ft*E2-IW*B|(s$Tdd#)
zvL+M6IUvPofqhE|><F+jSm-u`Tmed644~pdmt4D<i&IPR*(?CIxj3~1W^X%+z52-Z
zg8DBY4E8gqriS<#p4gc{Jz}umnILUuO(sZ`fD(l!GsOFd*Z`+BFab(%&@cd}ahO}W
z5pLmOEHa>qTfhmQkXwWyp#rW?V2<fWaf~4~9CM2uS~U|491*aCpsHamnvCKiV`LYB
zk|_v-qY9LYz~u>}A2{J@GD4yP9QI%W908D=fL!oR2U!40<qV8PCdkHvOa!&|z{b0v
z6?>IDDU2x$IgGW;;S6~)EDR+~*<3{?kU=l740A0rWZ<iYIg6zRGu5zyQ_U^*<N)xH
z%`IWy{FKt1RKNTZxBSw)6xX7n{GwZI1rSMAP{<T(V&rvmoOvCh5i;O23*<vkxd_VZ
zOpH8?e2jdIMV82Z1(hM7atrKNGc>=}Fk~@;QhhCR2@|-l!J5J}hozRKh9!%wh8f)V
zW`z{!n#_Jha^O-4rHstXD^a&YDdoYv0(HA$PzeMpNkLr%b-Q9v6B;H2>L=!;R;b%)
zvV(&Q+_b*MoS9d0ixq6dEf$cjTkKGUx0usYOElRaF%Qp#x7dqIlR#aPC{7R=pOTqe
z62)4QoLF1}PAU)roIE5!0Sz7dfHan1>2NV95JANnC=j`rm>5BjgPDu52-Ho}<UsKX
zsJV;gV{o~rZif;s#i^ih0fjrbL?{Nu5pvrfslf+kLL#*olrTXlSd$UlCcsXq-eQlB
zPsvY?kH5uSoLLdY2N@pHgA7X*gM;%HS1N2Au($|R=-uM>%q_?-DgmdyB2YTG#p#!r
zn+oRNVgdCMi$J5skopJIa|QQ8zzsccn*m&Z6oCrzB2cLcF3yTT#Yhn-=YsPWIN^Z<
z4P1d0ftrFvdZ2_1DnY>&Vv!YyYYif7K!h!b0HuptT%ZUn2FKtnzKrY~aAzFChr}W{
zr-KP_d~(?2g0qz!NT@iBfq{XAiGz`ciHT8wk&ls$i3wEeF>*0+Fmv#8DRJ?033Bmp
b@NlreAu|UjhX4yB6f!|+MjjSW^dm3;w1JAU
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_scipy.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/__pycache__/load_scipy.cpython-36.pyc
deleted file mode 100644
index aff088cfb6e7c7c16bc2fd5a3ae5dd05f77bdd41..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 3032
zcmXr!<>i`^S{?hDkAdMa0}|k1U|?`yU|=Xd#=yXk!jQt4!w?0b8KW3e7*d#W7;~AT
zm>5A~%sI@tEKw|s3@I!rtSt;ttSMqCY$@z5j8SYU94VYF3{mW<j9E-s94TC>Y^m%i
z+^MYTY+w=26dp7Yt`uH`NIG*AcRE89PlZE@xC=w9MHFv^QHq2ML#$R5Uy4X7Zz^9J
zlNdu3e=0+kK=FoD{w%>1(PqXd{#2eUp;Z26Mn;Ch3E>QdZQ%?ljKK_=k}p9X^wVU#
z#a3L9SX7*<$#_dJJ+&mhq_QA2J~1yPJ|(dvF_{shih+TFm4ShQg@J*=859gY3=9k<
z3?(3MFsCr4Gd43eGuARfMOfe>Oi&S4xCnCzLk&YSV+~^sQw?((b2eKoOASjfgC<ke
z3kBzl)Z}c1%ru2Wg@VeGjQl)>{G_bZ<PwF<VueJ7;^fSNN(G2Z6%vbz5-Yj56ciK`
z9E;M6t-vf0ub?tMKPk%!&47}8g=Dbd<r%3Z8L34InJ`1a9!EAVD7B=tD6bf9U}kZA
zUTJPYrGiFMetwRol>$_OLP$|*D#)?ejM7m^%_&Y*a7zS{nQ0K~^3+SfrWS!hRFmx%
z3n-j#v4Z_o#KgeBaEmLsxF|k1v7{)o0>n;+vy<TBnyk0jAbyNuEs0MqE`m_WAS$W2
zD2g+)I35y4D;bKo7#J9SRq7Y$7Zl}#0#ZLYE2}u!#7I9WGcR2~H%GTPwJfzLGfzJw
zGdVjaRX-;)NgopO#(HM@#hE3kx&?{J*@@|?#ZV=>#(E}t#<~TS#(HLYsp;uZVf~!^
z#FW&cVlW*K_K{ve<t=G&pvQxp1d2SU!{Wh_#|}zHAO|vV3NeBqQxOjX14A-1G`)f3
zKxxh!q;Li!149jC7DEbS4U-r{Gb1R~WHF|Qr!dtpWHDtj)-X?J>SL^BtzpSxOpySK
zGQ&mLz@n01Q5Lu;J6Kc-EXoQO<tSmRVJKlQ;mG1lVS}XWW~QR16!u!qJTtf)7n0n(
z6pmWXJSn&wH<H|r6wX@C5}qvH35-QTDO@$2VhrhwH7r?tDcm*;DLha<E0oU*<+DNg
zd{90+l+O?4b3picO(py#0x1F@(PqXJ!Dc3e*<cwV6d9;3U>RW)8K`|=84(m2sLfy*
zQ4|@d8&brA88oF6B^enQ6jDla3yKvI^HB1rLP@?thG{{iLRw}{DoGg@WSq4^aYlY=
zPKrWOszRayBp-rPk3xQtLSBJhZlWH9tC5>noSmAY3C>`kav&o<y(qu5KtUtJRL?Cl
zCshy3)wEIo(;!!Y%vS)Fa0;1uCHZ=AoymzgIq{h(3L2StC7M<WnRzLx6(IGQDXDoS
znQ4{KLP@<?p(r&cu_UuBRUs!oIk6-&KTjbuPoX3uRUxw^wJ0$uClz976+6h5c-72X
zEFi`$_RPGL%;ePKTWpzmDFr1(w^)la5(`qn#lkJ-<l-U_k$j6esko?^lYxOjwHR6s
z6oHE0Vm@Rca1mWBh#~|vzE}WVw4kJ@m>)$DY;hGQ+=g4MiN(p8nN@sn5wHbSf(Sm;
z^eO=qp@Nd4Dt?3r*jztN#v)z@1_n*WB9NOk8H+%cYcdvr9HuFFizm4#HL)ZW5>dDK
zpgd3z6{nVfb%OPQb%FKV;>s+}%quQQ%u7xMhXuIQEMj9|U?>6=Cq<xI`WAC>X2mV$
zfJmcTtcfKhMa8#RlPXJ6i*GR(mlSDo6@fzZ7CSU~++v5u_AR#T^2DO_;wat>Xek9w
zC)ivO#Q`d}<BLm*z~w81;06`nqEIhD%3pBE6{nU+fXXi&P@%@b$0!DYOk#{|j1bJi
z#K9;4g+feHj1X9)04fSW6$Mx$wE6{Cy&J$)FQ^iYVX9@U1r>mdwahilSxhO6&7igj
zLo1UcLoEwfgb7K66)eJxB*Iq1n8K3HRFqZ1oW)YZRKrrkn#Nqil*L-ZmcrW0T+5#4
zRKu9Umd#YuQ^K6ZhE;_^4Py#>HdE1>66P#+Y$|HlgBdhA{LtbOoDh?XVO1PBur!%&
zv8I3m7aZGA6|f2stO6{Cs0_hknjA%-D&rPgL1IyAUdb)y^wbha!T_fUMNsMhClOHU
z(B!zqmJy$qpOSiuy`U1@(gDXYNVF(FzXZ*dQ7pxv>KIfNA;q&esJ4S96KK{1yTl3<
z)1c^PVB%q9gH~!Bj8cpcEWuc$4pITi1&E3stkVtbF|^<Si=lb82$a=}gg|~01`(i)
zg~%XPJcul)T3jRo$$X$xq{)n?Q<E9ZtD3?^Ag33Bs$6jJMe#ve1$r5&IR&Xj#o$l|
zSJB|a3HA$m-s4NoNh~goFG@|%EG|hcx+RcWk(^VSk_s1z;>b;eG%%ysK)EwBDT+NK
zJ10InwX#SOl!w?-^O8Za!<JZ3keZhQ33asa=dj7mPbtkwwFCLA7?e{u7&#bun0S~t
Z1i6$rL_w_yF=iHKE=Dd!P(*Ms0svE~7pDLK
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_astropy.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_astropy.py
deleted file mode 100755
index dd8efce6..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_astropy.py
+++ /dev/null
@@ -1,237 +0,0 @@
-import numpy as np
-from astropy.units import Quantity
-from astropy.coordinates import Angle, SkyCoord
-from astropy.constants import Constant, EMConstant
-from astropy.table import Table
-from astropy.time import Time
-
-from hickle.helpers import get_type_and_data
-import six
-
-def create_astropy_quantity(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an astropy quantity
-
- Args:
- py_obj: python object to dump; should be a python type (int, float, bool etc)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- # kwarg compression etc does not work on scalars
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj.value,
- dtype='float64') #, **kwargs)
- d.attrs["type"] = [b'astropy_quantity']
- if six.PY3:
- unit = bytes(str(py_obj.unit), 'ascii')
- else:
- unit = str(py_obj.unit)
- d.attrs['unit'] = [unit]
-
-def create_astropy_angle(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an astropy quantity
-
- Args:
- py_obj: python object to dump; should be a python type (int, float, bool etc)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- # kwarg compression etc does not work on scalars
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj.value,
- dtype='float64') #, **kwargs)
- d.attrs["type"] = [b'astropy_angle']
- if six.PY3:
- unit = str(py_obj.unit).encode('ascii')
- else:
- unit = str(py_obj.unit)
- d.attrs['unit'] = [unit]
-
-def create_astropy_skycoord(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an astropy quantity
-
- Args:
- py_obj: python object to dump; should be a python type (int, float, bool etc)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- # kwarg compression etc does not work on scalars
- lat = py_obj.data.lat.value
- lon = py_obj.data.lon.value
- dd = np.column_stack((lon, lat))
-
- d = h_group.create_dataset('data_%i' % call_id, data=dd,
- dtype='float64') #, **kwargs)
- d.attrs["type"] = [b'astropy_skycoord']
- if six.PY3:
- lon_unit = str(py_obj.data.lon.unit).encode('ascii')
- lat_unit = str(py_obj.data.lat.unit).encode('ascii')
- else:
- lon_unit = str(py_obj.data.lon.unit)
- lat_unit = str(py_obj.data.lat.unit)
- d.attrs['lon_unit'] = [lon_unit]
- d.attrs['lat_unit'] = [lat_unit]
-
-def create_astropy_time(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an astropy Time object
-
- Args:
- py_obj: python object to dump; should be a python type (int, float, bool etc)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
-
- # kwarg compression etc does not work on scalars
- data = py_obj.value
- dtype = str(py_obj.value.dtype)
-
- # Need to catch string times
- if '<U' in dtype:
- dtype = dtype.replace('<U', '|S')
- print(dtype)
- data = []
- for item in py_obj.value:
- data.append(str(item).encode('ascii'))
-
- d = h_group.create_dataset('data_%i' % call_id, data=data, dtype=dtype) #, **kwargs)
- d.attrs["type"] = [b'astropy_time']
- if six.PY2:
- fmt = str(py_obj.format)
- scale = str(py_obj.scale)
- else:
- fmt = str(py_obj.format).encode('ascii')
- scale = str(py_obj.scale).encode('ascii')
- d.attrs['format'] = [fmt]
- d.attrs['scale'] = [scale]
-
-def create_astropy_constant(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an astropy constant
-
- Args:
- py_obj: python object to dump; should be a python type (int, float, bool etc)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- # kwarg compression etc does not work on scalars
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj.value,
- dtype='float64') #, **kwargs)
- d.attrs["type"] = [b'astropy_constant']
- d.attrs["unit"] = [str(py_obj.unit)]
- d.attrs["abbrev"] = [str(py_obj.abbrev)]
- d.attrs["name"] = [str(py_obj.name)]
- d.attrs["reference"] = [str(py_obj.reference)]
- d.attrs["uncertainty"] = [py_obj.uncertainty]
-
- if py_obj.system:
- d.attrs["system"] = [py_obj.system]
-
-
-def create_astropy_table(py_obj, h_group, call_id=0, **kwargs):
- """ Dump an astropy Table
-
- Args:
- py_obj: python object to dump; should be a python type (int, float, bool etc)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- data = py_obj.as_array()
- d = h_group.create_dataset('data_%i' % call_id, data=data, dtype=data.dtype, **kwargs)
- d.attrs['type'] = [b'astropy_table']
-
- if six.PY3:
- colnames = [bytes(cn, 'ascii') for cn in py_obj.colnames]
- else:
- colnames = py_obj.colnames
- d.attrs['colnames'] = colnames
- for key, value in py_obj.meta.items():
- d.attrs[key] = value
-
-
-def load_astropy_quantity_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- unit = h_node.attrs["unit"][0]
- q = Quantity(data, unit)
- return q
-
-def load_astropy_time_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- if six.PY3:
- fmt = h_node.attrs["format"][0].decode('ascii')
- scale = h_node.attrs["scale"][0].decode('ascii')
- else:
- fmt = h_node.attrs["format"][0]
- scale = h_node.attrs["scale"][0]
- q = Time(data, format=fmt, scale=scale)
- return q
-
-def load_astropy_angle_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- unit = h_node.attrs["unit"][0]
- q = Angle(data, unit)
- return q
-
-def load_astropy_skycoord_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- lon_unit = h_node.attrs["lon_unit"][0]
- lat_unit = h_node.attrs["lat_unit"][0]
- q = SkyCoord(data[:,0], data[:, 1], unit=(lon_unit, lat_unit))
- return q
-
-def load_astropy_constant_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- unit = h_node.attrs["unit"][0]
- abbrev = h_node.attrs["abbrev"][0]
- name = h_node.attrs["name"][0]
- ref = h_node.attrs["reference"][0]
- unc = h_node.attrs["uncertainty"][0]
-
- system = None
- if "system" in h_node.attrs.keys():
- system = h_node.attrs["system"][0]
-
- c = Constant(abbrev, name, data, unit, unc, ref, system)
- return c
-
-def load_astropy_table(h_node):
- py_type, data = get_type_and_data(h_node)
- metadata = dict(h_node.attrs.items())
- metadata.pop('type')
- metadata.pop('colnames')
-
- if six.PY3:
- colnames = [cn.decode('ascii') for cn in h_node.attrs["colnames"]]
- else:
- colnames = h_node.attrs["colnames"]
-
- t = Table(data, names=colnames, meta=metadata)
- return t
-
-def check_is_astropy_table(py_obj):
- return isinstance(py_obj, Table)
-
-def check_is_astropy_quantity_array(py_obj):
- if isinstance(py_obj, Quantity) or isinstance(py_obj, Time) or \
- isinstance(py_obj, Angle) or isinstance(py_obj, SkyCoord):
- if py_obj.isscalar:
- return False
- else:
- return True
- else:
- return False
-
-
-#####################
-# Lookup dictionary #
-#####################
-
-class_register = [
- [Quantity, b'astropy_quantity', create_astropy_quantity, load_astropy_quantity_dataset,
- True, check_is_astropy_quantity_array],
- [Time, b'astropy_time', create_astropy_time, load_astropy_time_dataset,
- True, check_is_astropy_quantity_array],
- [Angle, b'astropy_angle', create_astropy_angle, load_astropy_angle_dataset,
- True, check_is_astropy_quantity_array],
- [SkyCoord, b'astropy_skycoord', create_astropy_skycoord, load_astropy_skycoord_dataset,
- True, check_is_astropy_quantity_array],
- [Constant, b'astropy_constant', create_astropy_constant, load_astropy_constant_dataset,
- True, None],
- [Table, b'astropy_table', create_astropy_table, load_astropy_table,
- True, check_is_astropy_table]
-]
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_numpy.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_numpy.py
deleted file mode 100755
index 7a31b12e..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_numpy.py
+++ /dev/null
@@ -1,145 +0,0 @@
-# encoding: utf-8
-"""
-# load_numpy.py
-
-Utilities and dump / load handlers for handling numpy and scipy arrays
-
-"""
-import six
-import numpy as np
-
-
-from hickle.helpers import get_type_and_data
-
-
-def check_is_numpy_array(py_obj):
- """ Check if a python object is a numpy array (masked or regular)
-
- Args:
- py_obj: python object to check whether it is a numpy array
-
- Returns
- is_numpy (bool): Returns True if it is a numpy array, else False if it isn't
- """
-
- is_numpy = type(py_obj) in (type(np.array([1])), type(np.ma.array([1])))
-
- return is_numpy
-
-
-def create_np_scalar_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an np dtype object to h5py file
-
- Args:
- py_obj: python object to dump; should be a numpy scalar, e.g. np.float16(1)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
-
- # DO NOT PASS KWARGS TO SCALAR DATASETS!
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj) # **kwargs)
- d.attrs["type"] = [b'np_scalar']
-
- if six.PY2:
- d.attrs["np_dtype"] = str(d.dtype)
- else:
- d.attrs["np_dtype"] = bytes(str(d.dtype), 'ascii')
-
-
-def create_np_dtype(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an np dtype object to h5py file
-
- Args:
- py_obj: python object to dump; should be a numpy scalar, e.g. np.float16(1)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- d = h_group.create_dataset('data_%i' % call_id, data=[str(py_obj)])
- d.attrs["type"] = [b'np_dtype']
-
-
-def create_np_array_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an ndarray object to h5py file
-
- Args:
- py_obj: python object to dump; should be a numpy array or np.ma.array (masked)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- if isinstance(py_obj, type(np.ma.array([1]))):
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj, **kwargs)
- #m = h_group.create_dataset('mask_%i' % call_id, data=py_obj.mask, **kwargs)
- m = h_group.create_dataset('data_%i_mask' % call_id, data=py_obj.mask, **kwargs)
- d.attrs["type"] = [b'ndarray_masked_data']
- m.attrs["type"] = [b'ndarray_masked_mask']
- else:
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj, **kwargs)
- d.attrs["type"] = [b'ndarray']
-
-
-
-
-#######################
-## Lookup dictionary ##
-#######################
-
-types_dict = {
- np.ndarray: create_np_array_dataset,
- np.ma.core.MaskedArray: create_np_array_dataset,
- np.float16: create_np_scalar_dataset,
- np.float32: create_np_scalar_dataset,
- np.float64: create_np_scalar_dataset,
- np.int8: create_np_scalar_dataset,
- np.int16: create_np_scalar_dataset,
- np.int32: create_np_scalar_dataset,
- np.int64: create_np_scalar_dataset,
- np.uint8: create_np_scalar_dataset,
- np.uint16: create_np_scalar_dataset,
- np.uint32: create_np_scalar_dataset,
- np.uint64: create_np_scalar_dataset,
- np.complex64: create_np_scalar_dataset,
- np.complex128: create_np_scalar_dataset,
- np.dtype: create_np_dtype
-}
-
-def load_np_dtype_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- data = np.dtype(data[0])
- return data
-
-def load_np_scalar_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- subtype = h_node.attrs["np_dtype"]
- data = np.array([data], dtype=subtype)[0]
- return data
-
-def load_ndarray_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- return np.array(data, copy=False)
-
-def load_ndarray_masked_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- try:
- mask_path = h_node.name + "_mask"
- h_root = h_node.parent
- mask = h_root.get(mask_path)[:]
- except IndexError:
- mask = h_root.get(mask_path)
- except ValueError:
- mask = h_root.get(mask_path)
- data = np.ma.array(data, mask=mask)
- return data
-
-def load_nothing(h_hode):
- pass
-
-hkl_types_dict = {
- b"np_dtype" : load_np_dtype_dataset,
- b"np_scalar" : load_np_scalar_dataset,
- b"ndarray" : load_ndarray_dataset,
- b"numpy.ndarray" : load_ndarray_dataset,
- b"ndarray_masked_data" : load_ndarray_masked_dataset,
- b"ndarray_masked_mask" : load_nothing # Loaded autormatically
-}
-
-
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_pandas.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_pandas.py
deleted file mode 100755
index 0b518553..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_pandas.py
+++ /dev/null
@@ -1,4 +0,0 @@
-import pandas as pd
-
-# TODO: populate with classes to load
-class_register = []
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_python.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_python.py
deleted file mode 100755
index 58de921e..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_python.py
+++ /dev/null
@@ -1,141 +0,0 @@
-# encoding: utf-8
-"""
-# load_python.py
-
-Handlers for dumping and loading built-in python types.
-NB: As these are for built-in types, they are critical to the functioning of hickle.
-
-"""
-
-from hickle.helpers import get_type_and_data
-
-import sys
-if sys.version_info.major == 3:
- unicode = type(str)
- str = type(bytes)
- long = type(int)
- NoneType = type(None)
-else:
- from types import NoneType
-
-import h5py as h5
-
-def create_listlike_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Dumper for list, set, tuple
-
- Args:
- py_obj: python object to dump; should be list-like
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- dtype = str(type(py_obj))
- obj = list(py_obj)
- d = h_group.create_dataset('data_%i' % call_id, data=obj, **kwargs)
- d.attrs["type"] = [dtype]
-
-
-def create_python_dtype_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps a python dtype object to h5py file
-
- Args:
- py_obj: python object to dump; should be a python type (int, float, bool etc)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- # kwarg compression etc does not work on scalars
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj,
- dtype=type(py_obj)) #, **kwargs)
- d.attrs["type"] = ['python_dtype']
- d.attrs['python_subdtype'] = str(type(py_obj))
-
-
-def create_stringlike_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps a list object to h5py file
-
- Args:
- py_obj: python object to dump; should be string-like (unicode or string)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- if isinstance(py_obj, str):
- d = h_group.create_dataset('data_%i' % call_id, data=[py_obj], **kwargs)
- d.attrs["type"] = ['string']
- else:
- dt = h5.special_dtype(vlen=unicode)
- dset = h_group.create_dataset('data_%i' % call_id, shape=(1, ), dtype=dt, **kwargs)
- dset[0] = py_obj
- dset.attrs['type'] = ['unicode']
-
-
-def create_none_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Dump None type to file
-
- Args:
- py_obj: python object to dump; must be None object
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- d = h_group.create_dataset('data_%i' % call_id, data=[0], **kwargs)
- d.attrs["type"] = ['none']
-
-
-def load_list_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- return list(data)
-
-def load_tuple_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- return tuple(data)
-
-def load_set_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- return set(data)
-
-def load_string_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- return str(data[0])
-
-def load_unicode_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- return unicode(data[0])
-
-def load_none_dataset(h_node):
- return None
-
-def load_python_dtype_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- subtype = h_node.attrs["python_subdtype"]
- type_dict = {
- "<type 'int'>": int,
- "<type 'float'>": float,
- "<type 'long'>": long,
- "<type 'bool'>": bool,
- "<type 'complex'>": complex
- }
- tcast = type_dict.get(subtype)
- return tcast(data)
-
-types_dict = {
- list: create_listlike_dataset,
- tuple: create_listlike_dataset,
- set: create_listlike_dataset,
- str: create_stringlike_dataset,
- unicode: create_stringlike_dataset,
- int: create_python_dtype_dataset,
- float: create_python_dtype_dataset,
- long: create_python_dtype_dataset,
- bool: create_python_dtype_dataset,
- complex: create_python_dtype_dataset,
- NoneType: create_none_dataset,
-}
-
-hkl_types_dict = {
- "<type 'list'>" : load_list_dataset,
- "<type 'tuple'>" : load_tuple_dataset,
- "<type 'set'>" : load_set_dataset,
- "python_dtype" : load_python_dtype_dataset,
- "string" : load_string_dataset,
- "unicode" : load_unicode_dataset,
- "none" : load_none_dataset
-}
-
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_python3.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_python3.py
deleted file mode 100755
index c6b173fd..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_python3.py
+++ /dev/null
@@ -1,201 +0,0 @@
-# encoding: utf-8
-"""
-# load_python.py
-
-Handlers for dumping and loading built-in python types.
-NB: As these are for built-in types, they are critical to the functioning of hickle.
-
-"""
-
-import six
-from hickle.helpers import get_type_and_data
-
-try:
- from exceptions import Exception
-except ImportError:
- pass # above imports will fail in python3
-
-try:
- ModuleNotFoundError # This fails on Py3.5 and below
-except NameError:
- ModuleNotFoundError = ImportError
-
-import h5py as h5
-
-
-def get_py3_string_type(h_node):
- """ Helper function to return the python string type for items in a list.
-
- Notes:
- Py3 string handling is a bit funky and doesn't play too nicely with HDF5.
- We needed to add metadata to say if the strings in a list started off as
- bytes, string, etc. This helper loads
-
- """
- try:
- py_type = h_node.attrs["py3_string_type"][0]
- return py_type
- except:
- return None
-
-def create_listlike_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Dumper for list, set, tuple
-
- Args:
- py_obj: python object to dump; should be list-like
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- dtype = str(type(py_obj))
- obj = list(py_obj)
-
- # h5py does not handle Py3 'str' objects well. Need to catch this
- # Only need to check first element as this method
- # is only called if all elements have same dtype
- py3_str_type = None
- if type(obj[0]) in (str, bytes):
- py3_str_type = bytes(str(type(obj[0])), 'ascii')
-
- if type(obj[0]) is str:
- #print(py3_str_type)
- #print(obj, "HERE")
- obj = [bytes(oo, 'utf8') for oo in obj]
- #print(obj, "HERE")
-
-
- d = h_group.create_dataset('data_%i' % call_id, data=obj, **kwargs)
- d.attrs["type"] = [bytes(dtype, 'ascii')]
-
- # Need to add some metadata to aid in unpickling if it's a string type
- if py3_str_type is not None:
- d.attrs["py3_string_type"] = [py3_str_type]
-
-
-
-def create_python_dtype_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps a python dtype object to h5py file
-
- Args:
- py_obj: python object to dump; should be a python type (int, float, bool etc)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- # kwarg compression etc does not work on scalars
- d = h_group.create_dataset('data_%i' % call_id, data=py_obj,
- dtype=type(py_obj)) #, **kwargs)
- d.attrs["type"] = [b'python_dtype']
- d.attrs['python_subdtype'] = bytes(str(type(py_obj)), 'ascii')
-
-
-def create_stringlike_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps a list object to h5py file
-
- Args:
- py_obj: python object to dump; should be string-like (unicode or string)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- if isinstance(py_obj, bytes):
- d = h_group.create_dataset('data_%i' % call_id, data=[py_obj], **kwargs)
- d.attrs["type"] = [b'bytes']
- elif isinstance(py_obj, str):
- dt = h5.special_dtype(vlen=str)
- dset = h_group.create_dataset('data_%i' % call_id, shape=(1, ), dtype=dt, **kwargs)
- dset[0] = py_obj
- dset.attrs['type'] = [b'string']
-
-def create_none_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ Dump None type to file
-
- Args:
- py_obj: python object to dump; must be None object
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- d = h_group.create_dataset('data_%i' % call_id, data=[0], **kwargs)
- d.attrs["type"] = [b'none']
-
-
-def load_list_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- py3_str_type = get_py3_string_type(h_node)
-
- if py3_str_type == b"<class 'bytes'>":
- # Yuck. Convert numpy._bytes -> str -> bytes
- return [bytes(str(item, 'utf8'), 'utf8') for item in data]
- if py3_str_type == b"<class 'str'>":
- return [str(item, 'utf8') for item in data]
- else:
- return list(data)
-
-def load_tuple_dataset(h_node):
- data = load_list_dataset(h_node)
- return tuple(data)
-
-def load_set_dataset(h_node):
- data = load_list_dataset(h_node)
- return set(data)
-
-def load_bytes_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- return bytes(data[0])
-
-def load_string_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- return str(data[0])
-
-def load_unicode_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- return unicode(data[0])
-
-def load_none_dataset(h_node):
- return None
-
-def load_pickled_data(h_node):
- py_type, data = get_type_and_data(h_node)
- try:
- import cPickle as pickle
- except ModuleNotFoundError:
- import pickle
- return pickle.loads(data[0])
-
-
-def load_python_dtype_dataset(h_node):
- py_type, data = get_type_and_data(h_node)
- subtype = h_node.attrs["python_subdtype"]
- type_dict = {
- b"<class 'int'>": int,
- b"<class 'float'>": float,
- b"<class 'bool'>": bool,
- b"<class 'complex'>": complex
- }
-
- tcast = type_dict.get(subtype)
- return tcast(data)
-
-
-
-types_dict = {
- list: create_listlike_dataset,
- tuple: create_listlike_dataset,
- set: create_listlike_dataset,
- bytes: create_stringlike_dataset,
- str: create_stringlike_dataset,
- #bytearray: create_stringlike_dataset,
- int: create_python_dtype_dataset,
- float: create_python_dtype_dataset,
- bool: create_python_dtype_dataset,
- complex: create_python_dtype_dataset,
- type(None): create_none_dataset,
-}
-
-hkl_types_dict = {
- b"<class 'list'>" : load_list_dataset,
- b"<class 'tuple'>" : load_tuple_dataset,
- b"<class 'set'>" : load_set_dataset,
- b"bytes" : load_bytes_dataset,
- b"python_dtype" : load_python_dtype_dataset,
- b"string" : load_string_dataset,
- b"pickle" : load_pickled_data,
- b"none" : load_none_dataset,
-}
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_scipy.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_scipy.py
deleted file mode 100755
index ab09fe23..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/loaders/load_scipy.py
+++ /dev/null
@@ -1,92 +0,0 @@
-import six
-import scipy
-from scipy import sparse
-
-from hickle.helpers import get_type_and_data
-
-def check_is_scipy_sparse_array(py_obj):
- """ Check if a python object is a scipy sparse array
-
- Args:
- py_obj: python object to check whether it is a sparse array
-
- Returns
- is_numpy (bool): Returns True if it is a sparse array, else False if it isn't
- """
- t_csr = type(scipy.sparse.csr_matrix([0]))
- t_csc = type(scipy.sparse.csc_matrix([0]))
- t_bsr = type(scipy.sparse.bsr_matrix([0]))
- is_sparse = type(py_obj) in (t_csr, t_csc, t_bsr)
-
- return is_sparse
-
-
-def create_sparse_dataset(py_obj, h_group, call_id=0, **kwargs):
- """ dumps an sparse array to h5py file
-
- Args:
- py_obj: python object to dump; should be a numpy array or np.ma.array (masked)
- h_group (h5.File.group): group to dump data into.
- call_id (int): index to identify object's relative location in the iterable.
- """
- h_sparsegroup = h_group.create_group('data_%i' % call_id)
- data = h_sparsegroup.create_dataset('data', data=py_obj.data, **kwargs)
- indices = h_sparsegroup.create_dataset('indices', data=py_obj.indices, **kwargs)
- indptr = h_sparsegroup.create_dataset('indptr', data=py_obj.indptr, **kwargs)
- shape = h_sparsegroup.create_dataset('shape', data=py_obj.shape, **kwargs)
-
- if isinstance(py_obj, type(sparse.csr_matrix([0]))):
- type_str = 'csr'
- elif isinstance(py_obj, type(sparse.csc_matrix([0]))):
- type_str = 'csc'
- elif isinstance(py_obj, type(sparse.bsr_matrix([0]))):
- type_str = 'bsr'
-
- if six.PY2:
- h_sparsegroup.attrs["type"] = [b'%s_matrix' % type_str]
- data.attrs["type"] = [b"%s_matrix_data" % type_str]
- indices.attrs["type"] = [b"%s_matrix_indices" % type_str]
- indptr.attrs["type"] = [b"%s_matrix_indptr" % type_str]
- shape.attrs["type"] = [b"%s_matrix_shape" % type_str]
- else:
- h_sparsegroup.attrs["type"] = [bytes(str('%s_matrix' % type_str), 'ascii')]
- data.attrs["type"] = [bytes(str("%s_matrix_data" % type_str), 'ascii')]
- indices.attrs["type"] = [bytes(str("%s_matrix_indices" % type_str), 'ascii')]
- indptr.attrs["type"] = [bytes(str("%s_matrix_indptr" % type_str), 'ascii')]
- shape.attrs["type"] = [bytes(str("%s_matrix_shape" % type_str), 'ascii')]
-
-def load_sparse_matrix_data(h_node):
-
- py_type, data = get_type_and_data(h_node)
- h_root = h_node.parent
- indices = h_root.get('indices')[:]
- indptr = h_root.get('indptr')[:]
- shape = h_root.get('shape')[:]
-
- if py_type == b'csc_matrix_data':
- smat = sparse.csc_matrix((data, indices, indptr), dtype=data.dtype, shape=shape)
- elif py_type == b'csr_matrix_data':
- smat = sparse.csr_matrix((data, indices, indptr), dtype=data.dtype, shape=shape)
- elif py_type == b'bsr_matrix_data':
- smat = sparse.bsr_matrix((data, indices, indptr), dtype=data.dtype, shape=shape)
- return smat
-
-
-
-
-
-class_register = [
- [scipy.sparse.csr_matrix, b'csr_matrix_data', create_sparse_dataset, load_sparse_matrix_data, False, check_is_scipy_sparse_array],
- [scipy.sparse.csc_matrix, b'csc_matrix_data', create_sparse_dataset, load_sparse_matrix_data, False, check_is_scipy_sparse_array],
- [scipy.sparse.bsr_matrix, b'bsr_matrix_data', create_sparse_dataset, load_sparse_matrix_data, False, check_is_scipy_sparse_array],
-]
-
-exclude_register = []
-
-# Need to ignore things like csc_matrix_indices which are loaded automatically
-for mat_type in ('csr', 'csc', 'bsr'):
- for attrib in ('indices', 'indptr', 'shape'):
- hkl_key = "%s_matrix_%s" % (mat_type, attrib)
- if not six.PY2:
- hkl_key = hkl_key.encode('ascii')
- exclude_register.append(hkl_key)
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/lookup.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/lookup.py
deleted file mode 100755
index 99d13df9..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/hickle/lookup.py
+++ /dev/null
@@ -1,238 +0,0 @@
-"""
-#lookup.py
-
-This file contains all the mappings between hickle/HDF5 metadata and python types.
-There are four dictionaries and one set that are populated here:
-
-1) types_dict
-types_dict: mapping between python types and dataset creation functions, e.g.
- types_dict = {
- list: create_listlike_dataset,
- int: create_python_dtype_dataset,
- np.ndarray: create_np_array_dataset
- }
-
-2) hkl_types_dict
-hkl_types_dict: mapping between hickle metadata and dataset loading functions, e.g.
- hkl_types_dict = {
- "<type 'list'>" : load_list_dataset,
- "<type 'tuple'>" : load_tuple_dataset
- }
-
-3) container_types_dict
-container_types_dict: mapping required to convert the PyContainer object in hickle.py
- back into the required native type. PyContainer is required as
- some iterable types are immutable (do not have an append() function).
- Here is an example:
- container_types_dict = {
- "<type 'list'>": list,
- "<type 'tuple'>": tuple
- }
-
-4) container_key_types_dict
-container_key_types_dict: mapping specifically for converting hickled dict data back into
- a dictionary with the same key type. While python dictionary keys
- can be any hashable object, in HDF5 a unicode/string is required
- for a dataset name. Example:
- container_key_types_dict = {
- "<type 'str'>": str,
- "<type 'unicode'>": unicode
- }
-
-5) types_not_to_sort
-type_not_to_sort is a list of hickle type attributes that may be hierarchical,
-but don't require sorting by integer index.
-
-## Extending hickle to add support for other classes and types
-
-The process to add new load/dump capabilities is as follows:
-
-1) Create a file called load_[newstuff].py in loaders/
-2) In the load_[newstuff].py file, define your create_dataset and load_dataset functions,
- along with all required mapping dictionaries.
-3) Add an import call here, and populate the lookup dictionaries with update() calls:
- # Add loaders for [newstuff]
- try:
- from .loaders.load_[newstuff[ import types_dict as ns_types_dict
- from .loaders.load_[newstuff[ import hkl_types_dict as ns_hkl_types_dict
- types_dict.update(ns_types_dict)
- hkl_types_dict.update(ns_hkl_types_dict)
- ... (Add container_types_dict etc if required)
- except ImportError:
- raise
-"""
-
-import six
-from ast import literal_eval
-
-def return_first(x):
- """ Return first element of a list """
- return x[0]
-
-def load_nothing(h_hode):
- pass
-
-types_dict = {}
-
-hkl_types_dict = {}
-
-types_not_to_sort = [b'dict', b'csr_matrix', b'csc_matrix', b'bsr_matrix']
-
-container_types_dict = {
- b"<type 'list'>": list,
- b"<type 'tuple'>": tuple,
- b"<type 'set'>": set,
- b"<class 'list'>": list,
- b"<class 'tuple'>": tuple,
- b"<class 'set'>": set,
- b"csr_matrix": return_first,
- b"csc_matrix": return_first,
- b"bsr_matrix": return_first
- }
-
-# Technically, any hashable object can be used, for now sticking with built-in types
-container_key_types_dict = {
- b"<type 'str'>": literal_eval,
- b"<type 'float'>": float,
- b"<type 'bool'>": bool,
- b"<type 'int'>": int,
- b"<type 'complex'>": complex,
- b"<type 'tuple'>": literal_eval,
- b"<class 'str'>": literal_eval,
- b"<class 'float'>": float,
- b"<class 'bool'>": bool,
- b"<class 'int'>": int,
- b"<class 'complex'>": complex,
- b"<class 'tuple'>": literal_eval
- }
-
-if six.PY2:
- container_key_types_dict[b"<type 'unicode'>"] = literal_eval
- container_key_types_dict[b"<type 'long'>"] = long
-
-# Add loaders for built-in python types
-if six.PY2:
- from .loaders.load_python import types_dict as py_types_dict
- from .loaders.load_python import hkl_types_dict as py_hkl_types_dict
-else:
- from .loaders.load_python3 import types_dict as py_types_dict
- from .loaders.load_python3 import hkl_types_dict as py_hkl_types_dict
-
-types_dict.update(py_types_dict)
-hkl_types_dict.update(py_hkl_types_dict)
-
-# Add loaders for numpy types
-from .loaders.load_numpy import types_dict as np_types_dict
-from .loaders.load_numpy import hkl_types_dict as np_hkl_types_dict
-from .loaders.load_numpy import check_is_numpy_array
-types_dict.update(np_types_dict)
-hkl_types_dict.update(np_hkl_types_dict)
-
-#######################
-## ND-ARRAY checking ##
-#######################
-
-ndarray_like_check_fns = [
- check_is_numpy_array
-]
-
-def check_is_ndarray_like(py_obj):
- is_ndarray_like = False
- for ii, check_fn in enumerate(ndarray_like_check_fns):
- is_ndarray_like = check_fn(py_obj)
- if is_ndarray_like:
- break
- return is_ndarray_like
-
-
-
-
-#######################
-## loading optional ##
-#######################
-
-def register_class(myclass_type, hkl_str, dump_function, load_function,
- to_sort=True, ndarray_check_fn=None):
- """ Register a new hickle class.
-
- Args:
- myclass_type type(class): type of class
- dump_function (function def): function to write data to HDF5
- load_function (function def): function to load data from HDF5
- is_iterable (bool): Is the item iterable?
- hkl_str (str): String to write to HDF5 file to describe class
- to_sort (bool): If the item is iterable, does it require sorting?
- ndarray_check_fn (function def): function to use to check if
-
- """
- types_dict.update({myclass_type: dump_function})
- hkl_types_dict.update({hkl_str: load_function})
- if to_sort == False:
- types_not_to_sort.append(hkl_str)
- if ndarray_check_fn is not None:
- ndarray_like_check_fns.append(ndarray_check_fn)
-
-def register_class_list(class_list):
- """ Register multiple classes in a list
-
- Args:
- class_list (list): A list, where each item is an argument to
- the register_class() function.
-
- Notes: This just runs the code:
- for item in mylist:
- register_class(*item)
- """
- for class_item in class_list:
- register_class(*class_item)
-
-def register_class_exclude(hkl_str_to_ignore):
- """ Tell loading funciton to ignore any HDF5 dataset with attribute 'type=XYZ'
-
- Args:
- hkl_str_to_ignore (str): attribute type=string to ignore and exclude from loading.
- """
- hkl_types_dict[hkl_str_to_ignore] = load_nothing
-
-def register_exclude_list(exclude_list):
- """ Ignore HDF5 datasets with attribute type='XYZ' from loading
-
- ArgsL
- exclude_list (list): List of strings, which correspond to hdf5/hickle
- type= attributes not to load.
- """
- for hkl_str in exclude_list:
- register_class_exclude(hkl_str)
-
-########################
-## Scipy sparse array ##
-########################
-
-try:
- from .loaders.load_scipy import class_register, exclude_register
- register_class_list(class_register)
- register_exclude_list(exclude_register)
-except ImportError:
- pass
-except NameError:
- pass
-
-####################
-## Astropy stuff ##
-####################
-
-try:
- from .loaders.load_astropy import class_register
- register_class_list(class_register)
-except ImportError:
- pass
-
-##################
-## Pandas stuff ##
-##################
-
-try:
- from .loaders.load_pandas import class_register
- register_class_list(class_register)
-except ImportError:
- pass
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__init__.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__init__.py
deleted file mode 100755
index e69de29b..00000000
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/__init__.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/__init__.cpython-36.pyc
deleted file mode 100755
index 86d97d222a8780ca54c672085ff548dfa6a28be0..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 209
zcmXr!<>hkDTM@&+z`*brM8Gg30|SEt0|P@b3j+f~3PUi1CZpd<h9ZzKg7{UZU!b3n
zpPQ<mRa%@{RIHy-T3no)n3rd$pH*6vnw+Sgo1<HtT9#UrnWvwTnVg-I8lRb0T#}fR
zqo0$Rq+d{3l98WhtY@ZQoLQ2pTacKXotU0l3{|RYtY@NUtXoiNtY@Z|nx3v-l3H9+
gtREkrnU`4-AFo$Xd5gm)H$SB`C)Ez*u40he0o-RgO8@`>
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_astropy.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_astropy.cpython-36.pyc
deleted file mode 100755
index 8c672a706793402cf52bedb977e67f6a6b91ef51..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 3798
zcmXr!<>hkDTM-k%$H4HI0SPcMFfceUFfbH9V_;xNVMt-jVTfW(VMt-hVajEWVrB%1
zG3T)4vPQ9j*(^D1x$IHwxg1d(U_NUOXD(M1S1xxHcP>v94_J&Xhc}lmiVw_Y%;Ap`
zNMT4}&k@WOiV|XENEONwZf3|5X=Y$#Na0A~Y+;BJP2o!6ZefTLOW{f3ZDEKKPvJ}9
zZ()d%ND)X8Y+;C!Oc6>EZefU$N)bsBZDEL#P7zBHZ()d%NtI2J$Yv^flqw1`Ema&$
zN`Og8FewElrNN{Oh)iJ&X3&&;2?_u|O~zXsfu)IgC7C6anvAztLNaqxHJNU)I_9P4
zq~78P&aQOM&o4^RWWL4WoS#=*l9*R=i_6s)&SrGiWW2>1l9-f}3S#8sCnx44gG>ft
zW(EcZRt5$JXHeL7F)%Pxw4|`RFvQl>GL$f+u+%W5u=X<5GL|rAG1V}pur)I=GL$f8
zF{iLKGuATIFxD_-GZkBuurx8&FlMpVFlDi3GZlM4_%)1{3<Pu*K~%zZvVr)8Y~c(k
z48aVV9Dcu;bK{JEvFFAaE9ff}|6<Kf*DpwZ$<DyQ01AlGyv&j+u9DQ^lK8{|y$n-L
z<|1we28JRY5Wx#l#8w1KH@BEGvU6^+q?G0s++xYePfWSR=U7~vT2zvmpXXXslwVZD
z4^qs0i>b8q7GvTq#-x=DMdAz$48NN63-mMcb5r%RN{dsAiuE%}i;Hs;^YRS!vr3Cn
zlN0rGb99SS%TkLn^Yk+^le2SD<1;}qkdvdIlbNJnP+5|ZpJ%LRreB;{lB!#fn4F!M
zo>~l5s%xxgqGzmIP-(1Zrk9$Yt`G7_F_?@`EG{X^FR0WjsJtZ#5{3!H7eW&P$a}?N
zpcrK0V&q}sVdP@sVH9BEVB}%sV&q{gl4M|DfJZVT0|NsnqQQ}@#lXN&!jQ#S!j!@Y
zj!uRw<`l+e##)9FmJ+6BMn;AbmKp{S4UPmRKa?P5EfN5Qt{{lu01-kA3=Fqe3yL!H
zN;Daxn9|e1AqOEOKyp$c{x1IUD4s}$CQ~_3q63-20CF}H2V;>mva>;P2Tz-~z|Kx#
zOkqlAN?}f6Ne9t9DeNg6nT#o%z09?sbjg&$TEmdZn8F427$}``H8ZBLN`lfM)KehO
zWwC<1SHe=m2&3U53|VY7j9K7xUc->Z3QE=4OvQC2>`ja{P`PZT;yw_+gd>X+ls_aH
zvba*Xn;C1sHbL0UOvRg!)UY9`fwP;Lim!pxfP<OW?<FXfUV{Am5)@Z2SwNvH3L-$!
ze2XP7F*h{|lG;l$GV{_wMF2>w_!diUYDpq;EJd-W<>V)p7#nFa6~Rr=<Sya_nTCiz
zP}CJkfy9&;7#N~hQ%WieQj3H^k;ar)aEl``Cnq^4zc@9DEwLCRrO9-QvE&w0i4i1r
zK{=!d9K}K?v0DNvX;eW`3zB4Dlw%TL<YVGs<YMGt<YMAu6krr!QeiAoM~`ZdhruZ{
z03OvT%<0T2EZ`yt9KAIRXhjfX2}>3$EMl6OifupvTEmdVQNx(Uk<C=>1LA`NnBA{P
z2IN&)5CKYfw-|FZ8NoRpIp{Pw!9@mADhG#>GDw#SC}dd65_3u+p#;{99K`71B8?I*
zu&f)OSX7i)X$}e?E>H$$6ar^o5ym1=)`VwTP;kIgzaAq410?N(%H<jcP!338?PbPQ
zz9OY<aFzz8<198%`JBa;%~Tu!3YH3;5{?qi8pdYEW=0o=SeaU8kg6JHNd}NQ5S`6b
z90yWa7#Ge^!<fZY!<5AZOZ_#BS==>DS=`x7#q&UFKshmmGaZx<8EP0(xWLgyuKPZK
zOf2Ea;sv{iDT}q4X#!(08%Px4jtPv#5>OGSOC~TDYk@?-5zg&bC2MGDX{l>qsB2&t
zVqjneMtX)u#wMm_<`$M!!bS#$258bzESbgmCAXMMOOiF2Zn3517v(0F++r<GPRvQI
z;)cX$fstN@Y37W_3=EkwLBy=b3=C1sxmhWid_|zFi=KGESxgh0c#1O;VOfl|C^0WR
z^%iqZYF?2JD0%6E2zWx%WVyv!l9`)Ye2b~Xut*K0n<+CBJ%x!wTm-2t;z8;msSDgx
z0wpH~Mh-?kXz9wqD8eYjC<D%bTqvSqj6#e>Cde5Q<PUi217*kxuM{>HhFB+1`b%N1
zVMt*?Y8SA=%Uf1F?E=)=o!##iTTx<4W@6qgwv^QLqSVy%pdE6puLD2WW6PIJMW&$e
zF$2}qY(=2Vfga4Df(w!*nTtd~YT-p6dZ?k6SBao@R1_#Xf*KACp!yS(Awjh$D3ii7
zB&cWx*-OVVstD{|tfdslCFmglPdr7S)CMleApwF^4WNgG1WF15hXo`Ff&5*JRzO)G
zClQbfKo}ezPvB)Xq<ZdU0w)yK6gD`Yv4pXPp@y-Usg|jPDT}#=DTN)JY?!iGQrMdr
zYne+}!9^5%7Do+p7JC*)GgEOy31<@{RGbSe&IJ}P0E<J5#1sy2Vgxl<DYfqe*j$J^
zknMW_5(gJ6oPJTPhABn{xnRmL7deWe*o}=-3~~*Mqu4+U!(8N=C5kg6zZ8_`b5b?A
zAqg0i!EnamEvBMGL_FSNPDxGHWGS))C2T8jbzPL0lA0U^%0sEi7*#TAtFbt{G8xoI
z$N{BtP&LQ^Dg;0&oso-Cf>DA=fKdR>E3yU^HXQNsxrv#1@$s5+x7Z+UTd-53cwuc*
zQ1iPOTw6x*z{Ef$5V#DC5{3zZEzZnKEJ-a!_&Q1uAq{FU=am!}v4Ok>YTFlqnuG{@
z!0ice+vOH(UTJPYWf3TyN3nyN$B88wMW8eq#gm>|5?_*<TTqmrUvi5zIU^;r2;2w;
zHz~o54RG}eF6D|q`ML;{nBjf}wWQ+XkunFOb;Dtko1apelWGTQpcNZ1Fff2x**uH_
nj3SIYj2w(S%sfmSf*kA|LL8hNQXHHdoE(B2jBHG7j4%iQX4gY>
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_hickle.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_hickle.cpython-36.pyc
deleted file mode 100755
index 046d30d0527994b989f6c055be5a3cb573b6163f..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 18861
zcmXr!<>hkDTM^^u$H4HI0ST}%FfceUFfbGwFf%ZuFhnt=Fs3l&Fh((gX{IP9FwGpr
z0;XA`Siv+~6dRakk77?@OkvL9$mNLQU<B!9$>Gf9isAyZnR2+JxKkKYSaWzn7*e^i
zc$*or_?j6Q8B*9%*i$&tnNv7ZxKg;&nNxV?FcBrjJBJBH58oW-DE@SYD1j9I6oD3o
zD8UrL6rmP|D4`VL6p<E&DB%>*6tNbDD3KKL6p0pwDA5$j6sZ=5D6tgj6qy!=DDf28
z6uB0LD2WvL6onRsD9IGX6r~o1D5(_X6qOc+DCrc{6txzHD47)X6pa>!DA^Rv6s;D9
zD7lo36rB{^7RD&~6ulJv7KSK=6oVAQ7KSLr6r&X57KSLL6q6Lw7KSM06tfib7KSL5
z6pIwg7KSL*6sr{L7KSLb6q^*=7KSMG6uT7r7KSK|6o(YY7KSLz6sHvD7KSLT6qgj&
z7KSM86t@)j7KSLD6ps|o7KSL@6t5KT7KSLj6rU8|7KSMO6u%Vz7KSK;lz^1L7KSLp
zl%SO07KSLJl#rCr7KSL}l(3ZW7KSL3l!%nb7KSL(l&F;G7KSLZl$ey*7KSMEl(>}m
z7KSK`l!TPT7KSLxl%$m87KSLRl$4az7KSM6RGXBvY^Dj!MLMa5ptwu529uUx(i}{x
zgGn_osR|}#!K4V76a|xFU{V51N`gr#FewfuWx%8om{bOnDqvC(Olp8hO)x13CfTxV
zComPANEHI}gu$dBn6v<sY*}^_n2N5YT7h}mU{VK6>Vio<FsTnF4ZtMCZ$@CYF_<&~
zlcr$O3`|OcN&A%aW=2Ma6sBMX&CDtzE@g$1)Z&u(jLhWhoK(GnN-nO@yv!1?V6j44
zevtx1Mj<ypr8Fm1kBjRiDEs+oGTve<C@9LzE74@U#hR0!oS37@c#Bc%B_jg^!%HR*
z!3-i;Km;p@0A*E8rdu32sb#4-M)8JGFqRQq)EF*m9B-t_3{_zW7d4DGg6cPnH%<n*
z1B5|VfUq;D9Ef3HU?^e8Vl0YCVU1y`WvFFLVXkE=VX9%OVW?qjW@cn4VXk3lW~^l{
zVX0wiW-MW;VQyw*VPIy+X3BRcVa;aBV_{-oWGMD3VQXTnVX9%yW-5*VvGW*GSb`Ze
z*{b9eTuO5bGV{_E67y0Na`F>Xz-)!$k|Ge}7dt3G^)gIvF_u?phh(HG6qaTtXDcKX
z<(KCvq~%vAWR>O?6f5MHr4}iaWTYzOBvw@_q~xdjX)+b@F)%O`@iQ<mXtLj8F3zmD
z#T*c6bc-dWG`HXu3&^TlEG3l%skfMmONwssITjbE7L{b?=eZUY<ritP+~P>f%t_5l
z%uT(;lAE8BdW#F<ruh7%EI2zOJ7*<Bku(DX!>>mD0{x8q+*JLn(&E&jV*QNL;^N%I
zygWnwtkR;?<V5}49Nps7vecr?Jbg$qh|kO`E=kPE(a*_D(l4ki$;i($)-%&D&MZmQ
zEl5nxPE1cNhAP!H)-%yF)-9+s)-%&fO;6Vc1xqoQM5MM`+#u0-h%*g95hwtPQYJP=
z4n`J69!3sk6-GX$B3T9o2JG=&0gLaF6b?vymoTL;H#3$nrLZ(JiZhfjrLaP2HV|FQ
zRKkpu7FeKZffbw<Sdr2~ObHu2Eu@vO!_z_uh@Hof!XC_^$yud_BQ2EXWhUpRq++BA
zaNNEGWueSZ+zbqvE&L1&nFZnu44}xDgGW9`5hxgn1Q{3@Zn1&P$|x!l0?7)42oaDl
zJ4iS`CACNllv-Gd#6dC=AQ{HeDAv;WjO-k6Btr-pkOU7ntfBhsQ3G3nk%y6kNrjP*
ztw<R;ut7N)lzqXOD-IUeQ7MeI3?)o83@J=0%qc9r;B1w`)XZ4R2+C&JOvO4SEKQ6h
ztf1g6VXa|=XE8UVU=0APVF$@FGUPF&Fa$Gbvieo=KwStfgQ_?{j@ASD2AnMzi$H$V
zL<%6JfB^+X5h!$u6hWZ@a=RweEyl!Kj8%|e0HyjOa9HqyjE#rdWE2Pw3`RCa7DhfM
z4n`hE6-F+mA~oc|02N`}3=9n5z*xh;z>vZNNpqk;NMQxF0n%A(nMxQSDJ+Ep959Ta
zK&WM|^aGVbpp=P}$~;P#;i)VL6dceJCl160wH0z$YguYove>|8u-33<v4cwSaE3e`
z7KRd*Y_6hZj0_-H!jZxa(hDl#VEWP+niwM)@|YqRQn-Q{G<mCJaHQ6p%;FM-Uy$Sh
zV(Vp?f)i>HDCUbmzAj=11t<rI-~^>NNHi3I3VXjQ7T2I4D@~>%b&xbTmN=0j1r#5*
zSU|=Wf#T;DONMDd<t-Ms%$(F)tjRg~#i_SgAr%Q5QX=L6X#kg75Edu_XoAdF0TJMM
z1QX!M<OD}1NJlIvB0*vdOdO1SjC_n7%uI|bptQ}%#mK|N2VyZ5=_5xfsOo@6>NHTK
zazY}tgaJ8Pn;C1FN<b9?tQuhgM+tMK38+A6W-MVwidL@@26(iFfT9u<HQ7wX2_SYj
zL!KTq@+N?yj*+2+6=DXcVg|FBn;AjuCI)EqaRf7HauJ9=Xi1t{f*O0EC<4b`I;db}
zU=VUhPXhxr2T+*<Ds#bct;u$axj3~1E#kmYR-^-pa-<@rNEZ|ddLRNE319*g>bJnT
zHNH5tq!<(tp!fzwgb*m$!4V<A2#Nw8rXnNcC;+7})F|MBL;)-U(phSmD)efYTo_{2
zYMDXxdkT9Ea|%Z<Q!PsgQw<BK&;g}<Xx+|QC>737saC_91<E@t*-XV@C9F-1B@9_?
zHLNu(X-v(G#Tg)xaE3fJ7KRemY_1|$RIt~uf`Yz;y@mxu!=r#Rm_d`fN)c-g$}7z+
zs8mQSDoU(WfLH(EP`$;TmXn`YVr&Fun3>$-OwP|O$Vsg*Gr7eDWf~e;yaeSfP*W-e
zRKJ5_#uya5;M~M<iz%<*7E69!YB5suSc2q1A$*HDF((HSN60yd4Qx{JEvA%`Tby9G
zfwL2o3Cd64Xaf`AxZ?unBCuatL9r(VN-&_3kckTtYm7{c0*qpyOvT8-SmcNtgD9D3
z0w@!4Lt+pVS6Sdn0#rVxvw(U|Da<JxDIDo6DV)8awi%ZCxKgc#Ig7Q1DT_6msW<@C
z`YK__Vy|JYVFJfp5=aCZcWl{QMP1Nr!~x2Jj0`0lpxT^~0Umdt(vPP~QXwR@xI`g0
zu{b+5MFH$jtZ|05)H4BvHYhJ?G8KVB{uT?!98f$tfn>n(#93qos%DvT6K}DC0~}jy
z71@B)W8^4Mu@=Reh^^X&WNeUisVUPz5d(^JP_|NH0@c|lk>Z9NDWKAu2UcAPF=DH(
zQaDmL)1es(!fs|v;hMvSy~$j0riLktrJ2EnA@)EmOJzn4OARxKpUqaB0Lo4!Y&A?Z
zERqb(j3sQKh)-cpX9SgG*&IbPz!3<RtzibK0Y~Hvkghz25;lk!NFU6EY^LHZAlYz+
zLN7=boxoVc0gq^K<}BfW>4QW%$XrmSE#bgsVlxvYNfdfTFcf-)Go)~XD@9dEvPj8H
z2DPLUiz*dBt!+?C9=UmsnyrgLLF<y3msj}`ly#sbs)R!bTH4WMyTt-(jz=*krl;Ow
z%Sg>k&nSsvNy#tDzQvMPnp+&jl9yOi1WHEkpk#rRjEY=A0`NxPEtc%m%3?@y460F)
zk`XIXwaJu{a*Hh`1(b|#F=nGTdLZ>A$fqkn=|~Hb!I_yDA&`SngcX#c_&^z*Nrq8|
zMS)R(5!4joVU%Jl3P4U<ptcn#(}SBr8lVi%0?Y834DjL@-1Nz0$Yv^5DPd}2OkoDK
z@T#;SA)cCBP*SM?a!qD_9&zCg3Mf!p4P3TrG9%?LP@d3aLWBgm*ZIMP8`#Qtkd<dZ
z0RXCZ7??Pico>U<kUa+~i{bUrF_7o@AfAKOM=9)}b{;6hq;RD(*D_ZGf$Ju(T9yiz
z8fF)USesf_a0cM6VM*cXWvXQ>VX9#R1xpQU3U4n9xEIP=!ji?B1#a1al4c1zxIw{Q
zXb{d&s0OaT*t0ll*s?gXnTp#=IKlN7R}Fg&8#L>1HZelWK+bHgq65&Z!Cb=*&H~Ie
zY#<t527;Op{8h4$pa<1EMXANbnfZAN`FRQ{i6x1kIvX{S-(t<oD={`I0vCOzpwwsv
zBFq^W82oOrq*rAY++xnDO3P$uU|`T>1ec1MjJLRv%+_Qo3I=HbCu?qS=~!e95&$P_
zP#JiODZls@OF?2u#x3^r)RN-Ns?;KI9|u&%LQ*@Zywv1CidkDwA<GJ~rntxs#AZrP
zE&_FDia;ZQ=n*dfj(Cu2sE4nBVqO81^qDyr`IxwvxEMk8n*gH_qW~!3BWgHMV-kC>
z6O?mO*pbu39417&AKVfFC;eLH6t-XnO^zx}NVtQ>Q$VQ?oZvH4ixtW<OEMHP^GZ_F
zQ;QTpIUK$Hzyfl96f1}{(qsZRxWQ!%C?|lrW00J{gvbe+$n7*=P+<Fk2yitICO`$&
zEkTfh;EWKTnO72@om%-C6f9hzU||BanK_tw7>g2+GX*HKz;isPEPw?-2?O?)aSd|{
zvm`?bizGt|t0Y4$OAQkWpB2;;%VsL}DPeA6q%fF@ctL&yW!WN7g$D|pM34wL;j!G}
z_HzvmadnAz@pKMBgcLY=X(Bb5ir~Q&#g>x;Dnp}~axi+)kOT#a#rV9`;*!*qAE1x}
z4cvf=5*BcsEWjwhSOiKsAPfpq22h0oavHe2oCm6lc_GOMR7}7cNa@U=E@TROI=E?D
z!T=Kob?DMrn;BEM!O09ugJ}X|g$Jm)1ge`SFveQdvQ$0-*UqfX4B2eOMW6;#2|KKA
zEn$b44X>S7l`v#+Kx82`D!7PQ0n(LMfu@fU<SLlX35-QbC7ckQSzMq3r-Tz?0+h#)
z1?rZvNHSz`!Nr*-Fc!0b%q~I5@<3&InnAKkU|CRu57L@|yF42t>j07sXDEzfVJKnC
z=9<7*GzD7taY7pHFgHNm18#Ld*d<uZ1QouZVzMwQf}t=foFRoLm_d`TN)3`_K@p9b
zT_G(oNYk+xEx&^EW)Uc>fcsn_ZkmimpeQc_RVhW_t~1ju7EpP6ixor~fs$$pJgM?R
z%19TG0BZeT<O31|)pC#l4p6BGZu1wVf{H+n#GD*(Um9$iCQ_3Ck*JIOK}LYf05AbA
z2q38$;w5l-z{beHPz)-X8JIbkr5GicK`jCyaKR|UD8r}#?p4b%$}uT1A+-%M&`SzX
zh6H)M_!Fr5WJ9ENP|7P|$O5%FSxOkPK-DCutq4ky93>1{oFxodT%dGO!jQ#N!jQ$A
z##_r&;R3E+ZEBf8WlA$+ElUX>Qm2m}Tye72u+)IM3q>UYO^h`xS%NjJS%TS2#cm*e
zp<6gZCPOVtg$&pv5wJ-t&5SiHB|@N1gAGIBgm4CIrj`gp(o}^8Sf@fQGfa;pLk%lL
zw<M^Aa|&$Fsc?oA)?fxrc34#lD)Ngn^U_Om5{ne_3KSr{GKF8<c?I#{z9C9?9ok9%
zU=K<cnV|Rr7hxjcbP)yVOd6VDtCFKwGV@9-qCj=Fp;=J~0|Nt?ZDtb1S_+n715t)%
zQDEBG2tu2gpp}Q<YP?7kWHTrcAQz8OTwu4ygT`6F6)#fK0w*Xi0Zvtr(KVP4H9$!U
zG>`@=AbA)$7zG$bn7BYaOK{^-go%Zbi%|qLM!>|u$i-Nc1)45kU|?iO2DLIl#V4rV
z0hR8cDTQKCeFiEun?dysXbh6EnW<l|ld(drgt?ueoiU9GG%DG`k;2jp8Zh9iWd`+H
zK(WloP^cBo0E(eNh9HIr21tns9@JM-aL&&yC@o1<@O3d&C@#$f&wV5+q-U0;<|%-N
z)fK9ki?oyA1v<!CAm@QvE5)FCvx6~<p@X5BF_=M<5i~u*Qk0sQ0&b0HGTq`#%E?d8
z1~qSPF{Xh9A;XXyHaUs8NhyhTo}e%SnOP-~n_^mAnyY66o>VEe(}O9w#o(vOc8fhD
zu{a|qGwBv{Zi?wGmi&U$yjv`pC8<TX*h&jh5=&BVab~1eq-3V27MH-<0=b|ff(4}U
z7Ax2xU>9gI6@es*@<5Rdjy5)k)h-~nfV>2WFDXU|#-d741_${U<VH|*fGRXlfl>_W
zxq_kt+`46|VNQc|&A{_5X-qXt^O$Ow`@o@6rKAv?SeB}gSd?Cxo0?ZrtdO6kkeHJL
z?r<q2C+6f7D`+y`;&KV~4TyJkboOw)#g<r5keZhg#SNW)0CmkYS#B}sCuM<Kgtu4{
zi_(j4v1ONoXt1{+1lT`3$%)AsnR)5)AZ>miUx8fCz$C<2R1fkG$S)ua8r%f=2V7vC
zU|?WKXQ*YU2q*z{bUk8NY8h*pYME<UK*N6}Oj*p$Ohr1N*`q90(4b%nV+~6RlMSd7
z3mP6|PGhcN>0@MMsA0%rhm1sks-8lwaE21jEH2R81S3O5QVDkrLkR;ofri8|)v{u>
zWfH=c61Ex?(;+6;vX$`Eu-34FsxT0p%~YII!rR1Hm=g|}tzxaRQV2^e%1o<ND9K1H
zQ79=&OwNuk$;?eHR<KoYOUx-wRVXORPbp1KEmp`(NzE(COis)Jr}tt7w<vBT_28<_
zPm`~x3Y40-GK;}eL3zobel2+NFpAg16f~+EUIdD?B5)ycixZS};z502aF7>OgAyAk
zG2LPTx#AXUaZYM#!7b*h%mPSt2CCvT*`wG%sWr8@s0O4SKDSX+2QJq@a)wc?Aj$|5
z-H4(GQlTNaI|LN(ps5uGCKg7}NF*1y1QKH6Vv=FxVdMgLak&_a+K^KOsMG}&IN%g<
z2vmiMK&miYJwVV*eL7PuOA2oaUpiAQD`-YPMKGPYmaU=*Tya#@vR9OV`-VBS92IFb
zEG`VOakZQk5jCtX46y;VToqn5Y%UD3PPN=M95w7UoHblE+$lnHSZjGo7;1S-7;AXJ
zIgBGk7~Fqks^KYN&SJ>|PqBdFzXaTqtmP}r2xlma3uh<{0S%pmGZZ?6Go*lef8td(
zkR}aMX9&^(DbCMLRe)q6aIsgMQJR*PlM0^SOwK6H%La{xpmq+yH4L_T$ge0A6lGzc
zG91*fdHIWhfg!{V(Q^T}jx<?|!a>SF-Mm|DU^|L$u|w^R;x0~3%t_5pODj$-(PRXx
z)MP4Z2WbFBj3!f22Z#+KG?|JzL2MAA$zGHMV#9lvMG+uwB#4Lt5umbKlOJ4gfTxj)
zKwZhAXpmGFC}D9x9Z(#_gX988FS7_VvwVvwIpY>nF$m{_(_#@3{mytu$Yz4lA!ww8
z0n+be<Y44rg!Dc^Jy0$t$RG?8BL_1JBd9l8)Q_AZL2XJ<=><-aX`mDd?SF#GUQnq5
z>VIZ|XFWj`UJWxSHI^_VHIpE7ovbCS&=GF78rBl_W~O4l5{@Ru61EyvP&W<K1Z!q0
zP6CNRJ7pTMRFhW*nr#Nndn<t20HD$vX%rQ`DV)i`!oZOEf|-E<9QU9G4QOB((qZKS
zH)z0v0k>GIQj7ADN47x?kD>-ptbl^62vj26V$Vwf52+$INQyvJ&n?EJTa3v?phf|D
zQ2-h90Xa1u?9O&j41f{`12YG>*AE)`;b7um;$tkDiZiT1O<P1)H;WNlSl2M6f$9fh
zLOTEyvS^{50urNoXio!$AUMAug?7<IkSHh+Amc3Xu)f7wQd*Fc3J&d}NuUWI<eo7o
zq>Cnlv`zsL*g_jx@PZ711ouqz;N}ANjlsb^139=s$r^;gNgFhRodO+81TWoyPNpIc
zsz4@E89_79%%I8@H1}4*0*aVsreY`1s4;Y;5jy1Wm!Fra0Gd~UEM~zh%|MBBCdgyp
z;9^5+6T`y?((MAJMOdo}G!YF>f>;wDWUMtWKQDDPC>+4!1I!#u;PC-a)@Lf3g)<D0
zM^_*f4!GY>Tu3mbut_qMuz;37G=oN0SeqEZ3s*Q`6CeabqX^U%0tE>o)<IFP$y79(
zx?#kJ7=;1_)Gp+J0w)$oK+Qo8C{PywUNeIV3GhljhFVa2ixCu9%}lk7pjH}a-A64G
zs0Eb*ZXcntY8X>kQrJQB^EHfFte`pdN+FO6gm^eZo&soqA&V`GJ)5hj3OdBVk-`B=
za?puc&?+j<Dno^gl9GaAD}DX+%#w`KB)#PPT>X;NoYeH9#Da{>Wax?*{mkOx($r#o
zBLf9UkqBGdkdg*26~W^JFtO|$O~@pBl`?4Bs@O`w88la(0&bfrq*f%S7J%j}6p|B5
z(=$pmIg7xv<Dh{_aM5~;EwLyuFFh5Kl(j+e)(EQN_#vi%+G*fLL*Pc2CNo4gxF-OP
zbZ~nLOn@8GBH)MuH|0Pbj(G4;!%0wdfieVWE{X?Il`-)#axgM63NZ>X7R^JBHc<YC
zN82_~^~IRN)WQIoV{c}30WHX4C}GS3_2yW>jTy!)W>5;QWvoo705!H5K&oODVwgZp
z@fwC2CP{`YR#1}|L}xP<n}EiI3QfWpFjbd;hS9+~AnL)@8(R}&VG^h>n5O}X<ZP~@
z4bTYYgqRLf3z;(lFQNeT>)A+-Y%_&q(7Yk2p8@JWf$|n;P`ns4y_(KY!w@S}3!0f`
ztYJ)H5NBv+oXAuNT4%-x&V`yx;Mgph3o4MAZZYW@++xf_tLJ`k*?{UeP^-@F4k%7Q
z*03>DDT5;oBo&`ul#*J6+@s7aE&;2537Si)f~{Ed)8x9v3Q0Dg!~#h>phnXz&eXip
z+|(ja_mZ_ZBe4KlPk`n-i$G%*kYvPii!tRED`fVGDKqmHQ$+=uTW=xN$zVs_0y!9x
zh`?=gCQ#y$VH5%nGzl>ltwK&jpf(T)gDZ-4(8Wlg{GSe5$;SYyn^Ra)SksxC8Pk~{
z?V=PmxEy#gGMy=%sg?m$KGZOzaDbB<Qx*#-PuDV5ctI8{xiG}qK@wVpNeOsOm<vO!
zMhsIe3wTxsG>^k1$pD&cgd`FWtC^`d3$!w<Fe{v)Fd`f@2Mu1rQNjTloML210j(e4
zsxnmwPE94Tv@lTsg)jpHgNg%`VdC%-)IBd+1Zrf1x;Ef8HB|D0JyLu^rlcWriKtOp
zv;<Tpfb>ABd{9NF30fE(pI?-jUbGw}3z{V?S_vvPn6k2NF;!Ng$EGkiiKXTyrKY5$
zro@9=Xs<vq3R=F#z-$5PUon9S0}Un)#v)LR!N*TH85kJ2VUyG6pant-b2>AqB(Gtp
zVMu3&EGA1~NoTGFk0#WBS8=AZf(Gwum_aawF&(r%ilGM72WOqbTnnC5s9{cFo5Nho
zRs(9&vd>|vWv^ic_nnwPgUq0QGH4d9hCQ9T7Bu7tUMmh-laS8R3|`g-N>Lzw3MXjL
zy_PYB58Sq9&Vs814alc(r|_h6fL4;i&E`nqox@toQ3L92@%w=@FgRwwW5Kr=i#3@c
z?FCS6dW$htlNmDh0xofvL2@`_#x2Hl#AtC5D4t(}#wUs_Knhq91{8r?Vjyirst_qq
z(SooKVTdLRq!0m*C>w$z7Bs-D$%)mh4Ip`NMX(9P+6W>*=A%bFe|l;jxIlo;g8TzT
zJ18wOFoL@?;BjIePy%3-0@nvDj545EkuAs>8dOq&Ds*s$HbO}RDU6_E5tMmSKnr-}
zN*F;CprEyaDWFkaJrEzfVvr3zphMg=nW6-fGg6bY<1>rnL1SZyNzj(9CKHmU!Sx%M
z09SI51}u^e0VW29V$i@n$V)6t0*pmlksS{ziQtY0&BZ~|4XzYi!T|CNX!R1v6SXX$
zRf5fowX8MFC5+h|#ZI6J@*0*B=4_5)KQNmW?9nI?8@oq~)-y0LxM{LPNkBbYl3E1v
zDagyY@u2qZE#}O;5=j0Ab)_`fkbDnHK2dBSRq=^XItfZAqer$ZIG^LPMFJE~pe_Xi
zxL3u+1nSv>`c^!QMca`>3{($-5-B*kLCxn9hGvFZh7!ga(B2EC6lPF+4BS{{t<nRp
zv{Fb^0M%1DnaPPIphY|22{(oOq^#8B5`~g{1rHZDQ!WJs1qDR23EY%|ww*wYSWtok
z*JPTE;EaOqH%PRDMhKxPSsmmZP<w@enT@e%2eJo1j)%|U6@b%gI%6#ZD6OV2rZ8rL
z)(C>CAx6;rSS>TCngOkf?YFCCDFH1~V*xc#L7psOffQNLwX-as8MG8OIFGpov``AX
zt~Q(@g#j{J4Ne3ckkpRwZWW7uPBCbd1T@T1#iCzQkW<B?pORcs#iE~BRHVs@l=l%$
zL{MyLvfg6ODK5FiTvCv8i#a8^<Q8*cQPC|HP%{rbkRUB}&}=n$EL9H_L?AO5!0Tl>
z7<rg@z!{#8v1lK9fPsoIaEwD*>qG^ZUo9xWKwUbf8qnm54X9IB!cqh3)-csDW`jm8
zKr8w{BCO3!HB8{el`s*uX2`nzB9Jb$0A$A=fLKC~4Qt4OnnU0mrO8^f6I3|v0uj4G
z#2yf_7es&y6)*uVS&%{xw1(drTj+smNg>9f{m7vQN`#;!2M#^Z2vrF*pQNxr>naA&
zsyfh`IxwHHgaKYNf$B17t(nFQsWH=;AT+4vtYJ=Ltz|A@OyLFfW=fbK(<7jcbP6A6
z>=K+v8B+wovMeclpn49b3Y-vyAmNCuz5!QSMVmnZ0jjiKf~xwW)u6TtsKS2<%0N||
z{zaMTnR$shR*0mGoO7%Axq?!Y^Ye;Jib|79Qd6unS&P6`7pR?51gbZ{c^W*3p~-?)
ztYOXPkPbO0{5?S74=R2bm_QTt0*s&$VbGdy(0UM1*;RBHITk>c!D9im!~{~#gIk)c
zpv(`NZi4a{A*DPcLkVb^Fe4+lY-?r&r&t@LlDrsHd4MK>;|&l@Lj=<Z!89%gRsUeM
zh6b9<;Fk6+&f@$c@Ujiak}yps@ceKTH)sh-JV-7TGW7(>$4H3~BnMjVQ5*#d1W?vw
z0A*ZokzI5Y+1sG14eo8wxGK0o%?R==sDWJz-ev>hff{JAmH%m=Wia3^7vLpMOpxSL
z1Kl16PE<@OoM2Jb6gJTG31sCnBV_eI7uG~YMq)wEyVz0;QbP=DVu7rMhmJXdrXh>L
zr5!lEaDW2~l3t2HtHa=JT9hCIRS%#b%L4B~1hs6ztxJY1&;knZir92!aPDPIXHH@7
zWrDPJ!2!q&>W?rofW{*rK~)3YHwX?u<`f>VC>tmsAtM|hJ~&tMVhup}G6XV0k{wbC
zf?AA_ZUi~O3F#+5gEJno+5nV^p`j|oD8q;ns>hKNA}IfXLKa*Kg4U};^4}a#KaQb>
z0W{4E%4d+_0??)uP-v7efm&J6WW|aUDw<5-rIdzX(g;!bpsXeYSqW<@Vr%?D+6={+
zpoQGvRTGVf+yolZ2YIIG6tX8!it-Y0%K?;kAfCu#EK*2efoxSuVVlEL%b3Sf1KPUA
z4q9Re8vBEEF;ckZFxN7dFsE>Xy$IgA2HHJP!<@ncns|aH4}P#L8z`B8x4D7GNWsZN
z03{%zSQ4{SE2CIJq!Fkm2pXe+^)WddKp~X`RsbT6qgXRib8_;lI5ILp+t!LwHJPGV
zlELyI60~2BwJJNc(l82489^vxvXUIA`v}?n#FCSrmkuuTK<R~n)IuM!&JEmzOsxRN
zR5vK5z=b~}7b6>!0C@2M4|0}1iym*FvKpNDKqI!0#0M(#QdnWJm&F7s%Zk)M>sml8
zaMp!x6Heg-PuH>3fcRV~+$ek=G(In!58h0a!q*E*<RCT|n9W!MV)KC6Okg%In9U51
zU4Flp;OSLRsST-@L7f*^ItFz*5y`!Z6}%G<Qspugff|qCDX}Uxu((kbQ;{~5Z49MN
zG#SC_G}#dIYPZ-@D>92qif^$MrRIWmG=hpr9F;P-s|s!ufeCP>1c{&A(wve^=%&EA
zp!fj~2r&vUL8fQHl{6EmFUiALbP+jrK*a()cA#A_(CQ)ZKyV6cGjj@?I715i9A?mv
z4}8D}T)fsYR_N3)fL5=&FvKd<GJ%GEK%<Jypj~T>5LPx*u^D(=(JUM?fWuvt0-Fp2
zW&H|;vc#OyRLGhhh0J1wjKsW@oYWMB<ou$d)Z~(!N-l8cgtXCNOW1hN4|~u=IVe*?
z@dtbGxG`j40Tj1r@(d1|%(oaFZ!tREVsu8twI5X3571;K0|Ubcdre5G9>tN9nO9tp
zm<$~z2Dfd&3(Y{of{<1&GkAen5opHo7IStg#t;ysriA9OHK51?H!4`AAgwCKqHD+z
z1}X+YnGc+uKs^Oe4Oz6Sgeirw8MJ4U3A9HOvV#-6cN4savliaUP2mL1d4qP{r|^Ip
zA)s-h6kbr#2Wqo|cF(5>r^u#=q{yXkrAVZRrZcB8fcDI%@MiMVf+H4EG}bVtNQ1}z
z*i)pzjS%quPez6cg&M{z4$vaXSnv|D8YU1go2_^TX#I8JjBth$sLDKs6q#TKP5G*7
zNP+;bt^l>Z6cUS46^f9Xt_n$s$=RUJYDq?FW|0DDcn#F>1+TgWb-#;APb0==U_C|h
z3=9lt8%c}eLG5nPfZ8qQl+?7$6O0TDQOp^c*^pEN$}QmGC~!&&xy6)`nLcYK2z<1U
zV$Mjdgr*f}ieva-A0+^d-1v;d;tbIELKJgSetyv<1_p-b;GJY?h7K>m6Az#!e-)ph
zv7vFQLV04bLPly)>MceCq?D-1gm8u?GuVuxn;-|@0u5ntA-261fmUcknjYX`D{!j=
zUbui;yO8wBQltdZ4_XBYPNYRwL0oVN117-547B|L?oNQ#1A!uHJ1DJ!jAdY9V^m;b
zVFb;(gGd%e7BG*Ck%f_inFE}TL6erCHETt8kW(|L_yHBR;M5FiBBy{y@<CmGP)8qL
z?0^bw@LW>~Gk8`CwCfx^It(s?Si$KHR0KgsmciR#5Jiv`cwE^Ex{8CdY9=JbC6%Tt
zl!HPTJVF9p+zy%!Do!m14Vx4t7H5DK_k&h4Bo={F8)%I`q(ssPPEO28EGh=4*33Lm
z41-Nb%1Kq!1g9;M$}D4ZupV%l0@pkt$U~jrkxE1&051@MF6;2q<N!zVEmrVaNJRAx
zsq%4`MOdR*96U?~@o+rcPx*N_K~W9rb~3ODfo7<{dqa!vAx9;sU;<%qRD#-P;DV`u
z@F4<_0s&mCus|jYz-(|a0pUXr7l<f9Y$#6xMJaSoXaR@~KU_cuJnRLUp+??)lnC8O
zuBQ+X8RFsZXRP4w=M#yx_P7`{%SYmQ1NT9h1w82u*-;HTZ@?Hm7C@E3E!L#Ul2mAK
z6M2h)B*+BNTo$COiO6%1?Sx=mFF@fBGKB$YV?U@q2XF3wgfkvM(GS{bo&ufBX$Fn3
zF@l=*H4M$5o+dMNNEOi%W=vrM?>}R$VTQ1>nTkC?bvSfbF+jl&yjc^HNWi%QR0czG
z1xab%)Bv33tH9$5dZ10s;G!4Wj#hE-gI2Od;H@)=q6<{yATK6}Vou9Uxy4ju03Oc0
z#aM(MkdV=%yn=X$n}2`;6Ox!&82KQj6Jya+(4st!`1st!%)I#cDnUmdAB7Ot;E-U2
z0LS29R~JQ1#-eOc8S8M1Jw84qKRG@gGUr<e(#Q&49|O*bkhUGT0p1LXT2NOgioKvx
z541_A2vp(~fhLlocp)n?^&nIfPkL%ed`W6<K~a8w$t~98jFikG$Z9OmazXF`4&Z|Y
zz#Hnp8?cK&<Ne@Wq(z`LAK>j};B8n%pnW&sy(i!;6h)u~IN&wG;6<pAWrU!`df@d@
z;DtLypyeds^%O;*wF2OoXYkA+c*>~=JOT_Fx(AO&gGY41qm<ymHt<*vc;o}zmoEZ!
zJ;7bKB2Yg7+}s2=VZbFnxXdU5C0KZXgO)bN$Add7MGrs$2995FHb4;IN{GWI7kq|_
z9jN723~Cm!fV&-_5hoDlVFGtUSa?`C7<rgE7{N4?5DN=42S0}p2M>oBhY*JlhaiUv
z2M32Z2QLR7hZ2Vj7Z(Q`hbV^}SdO2ApF@yCfJ1~sh=YTJpM#x)mxGr>h=Y>@B*Voa
j#KF(O&7s7>&mq9U%f-aT#KuT8Wa5CS1+iFg>SG1~A}ZGe
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_hickle_helpers.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_hickle_helpers.cpython-36.pyc
deleted file mode 100755
index d10d047823c6177abba7330f72046787d86a5e46..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 1969
zcmXr!<>hkDTM=`Eje+4Y0}|k7U|?`yU|=Xd!N9<f!Vtxf!kEI8!x+U>DFbFPM=_@`
zrZDF)=dwhxFfydDM6rf5<VmnFq_Smm6*(|6fM5!16l)|y9#aHJH(L&SE=LqcE@u>H
zE>{#6BS?%rhdY-iiieRQm9?lSg)xOAhc}lmijR>Yg)@b#g&~SRg*%0(g&|5Hg*Sz-
zg&|5Xg+E20g&|5PMKDFEg&|5fRU}0?o2lqXsz5UXBSR`bm=pq&f?$%h=t!zCh?T+=
z%%CY!<;SJ0P?B0)5}%QooSl;zpOKnVkXlr%S5V2t6`Ge>0+uXRNXsu$fT&Q&%}*)K
zNmbC*Re-2fNGr`tF3HT#E7s%UdI<_LKTXD4Y{dnMMa8KhZko)u1d}sTle6P9i{mpA
zi!&0Fa#E3mGD}j6z(NvGRZxC>W=U#pd`V?NswU$t5vW{od|qj8L1lbmQBh*0CgUw>
zxOj1LW<g~<#8QZKGRP?~%)-FH;0%gBEd~aL5>QYxr7)*6H#4TNWOEdYl`u9jf<&7c
zQ&_V(iq$~;6oz01O*X%mj0_A6FPT6DGXn!dh?^!;5y-S#e2&G%sYM`<yA~DY7p-I{
z;$dKb5WhP03-mMcb5r%RN{dsAiuE%}i;Hs;^YRS!vr3CnlN0rGb99SS%TkLn^YkIX
z8lRb0T#}fRqo0$Rq+d{3l98WhtY@ZQoLQ2pTacKXotU0l3{|RYtY@NUtXoiNtY@Z|
znx3u?ioaqoi5ZKx#6Ys~@Nj|!w-hKaK>lE06k{ynV_;ywA70IjDNNZM#R8y^0fkfw
za|#PM{LsUzhyxT#MVz3p;9_840EHJQ;EF&w2oXdO7B2$>1G)=gX+$33LIK7iew;3J
zLyOT8MsSQWfnv0TF@>?2v4j~!rZYif+o%LAm(Bp?JAwG%*k<(uyB^u`EJds!H!^4D
zm4JP4i#07LKd}Vje0YQxfjuIN9Ml-eSOw%CHc*0O6l0WOED}QZ5y&EN8Vv&bsDv?#
z2~@N|(kaxfG9}Cqe`T>?k+p)zf^>m>3yK>*kQmr^Z0NoVanoci0=eiGQ(nO>R&b^Q
zdoqeCHxa83C6RrIk{Lnysu*Mz1EUP19Al9vvhP6g55i#Im4SVi1@>J!V>4qdLkU9`
zV-|A?Bbdhs=CMF|OeL%}pm+kMJ*yJ7CdLxh8b)YZ^n&t1@d}cUg78b&vp7<iL1sb2
ztOz6y4gprbB2aD)0mrhNCg&|yaIsMY%H2_1$;CzSxrrr3nH3;*GMt?R7cXK5rD||O
z76!3I7#J8dnWI=s;**PuAXGAlN-8cw3O;b!0GFePYzVd&rMNNxB@a+OWMJe2Ljgt*
zED{GLKaTkL+{Dbh_;^jxTkP@iDf!9q@wZsPWnmH6C1N1wK)eAeWaGi+-{SVnEyyn_
z0hdove2{`h4^nCu7l9%KT#|uPc$6emIwwCd1ylimXh>-bP9oUl!PY=R1>|&aE(IrS
zaQwsFo|l-L8Xu1oB;bq#4i^raTyTA22P!s-^%xi!SeQ5%nHV`3c^Eku`53vFM3{IO
ZSr~a(I9NGYIoLSZI5;^N*`Sbx5di#C)lvWe
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_legacy_load.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_legacy_load.cpython-36.pyc
deleted file mode 100755
index 387c9076b34bdbd560ab83a5798a241b2adc1ece..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 840
zcmXr!<>hkDTM@&=%)s!N0SPcMFfceUFfbH5F)%QsFr+Z%FhoIU#wbQGpDBtNOtVC>
zfN9n!))b}`<`#x1wp8{MmTaaXyHvJj21bSy#$X0b)|Vi|{gOc>2(vOUFbFa*FgSzk
z*uucTP{NSKShS*qsf0O;Ifb#Av6-=!p@b!iHHE2}k&&TdLJflpLu_3wW2IOLTMZ*f
ztb{#_1H`Ij3TG%(VPpWo63#3xkXXN4EmH|w4O13(7Edz+ST%?TWmhQGFoDEf7-Gd@
zm};48SxVSySP*Ip%fcC&7{VD+7y=oB7$O)V846V*7z$Ow8S*$v*ix85=A|$MGib8-
zRSD*#rY9y>#%E;b6zgm0Wn|~vVoS^^PpmAi((nOGDdglQrex-&E9B?pR4SC`7iAYK
zWacRZR2u0i1m_nORqAQ--eNA!thmJ-5NULatvJ7^BsJw0OL|Uz(k+hi#G<^+y!7H*
zJjI#01v#l{nK>n?MYmWBiZb&`ZZU)GV*y!kizUOfpfZZZEi)(e7HeWjNm21F*36RB
z++t1ETO4VbIjK3B#U&6{USe+QEyl!K%-N}xw^%@mRx%WcFfcIuYSS;!&&bbB)z2y|
zPAw|d&nPV}&P~kAGt|#2ElN#J)X&Y)Elw><Ey~Q(&&W*9&Pk2W%quQQ%*oNu$xPBO
zs4U6I&okCD(=W~}N!2Y#OwLYBPc4Qj)iu^L(KFU9s5I6y(@RZH*DpyeE-40+@z78N
z`9!av@|FNd3R#dHlui^tDTk4bk&l^+iHnhmk%N(sk&Tguk&lswQHW89QG$tsu}F-8
zf#DWMe0*+VW?p=}CPxt)0|P@5NdGN1h?9%BKq5RKf*C}>bbx{-J{}wpNQ5ZZJe%D7
Zl+v73J4TQ@_!t-%I2fUji9?Ko5dg#O+^7Hm
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_scipy.cpython-36.pyc b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/__pycache__/test_scipy.cpython-36.pyc
deleted file mode 100755
index 6f7dbb8741560fc88a3fb0bd2356c226bae2cd69..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 1718
zcmXr!<>hkDTM<*r#=!8H0SPcMFfceUFfbGsFfcHrFr+Z%Fhnt?Fr+Z$Fy%5wG3T;G
zvE;Hwu`)vB*rM1{7*d#X*mF6eIKW~oIh?s%QCy4+sa#py%?w#Q%?ykTDXb}MEeuh-
zDeNg6Eeuh7sr)IN*-S+_seB;WR9+C7!Whh;$@LQC6hBSoTU^P-Me(_bB}JJPAa*jG
zodg%xWV|JvoRONG9iLeoU!0s-P#IrbkXTfl8lPBHlvt_Bc#AbBKRGcc8Du61gZ#$A
zz`)=P@}Cg{149Wz3R?+d7E=myFH<c;33Cd2Gh;1d2}=z_Gh?w*32PIG&DhLXtOsJJ
zFa$Gba`?SuW?*1=2{H#{LrO_yLF!9JkPygtO{OAt1_p*A4h9AWO{O9)5StssW-9_o
z-(t!uxW$s0S7K2FGNp)>fq~%`pJQ=xYEemMex7SlQGSso(=F!WTmukcxRRkrkb!~W
zSA%|menx(7s(w~!acWVqenx3=ac*K>o}qqLX;Er&qJD0UZgFZ^YEfpMenw_;c1~)1
zW?pegVor{JPG*vRL1jrsex9+OnSOC*Nvdu^Vsdt3dTKFLsjjh}iJq};L8Y;tnO<so
zx_(J&aY-?lgaoTzLFFwzkRUjq!9mOm3PO-485sE(Ihc4Di-Z{%7~mnz!@$4*3I}jV
zpJD`uG%qZqvzVJ1Y8grxvsh9<kTr!31k>5TVGj`niLiiD4Oo;3tQMq>F@*_~hQOlC
zB@8vpHH<Y(X-sKMDSRoMy-c+%C9Ftd{9rLwux^$VwiLz`j&zo0##&IEgKS2%g&nL8
z&6X^-6sBgzEcO%*kQ<wsYne+pYM5)-YuM75Qv|@~bChsqan-P-a5XbAGL&#;ai?%K
zGuCo~MOjn0;iBBljI~@~QH~TIxF}CEV=Z?HPYq`YLo?F^P)g%%Vl3gQ;VNMT^X*C?
zeC`qsFy9Z%XUyWO;mqQ%VaehLi>AOuxw80cSfQd-a8d3o{u&Ob=md~x4QG}>4NI0l
zHptFJU=gk?ff`n*$QG~&ca}g62UO%3NCcb%1pSbbIAakfD2ap8KO(We1SRX2Y@lS%
zc#E|-Be5X0iVK{6iVO5IOslw|Y$Ifjv0jELTJ8a*Ah2C9agcx}Q;{ghN)Vwb2+mbS
zpyXBrN|I6RCHdgYe~T4Nfpgg{wxZO6)Wnh~_M%j<V~RkT_ZAx@qugRiDa|do#gdbs
zm~x9NvlyIj@{&`F_!t-%qBs(Ba*}iMi&Jm0q$HLk7Kwp8#Z$x&%D2o#`Q^8mlk;<m
zWI+xR0I6p#&NYf+%gjqDC@H$do|%`DnVec2#ax_g9K~9kYZ#vaqKqJvaeM|icS8sv
zkdZ>rkO7xskkSw2&tfA`@x>^_B*qAWB8(zTLX2XJYK%fmJd9k7P|PI2D8MKHQNtv}
zD8yK#2r`o+K0Y@wGcP_~Q{WbBUTJPYB{;mIc)&%b9;C>Ggd@020tHTyB*-(8kPy-X
z1x;#EF_?yw2;hP%ioKvxuOP7`qX-n$Q9S9XCGjPxxdlb}`6ahllQU8>i@>D<+!=X^
zxvBB-5buJDH3$praSof@{FKt1R69^gDi&d2VBlaBVd7xqVd7!tVdmiAP~u=@V*~(3
CXrK@P
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_astropy.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_astropy.py
deleted file mode 100755
index 2086ec37..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_astropy.py
+++ /dev/null
@@ -1,133 +0,0 @@
-import hickle as hkl
-from astropy.units import Quantity
-from astropy.time import Time
-from astropy.coordinates import Angle, SkyCoord
-from astropy.constants import Constant, EMConstant, G
-from astropy.table import Table
-import numpy as np
-from py.path import local
-
-# Set the current working directory to the temporary directory
-local.get_temproot().chdir()
-
-def test_astropy_quantity():
-
- for uu in ['m^3', 'm^3 / s', 'kg/pc']:
- a = Quantity(7, unit=uu)
-
- hkl.dump(a, "test_ap.h5")
- b = hkl.load("test_ap.h5")
-
- assert a == b
- assert a.unit == b.unit
-
- a *= a
- hkl.dump(a, "test_ap.h5")
- b = hkl.load("test_ap.h5")
- assert a == b
- assert a.unit == b.unit
-
-def TODO_test_astropy_constant():
- hkl.dump(G, "test_ap.h5")
- gg = hkl.load("test_ap.h5")
-
- print(G)
- print(gg)
-
-def test_astropy_table():
- t = Table([[1, 2], [3, 4]], names=('a', 'b'), meta={'name': 'test_thing'})
-
- hkl.dump({'a': t}, "test_ap.h5")
- t2 = hkl.load("test_ap.h5")['a']
-
- print(t)
- print(t.meta)
- print(t2)
- print(t2.meta)
-
- print(t.dtype, t2.dtype)
- assert t.meta == t2.meta
- assert t.dtype == t2.dtype
-
- assert np.allclose(t['a'].astype('float32'), t2['a'].astype('float32'))
- assert np.allclose(t['b'].astype('float32'), t2['b'].astype('float32'))
-
-def test_astropy_quantity_array():
- a = Quantity([1,2,3], unit='m')
-
- hkl.dump(a, "test_ap.h5")
- b = hkl.load("test_ap.h5")
-
- assert np.allclose(a.value, b.value)
- assert a.unit == b.unit
-
-def test_astropy_time_array():
- times = ['1999-01-01T00:00:00.123456789', '2010-01-01T00:00:00']
- t1 = Time(times, format='isot', scale='utc')
- hkl.dump(t1, "test_ap2.h5")
- t2 = hkl.load("test_ap2.h5")
-
- print(t1)
- print(t2)
- assert t1.value.shape == t2.value.shape
- for ii in range(len(t1)):
- assert t1.value[ii] == t2.value[ii]
- assert t1.format == t2.format
- assert t1.scale == t2.scale
-
- times = [58264, 58265, 58266]
- t1 = Time(times, format='mjd', scale='utc')
- hkl.dump(t1, "test_ap2.h5")
- t2 = hkl.load("test_ap2.h5")
-
- print(t1)
- print(t2)
- assert t1.value.shape == t2.value.shape
- assert np.allclose(t1.value, t2.value)
- assert t1.format == t2.format
- assert t1.scale == t2.scale
-
-def test_astropy_angle():
- for uu in ['radian', 'degree']:
- a = Angle(1.02, unit=uu)
-
- hkl.dump(a, "test_ap.h5")
- b = hkl.load("test_ap.h5")
- assert a == b
- assert a.unit == b.unit
-
-def test_astropy_angle_array():
- a = Angle([1,2,3], unit='degree')
-
- hkl.dump(a, "test_ap.h5")
- b = hkl.load("test_ap.h5")
-
- assert np.allclose(a.value, b.value)
- assert a.unit == b.unit
-
-def test_astropy_skycoord():
- ra = Angle(['1d20m', '1d21m'], unit='degree')
- dec = Angle(['33d0m0s', '33d01m'], unit='degree')
- radec = SkyCoord(ra, dec)
- hkl.dump(radec, "test_ap.h5")
- radec2 = hkl.load("test_ap.h5")
- assert np.allclose(radec.ra.value, radec2.ra.value)
- assert np.allclose(radec.dec.value, radec2.dec.value)
-
- ra = Angle(['1d20m', '1d21m'], unit='hourangle')
- dec = Angle(['33d0m0s', '33d01m'], unit='degree')
- radec = SkyCoord(ra, dec)
- hkl.dump(radec, "test_ap.h5")
- radec2 = hkl.load("test_ap.h5")
- assert np.allclose(radec.ra.value, radec2.ra.value)
- assert np.allclose(radec.dec.value, radec2.dec.value)
-
-if __name__ == "__main__":
- test_astropy_quantity()
- #test_astropy_constant()
- test_astropy_table()
- test_astropy_quantity_array()
- test_astropy_time_array()
- test_astropy_angle()
- test_astropy_angle_array()
- test_astropy_skycoord()
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_hickle.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_hickle.py
deleted file mode 100755
index 54910542..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_hickle.py
+++ /dev/null
@@ -1,826 +0,0 @@
-#! /usr/bin/env python
-# encoding: utf-8
-"""
-# test_hickle.py
-
-Unit tests for hickle module.
-
-"""
-
-import h5py
-import hashlib
-import numpy as np
-import os
-import six
-import time
-from pprint import pprint
-
-from py.path import local
-
-import hickle
-from hickle.hickle import *
-
-
-# Set current working directory to the temporary directory
-local.get_temproot().chdir()
-
-NESTED_DICT = {
- "level1_1": {
- "level2_1": [1, 2, 3],
- "level2_2": [4, 5, 6]
- },
- "level1_2": {
- "level2_1": [1, 2, 3],
- "level2_2": [4, 5, 6]
- },
- "level1_3": {
- "level2_1": {
- "level3_1": [1, 2, 3],
- "level3_2": [4, 5, 6]
- },
- "level2_2": [4, 5, 6]
- }
-}
-
-DUMP_CACHE = [] # Used in test_track_times()
-
-
-def test_string():
- """ Dumping and loading a string """
- if six.PY2:
- filename, mode = 'test.h5', 'w'
- string_obj = "The quick brown fox jumps over the lazy dog"
- dump(string_obj, filename, mode)
- string_hkl = load(filename)
- #print "Initial list: %s"%list_obj
- #print "Unhickled data: %s"%list_hkl
- assert type(string_obj) == type(string_hkl) == str
- assert string_obj == string_hkl
- else:
- pass
-
-
-def test_unicode():
- """ Dumping and loading a unicode string """
- if six.PY2:
- filename, mode = 'test.h5', 'w'
- u = unichr(233) + unichr(0x0bf2) + unichr(3972) + unichr(6000)
- dump(u, filename, mode)
- u_hkl = load(filename)
-
- assert type(u) == type(u_hkl) == unicode
- assert u == u_hkl
- # For those interested, uncomment below to see what those codes are:
- # for i, c in enumerate(u_hkl):
- # print i, '%04x' % ord(c), unicodedata.category(c),
- # print unicodedata.name(c)
- else:
- pass
-
-
-def test_unicode2():
- if six.PY2:
- a = u"unicode test"
- dump(a, 'test.hkl', mode='w')
-
- z = load('test.hkl')
- assert a == z
- assert type(a) == type(z) == unicode
- pprint(z)
- else:
- pass
-
-def test_list():
- """ Dumping and loading a list """
- filename, mode = 'test_list.h5', 'w'
- list_obj = [1, 2, 3, 4, 5]
- dump(list_obj, filename, mode=mode)
- list_hkl = load(filename)
- #print(f'Initial list: {list_obj}')
- #print(f'Unhickled data: {list_hkl}')
- try:
- assert type(list_obj) == type(list_hkl) == list
- assert list_obj == list_hkl
- import h5py
- a = h5py.File(filename)
- a.close()
-
- except AssertionError:
- print("ERR:", list_obj, list_hkl)
- import h5py
-
- raise()
-
-
-def test_set():
- """ Dumping and loading a list """
- filename, mode = 'test_set.h5', 'w'
- list_obj = set([1, 0, 3, 4.5, 11.2])
- dump(list_obj, filename, mode)
- list_hkl = load(filename)
- #print "Initial list: %s"%list_obj
- #print "Unhickled data: %s"%list_hkl
- try:
- assert type(list_obj) == type(list_hkl) == set
- assert list_obj == list_hkl
- except AssertionError:
- print(type(list_obj))
- print(type(list_hkl))
- #os.remove(filename)
- raise
-
-
-def test_numpy():
- """ Dumping and loading numpy array """
- filename, mode = 'test.h5', 'w'
- dtypes = ['float32', 'float64', 'complex64', 'complex128']
-
- for dt in dtypes:
- array_obj = np.ones(8, dtype=dt)
- dump(array_obj, filename, mode)
- array_hkl = load(filename)
- try:
- assert array_hkl.dtype == array_obj.dtype
- assert np.all((array_hkl, array_obj))
- except AssertionError:
- print(array_hkl)
- print(array_obj)
- raise
-
-
-def test_masked():
- """ Test masked numpy array """
- filename, mode = 'test.h5', 'w'
- a = np.ma.array([1,2,3,4], dtype='float32', mask=[0,1,0,0])
-
- dump(a, filename, mode)
- a_hkl = load(filename)
-
- try:
- assert a_hkl.dtype == a.dtype
- assert np.all((a_hkl, a))
- except AssertionError:
- print(a_hkl)
- print(a)
- raise
-
-
-def test_dict():
- """ Test dictionary dumping and loading """
- filename, mode = 'test.h5', 'w'
-
- dd = {
- 'name' : b'Danny',
- 'age' : 28,
- 'height' : 6.1,
- 'dork' : True,
- 'nums' : [1, 2, 3],
- 'narr' : np.array([1,2,3]),
- #'unic' : u'dan[at]thetelegraphic.com'
- }
-
-
- dump(dd, filename, mode)
- dd_hkl = load(filename)
-
- for k in dd.keys():
- try:
- assert k in dd_hkl.keys()
-
- if type(dd[k]) is type(np.array([1])):
- assert np.all((dd[k], dd_hkl[k]))
- else:
- #assert dd_hkl[k] == dd[k]
- pass
- assert type(dd_hkl[k]) == type(dd[k])
- except AssertionError:
- print(k)
- print(dd_hkl[k])
- print(dd[k])
- print(type(dd_hkl[k]), type(dd[k]))
- raise
-
-
-def test_empty_dict():
- """ Test empty dictionary dumping and loading """
- filename, mode = 'test.h5', 'w'
-
- dump({}, filename, mode)
- assert load(filename) == {}
-
-
-def test_compression():
- """ Test compression on datasets"""
-
- filename, mode = 'test.h5', 'w'
- dtypes = ['int32', 'float32', 'float64', 'complex64', 'complex128']
-
- comps = [None, 'gzip', 'lzf']
-
- for dt in dtypes:
- for cc in comps:
- array_obj = np.ones(32768, dtype=dt)
- dump(array_obj, filename, mode, compression=cc)
- print(cc, os.path.getsize(filename))
- array_hkl = load(filename)
- try:
- assert array_hkl.dtype == array_obj.dtype
- assert np.all((array_hkl, array_obj))
- except AssertionError:
- print(array_hkl)
- print(array_obj)
- raise
-
-
-def test_dict_int_key():
- """ Test for dictionaries with integer keys """
- filename, mode = 'test.h5', 'w'
-
- dd = {
- 0: "test",
- 1: "test2"
- }
-
- dump(dd, filename, mode)
- dd_hkl = load(filename)
-
-
-def test_dict_nested():
- """ Test for dictionaries with integer keys """
- filename, mode = 'test.h5', 'w'
-
- dd = NESTED_DICT
-
- dump(dd, filename, mode)
- dd_hkl = load(filename)
-
- ll_hkl = dd_hkl["level1_3"]["level2_1"]["level3_1"]
- ll = dd["level1_3"]["level2_1"]["level3_1"]
- assert ll == ll_hkl
-
-
-def test_masked_dict():
- """ Test dictionaries with masked arrays """
-
- filename, mode = 'test.h5', 'w'
-
- dd = {
- "data" : np.ma.array([1,2,3], mask=[True, False, False]),
- "data2" : np.array([1,2,3,4,5])
- }
-
- dump(dd, filename, mode)
- dd_hkl = load(filename)
-
- for k in dd.keys():
- try:
- assert k in dd_hkl.keys()
- if type(dd[k]) is type(np.array([1])):
- assert np.all((dd[k], dd_hkl[k]))
- elif type(dd[k]) is type(np.ma.array([1])):
- print(dd[k].data)
- print(dd_hkl[k].data)
- assert np.allclose(dd[k].data, dd_hkl[k].data)
- assert np.allclose(dd[k].mask, dd_hkl[k].mask)
-
- assert type(dd_hkl[k]) == type(dd[k])
-
- except AssertionError:
- print(k)
- print(dd_hkl[k])
- print(dd[k])
- print(type(dd_hkl[k]), type(dd[k]))
- raise
-
-
-def test_np_float():
- """ Test for singular np dtypes """
- filename, mode = 'np_float.h5', 'w'
-
- dtype_list = (np.float16, np.float32, np.float64,
- np.complex64, np.complex128,
- np.int8, np.int16, np.int32, np.int64,
- np.uint8, np.uint16, np.uint32, np.uint64)
-
- for dt in dtype_list:
-
- dd = dt(1)
- dump(dd, filename, mode)
- dd_hkl = load(filename)
- assert dd == dd_hkl
- assert dd.dtype == dd_hkl.dtype
-
- dd = {}
- for dt in dtype_list:
- dd[str(dt)] = dt(1.0)
- dump(dd, filename, mode)
- dd_hkl = load(filename)
-
- print(dd)
- for dt in dtype_list:
- assert dd[str(dt)] == dd_hkl[str(dt)]
-
-
-def md5sum(filename, blocksize=65536):
- """ Compute MD5 sum for a given file """
- hash = hashlib.md5()
-
- with open(filename, "r+b") as f:
- for block in iter(lambda: f.read(blocksize), ""):
- hash.update(block)
- return hash.hexdigest()
-
-
-def caching_dump(obj, filename, *args, **kwargs):
- """ Save arguments of all dump calls """
- DUMP_CACHE.append((obj, filename, args, kwargs))
- return hickle_dump(obj, filename, *args, **kwargs)
-
-
-def test_track_times():
- """ Verify that track_times = False produces identical files """
- hashes = []
- for obj, filename, mode, kwargs in DUMP_CACHE:
- if isinstance(filename, hickle.H5FileWrapper):
- filename = str(filename.file_name)
- kwargs['track_times'] = False
- caching_dump(obj, filename, mode, **kwargs)
- hashes.append(md5sum(filename))
-
- time.sleep(1)
-
- for hash1, (obj, filename, mode, kwargs) in zip(hashes, DUMP_CACHE):
- if isinstance(filename, hickle.H5FileWrapper):
- filename = str(filename.file_name)
- caching_dump(obj, filename, mode, **kwargs)
- hash2 = md5sum(filename)
- print(hash1, hash2)
- assert hash1 == hash2
-
-
-def test_comp_kwargs():
- """ Test compression with some kwargs for shuffle and chunking """
-
- filename, mode = 'test.h5', 'w'
- dtypes = ['int32', 'float32', 'float64', 'complex64', 'complex128']
-
- comps = [None, 'gzip', 'lzf']
- chunks = [(100, 100), (250, 250)]
- shuffles = [True, False]
- scaleoffsets = [0, 1, 2]
-
- for dt in dtypes:
- for cc in comps:
- for ch in chunks:
- for sh in shuffles:
- for so in scaleoffsets:
- kwargs = {
- 'compression' : cc,
- 'dtype': dt,
- 'chunks': ch,
- 'shuffle': sh,
- 'scaleoffset': so
- }
- #array_obj = np.random.random_integers(low=-8192, high=8192, size=(1000, 1000)).astype(dt)
- array_obj = NESTED_DICT
- dump(array_obj, filename, mode, compression=cc)
- print(kwargs, os.path.getsize(filename))
- array_hkl = load(filename)
-
-
-def test_list_numpy():
- """ Test converting a list of numpy arrays """
-
- filename, mode = 'test.h5', 'w'
-
- a = np.ones(1024)
- b = np.zeros(1000)
- c = [a, b]
-
- dump(c, filename, mode)
- dd_hkl = load(filename)
-
- print(dd_hkl)
-
- assert isinstance(dd_hkl, list)
- assert isinstance(dd_hkl[0], np.ndarray)
-
-
-def test_tuple_numpy():
- """ Test converting a list of numpy arrays """
-
- filename, mode = 'test.h5', 'w'
-
- a = np.ones(1024)
- b = np.zeros(1000)
- c = (a, b, a)
-
- dump(c, filename, mode)
- dd_hkl = load(filename)
-
- print(dd_hkl)
-
- assert isinstance(dd_hkl, tuple)
- assert isinstance(dd_hkl[0], np.ndarray)
-
-
-def test_none():
- """ Test None type hickling """
-
- filename, mode = 'test.h5', 'w'
-
- a = None
-
- dump(a, filename, mode)
- dd_hkl = load(filename)
- print(a)
- print(dd_hkl)
-
- assert isinstance(dd_hkl, type(None))
-
-
-def test_dict_none():
- """ Test None type hickling """
-
- filename, mode = 'test.h5', 'w'
-
- a = {'a': 1, 'b' : None}
-
- dump(a, filename, mode)
- dd_hkl = load(filename)
- print(a)
- print(dd_hkl)
-
- assert isinstance(a['b'], type(None))
-
-
-def test_file_open_close():
- """ https://github.com/telegraphic/hickle/issues/20 """
- import h5py
- f = h5py.File('test.hdf', 'w')
- a = np.arange(5)
-
- dump(a, 'test.hkl')
- dump(a, 'test.hkl')
-
- dump(a, f, mode='w')
- f.close()
- try:
- dump(a, f, mode='w')
- except hickle.hickle.ClosedFileError:
- print("Tests: Closed file exception caught")
-
-
-def test_list_order():
- """ https://github.com/telegraphic/hickle/issues/26 """
- d = [np.arange(n + 1) for n in range(20)]
- hickle.dump(d, 'test.h5')
- d_hkl = hickle.load('test.h5')
-
- try:
- for ii, xx in enumerate(d):
- assert d[ii].shape == d_hkl[ii].shape
- for ii, xx in enumerate(d):
- assert np.allclose(d[ii], d_hkl[ii])
- except AssertionError:
- print(d[ii], d_hkl[ii])
- raise
-
-
-def test_embedded_array():
- """ See https://github.com/telegraphic/hickle/issues/24 """
-
- d_orig = [[np.array([10., 20.]), np.array([10, 20, 30])], [np.array([10, 2]), np.array([1.])]]
- hickle.dump(d_orig, 'test.h5')
- d_hkl = hickle.load('test.h5')
-
- for ii, xx in enumerate(d_orig):
- for jj, yy in enumerate(xx):
- assert np.allclose(d_orig[ii][jj], d_hkl[ii][jj])
-
- print(d_hkl)
- print(d_orig)
-
-
-################
-## NEW TESTS ##
-################
-
-
-def generate_nested():
- a = [1, 2, 3]
- b = [a, a, a]
- c = [a, b, 's']
- d = [a, b, c, c, a]
- e = [d, d, d, d, 1]
- f = {'a' : a, 'b' : b, 'e' : e}
- g = {'f' : f, 'a' : e, 'd': d}
- h = {'h': g, 'g' : f}
- z = [f, a, b, c, d, e, f, g, h, g, h]
- a = np.array([1, 2, 3, 4])
- b = set([1, 2, 3, 4, 5])
- c = (1, 2, 3, 4, 5)
- d = np.ma.array([1, 2, 3, 4, 5, 6, 7, 8])
- z = {'a': a, 'b': b, 'c': c, 'd': d, 'z': z}
- return z
-
-
-def test_is_iterable():
- a = [1, 2, 3]
- b = 1
-
- assert check_is_iterable(a) == True
- assert check_is_iterable(b) == False
-
-
-def test_check_iterable_item_type():
-
- a = [1, 2, 3]
- b = [a, a, a]
- c = [a, b, 's']
-
- type_a = check_iterable_item_type(a)
- type_b = check_iterable_item_type(b)
- type_c = check_iterable_item_type(c)
-
- assert type_a is int
- assert type_b is list
- assert type_c == False
-
-
-def test_dump_nested():
- """ Dump a complicated nested object to HDF5
- """
- z = generate_nested()
- dump(z, 'test.hkl', mode='w')
-
-
-def test_with_dump():
- lst = [1]
- tpl = (1)
- dct = {1: 1}
- arr = np.array([1])
-
- with h5py.File('test.hkl') as file:
- dump(lst, file, path='/lst')
- dump(tpl, file, path='/tpl')
- dump(dct, file, path='/dct')
- dump(arr, file, path='/arr')
-
-
-def test_with_load():
- lst = [1]
- tpl = (1)
- dct = {1: 1}
- arr = np.array([1])
-
- with h5py.File('test.hkl') as file:
- assert load(file, '/lst') == lst
- assert load(file, '/tpl') == tpl
- assert load(file, '/dct') == dct
- assert load(file, '/arr') == arr
-
-
-def test_load():
-
- a = set([1, 2, 3, 4])
- b = set([5, 6, 7, 8])
- c = set([9, 10, 11, 12])
- z = (a, b, c)
- z = [z, z]
- z = (z, z, z, z, z)
-
- print("Original:")
- pprint(z)
- dump(z, 'test.hkl', mode='w')
-
- print("\nReconstructed:")
- z = load('test.hkl')
- pprint(z)
-
-
-def test_sort_keys():
- keys = [b'data_0', b'data_1', b'data_2', b'data_3', b'data_10']
- keys_sorted = [b'data_0', b'data_1', b'data_2', b'data_3', b'data_10']
-
- print(keys)
- print(keys_sorted)
- assert sort_keys(keys) == keys_sorted
-
-
-def test_ndarray():
-
- a = np.array([1,2,3])
- b = np.array([2,3,4])
- z = (a, b)
-
- print("Original:")
- pprint(z)
- dump(z, 'test.hkl', mode='w')
-
- print("\nReconstructed:")
- z = load('test.hkl')
- pprint(z)
-
-
-def test_ndarray_masked():
-
- a = np.ma.array([1,2,3])
- b = np.ma.array([2,3,4], mask=[True, False, True])
- z = (a, b)
-
- print("Original:")
- pprint(z)
- dump(z, 'test.hkl', mode='w')
-
- print("\nReconstructed:")
- z = load('test.hkl')
- pprint(z)
-
-
-def test_simple_dict():
- a = {'key1': 1, 'key2': 2}
-
- dump(a, 'test.hkl')
- z = load('test.hkl')
-
- pprint(a)
- pprint(z)
-
-
-def test_complex_dict():
- a = {'akey': 1, 'akey2': 2}
- if six.PY2:
- # NO LONG TYPE IN PY3!
- b = {'bkey': 2.0, 'bkey3': long(3.0)}
- else:
- b = a
- c = {'ckey': "hello", "ckey2": "hi there"}
- z = {'zkey1': a, 'zkey2': b, 'zkey3': c}
-
- print("Original:")
- pprint(z)
- dump(z, 'test.hkl', mode='w')
-
- print("\nReconstructed:")
- z = load('test.hkl')
- pprint(z)
-
-def test_multi_hickle():
- a = {'a': 123, 'b': [1, 2, 4]}
-
- if os.path.exists("test.hkl"):
- os.remove("test.hkl")
- dump(a, "test.hkl", path="/test", mode="w")
- dump(a, "test.hkl", path="/test2", mode="r+")
- dump(a, "test.hkl", path="/test3", mode="r+")
- dump(a, "test.hkl", path="/test4", mode="r+")
-
- a = load("test.hkl", path="/test")
- b = load("test.hkl", path="/test2")
- c = load("test.hkl", path="/test3")
- d = load("test.hkl", path="/test4")
-
-def test_complex():
- """ Test complex value dtype is handled correctly
-
- https://github.com/telegraphic/hickle/issues/29 """
-
- data = {"A":1.5, "B":1.5 + 1j, "C":np.linspace(0,1,4) + 2j}
- dump(data, "test.hkl")
- data2 = load("test.hkl")
- for key in data.keys():
- assert type(data[key]) == type(data2[key])
-
-def test_nonstring_keys():
- """ Test that keys are reconstructed back to their original datatypes
- https://github.com/telegraphic/hickle/issues/36
- """
- if six.PY2:
- u = unichr(233) + unichr(0x0bf2) + unichr(3972) + unichr(6000)
-
- data = {u'test': 123,
- 'def': 456,
- 'hik' : np.array([1,2,3]),
- u: u,
- 0: 0,
- True: 'hi',
- 1.1 : 'hey',
- #2L : 'omg',
- 1j: 'complex_hashable',
- (1, 2): 'boo',
- ('A', 17.4, 42): [1, 7, 'A'],
- (): '1313e was here',
- '0': 0
- }
- #data = {'0': 123, 'def': 456}
- print(data)
- dump(data, "test.hkl")
- data2 = load("test.hkl")
- print(data2)
-
- for key in data.keys():
- assert key in data2.keys()
-
- print(data2)
- else:
- pass
-
-def test_scalar_compression():
- """ Test bug where compression causes a crash on scalar datasets
-
- (Scalars are incompressible!)
- https://github.com/telegraphic/hickle/issues/37
- """
- data = {'a' : 0, 'b' : np.float(2), 'c' : True}
-
- dump(data, "test.hkl", compression='gzip')
- data2 = load("test.hkl")
-
- print(data2)
- for key in data.keys():
- assert type(data[key]) == type(data2[key])
-
-def test_bytes():
- """ Dumping and loading a string. PYTHON3 ONLY """
- if six.PY3:
- filename, mode = 'test.h5', 'w'
- string_obj = b"The quick brown fox jumps over the lazy dog"
- dump(string_obj, filename, mode)
- string_hkl = load(filename)
- #print "Initial list: %s"%list_obj
- #print "Unhickled data: %s"%list_hkl
- print(type(string_obj))
- print(type(string_hkl))
- assert type(string_obj) == type(string_hkl) == bytes
- assert string_obj == string_hkl
- else:
- pass
-
-def test_np_scalar():
- """ Numpy scalar datatype
-
- https://github.com/telegraphic/hickle/issues/50
- """
-
- fid='test.h5py'
- r0={'test': np.float64(10.)}
- s = dump(r0, fid)
- r = load(fid)
- print(r)
- assert type(r0['test']) == type(r['test'])
-
-if __name__ == '__main__':
- """ Some tests and examples """
- test_sort_keys()
-
- test_np_scalar()
- test_scalar_compression()
- test_complex()
- test_file_open_close()
- test_dict_none()
- test_none()
- test_masked_dict()
- test_list()
- test_set()
- test_numpy()
- test_dict()
- test_empty_dict()
- test_compression()
- test_masked()
- test_dict_nested()
- test_comp_kwargs()
- test_list_numpy()
- test_tuple_numpy()
- test_track_times()
- test_list_order()
- test_embedded_array()
- test_np_float()
-
- if six.PY2:
- test_unicode()
- test_unicode2()
- test_string()
- test_nonstring_keys()
-
- if six.PY3:
- test_bytes()
-
-
- # NEW TESTS
- test_is_iterable()
- test_check_iterable_item_type()
- test_dump_nested()
- test_with_dump()
- test_with_load()
- test_load()
- test_sort_keys()
- test_ndarray()
- test_ndarray_masked()
- test_simple_dict()
- test_complex_dict()
- test_multi_hickle()
- test_dict_int_key()
-
- # Cleanup
- print("ALL TESTS PASSED!")
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_hickle_helpers.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_hickle_helpers.py
deleted file mode 100755
index 253839e9..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_hickle_helpers.py
+++ /dev/null
@@ -1,63 +0,0 @@
-#! /usr/bin/env python
-# encoding: utf-8
-"""
-# test_hickle_helpers.py
-
-Unit tests for hickle module -- helper functions.
-
-"""
-
-import numpy as np
-try:
- import scipy
- from scipy import sparse
- _has_scipy = True
-except ImportError:
- _has_scipy = False
-
-from hickle.helpers import check_is_hashable, check_is_iterable, check_iterable_item_type
-
-from hickle.loaders.load_numpy import check_is_numpy_array
-if _has_scipy:
- from hickle.loaders.load_scipy import check_is_scipy_sparse_array
-
-
-
-def test_check_is_iterable():
- assert check_is_iterable([1,2,3]) is True
- assert check_is_iterable(1) is False
-
-
-def test_check_is_hashable():
- assert check_is_hashable(1) is True
- assert check_is_hashable([1,2,3]) is False
-
-
-def test_check_iterable_item_type():
- assert check_iterable_item_type([1,2,3]) is int
- assert check_iterable_item_type([int(1), float(1)]) is False
- assert check_iterable_item_type([]) is False
-
-
-def test_check_is_numpy_array():
- assert check_is_numpy_array(np.array([1,2,3])) is True
- assert check_is_numpy_array(np.ma.array([1,2,3])) is True
- assert check_is_numpy_array([1,2]) is False
-
-
-def test_check_is_scipy_sparse_array():
- t_csr = scipy.sparse.csr_matrix([0])
- t_csc = scipy.sparse.csc_matrix([0])
- t_bsr = scipy.sparse.bsr_matrix([0])
- assert check_is_scipy_sparse_array(t_csr) is True
- assert check_is_scipy_sparse_array(t_csc) is True
- assert check_is_scipy_sparse_array(t_bsr) is True
- assert check_is_scipy_sparse_array(np.array([1])) is False
-
-if __name__ == "__main__":
- test_check_is_hashable()
- test_check_is_iterable()
- test_check_is_numpy_array()
- test_check_iterable_item_type()
- if _has_scipy:
- test_check_is_scipy_sparse_array()
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_legacy_load.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_legacy_load.py
deleted file mode 100755
index e849bcf6..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_legacy_load.py
+++ /dev/null
@@ -1,30 +0,0 @@
-import glob
-import warnings
-import hickle as hkl
-import h5py
-import six
-
-def test_legacy_load():
- if six.PY2:
- filelist = sorted(glob.glob('legacy_hkls/*.hkl'))
-
- # Make all warnings show
- warnings.simplefilter("always")
-
- for filename in filelist:
- try:
- print(filename)
- a = hkl.load(filename)
- except:
- with h5py.File(filename) as a:
- print(a.attrs.items())
- print(a.items())
- for key, item in a.items():
- print(item.attrs.items())
- raise
- else:
- print("Legacy loading only works in Py2. Sorry.")
- pass
-
-if __name__ == "__main__":
- test_legacy_load()
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_scipy.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_scipy.py
deleted file mode 100755
index ab78311d..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/hickle-3.4.3-py3.6.egg/tests/test_scipy.py
+++ /dev/null
@@ -1,57 +0,0 @@
-import numpy as np
-from scipy.sparse import csr_matrix, csc_matrix, bsr_matrix
-
-import hickle
-from hickle.loaders.load_scipy import check_is_scipy_sparse_array
-
-from py.path import local
-
-# Set the current working directory to the temporary directory
-local.get_temproot().chdir()
-
-
-def test_is_sparse():
- sm0 = csr_matrix((3, 4), dtype=np.int8)
- sm1 = csc_matrix((1, 2))
-
- assert check_is_scipy_sparse_array(sm0)
- assert check_is_scipy_sparse_array(sm1)
-
-
-def test_sparse_matrix():
- sm0 = csr_matrix((3, 4), dtype=np.int8).toarray()
-
- row = np.array([0, 0, 1, 2, 2, 2])
- col = np.array([0, 2, 2, 0, 1, 2])
- data = np.array([1, 2, 3, 4, 5, 6])
- sm1 = csr_matrix((data, (row, col)), shape=(3, 3))
- sm2 = csc_matrix((data, (row, col)), shape=(3, 3))
-
- indptr = np.array([0, 2, 3, 6])
- indices = np.array([0, 2, 2, 0, 1, 2])
- data = np.array([1, 2, 3, 4, 5, 6]).repeat(4).reshape(6, 2, 2)
- sm3 = bsr_matrix((data,indices, indptr), shape=(6, 6))
-
- hickle.dump(sm1, 'test_sp.h5')
- sm1_h = hickle.load('test_sp.h5')
- hickle.dump(sm2, 'test_sp2.h5')
- sm2_h = hickle.load('test_sp2.h5')
- hickle.dump(sm3, 'test_sp3.h5')
- sm3_h = hickle.load('test_sp3.h5')
-
- assert isinstance(sm1_h, csr_matrix)
- assert isinstance(sm2_h, csc_matrix)
- assert isinstance(sm3_h, bsr_matrix)
-
- assert np.allclose(sm1_h.data, sm1.data)
- assert np.allclose(sm2_h.data, sm2.data)
- assert np.allclose(sm3_h.data, sm3.data)
-
- assert sm1_h. shape == sm1.shape
- assert sm2_h. shape == sm2.shape
- assert sm3_h. shape == sm3.shape
-
-
-if __name__ == "__main__":
- test_sparse_matrix()
- test_is_sparse()
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/site.py b/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/site.py
deleted file mode 100755
index 0d2d2ff8..00000000
--- a/workflow_parallel_frame_prediction/Training/hickle/lib/python3.6/site-packages/site.py
+++ /dev/null
@@ -1,74 +0,0 @@
-def __boot():
- import sys
- import os
- PYTHONPATH = os.environ.get('PYTHONPATH')
- if PYTHONPATH is None or (sys.platform == 'win32' and not PYTHONPATH):
- PYTHONPATH = []
- else:
- PYTHONPATH = PYTHONPATH.split(os.pathsep)
-
- pic = getattr(sys, 'path_importer_cache', {})
- stdpath = sys.path[len(PYTHONPATH):]
- mydir = os.path.dirname(__file__)
-
- for item in stdpath:
- if item == mydir or not item:
- continue # skip if current dir. on Windows, or my own directory
- importer = pic.get(item)
- if importer is not None:
- loader = importer.find_module('site')
- if loader is not None:
- # This should actually reload the current module
- loader.load_module('site')
- break
- else:
- try:
- import imp # Avoid import loop in Python >= 3.3
- stream, path, descr = imp.find_module('site', [item])
- except ImportError:
- continue
- if stream is None:
- continue
- try:
- # This should actually reload the current module
- imp.load_module('site', stream, path, descr)
- finally:
- stream.close()
- break
- else:
- raise ImportError("Couldn't find the real 'site' module")
-
- known_paths = dict([(makepath(item)[1], 1) for item in sys.path]) # 2.2 comp
-
- oldpos = getattr(sys, '__egginsert', 0) # save old insertion position
- sys.__egginsert = 0 # and reset the current one
-
- for item in PYTHONPATH:
- addsitedir(item)
-
- sys.__egginsert += oldpos # restore effective old position
-
- d, nd = makepath(stdpath[0])
- insert_at = None
- new_path = []
-
- for item in sys.path:
- p, np = makepath(item)
-
- if np == nd and insert_at is None:
- # We've hit the first 'system' path entry, so added entries go here
- insert_at = len(new_path)
-
- if np in known_paths or insert_at is None:
- new_path.append(item)
- else:
- # new path after the insert point, back-insert it
- new_path.insert(insert_at, item)
- insert_at += 1
-
- sys.path[:] = new_path
-
-
-if __name__ == 'site':
- __boot()
- del __boot
diff --git a/workflow_parallel_frame_prediction/Training/horovodJob.sh b/workflow_parallel_frame_prediction/Training/horovodJob.sh
deleted file mode 100644
index 236a08d9..00000000
--- a/workflow_parallel_frame_prediction/Training/horovodJob.sh
+++ /dev/null
@@ -1,44 +0,0 @@
-#!/bin/bash
-#SBATCH --account=deepacf
-# budget account where contingent is taken from# TASKS = NODES * GPUS_PER_NODE
-#SBATCH --nodes=3
-#SBATCH --ntasks-per-node=4
-#SBATCH --ntasks=12
-# can be omitted if --nodes and --ntasks-per-node
-# are given
-# SBATCH --cpus-per-task=1
-# for OpenMP/hybrid jobs only
-#SBATCH --output=horovod-%j.out
-# if keyword omitted: Default is slurm-%j.out in
-# the submission directory (%j is replaced by
-# the job ID).
-#SBATCH --error=horovod-%j.err
-# if keyword omitted: Default is slurm-%j.out in
-# the submission directory.
-#SBATCH --time=20:00:00
-#SBATCH --gres=gpu:4
-#SBATCH --partition=gpus
-#SBATCH --mail-user=b.gong@fz-juelich.de
-#SBATCH --mail-type=ALL
-
-#create a folder to save the output
-jutil env activate -p deepacf
-module --force purge
-module load Stages/Devel-2019a
-module load GCC/8.3.0
-module load MVAPICH2/2.3.2-GDR
-module load Stages/2019a
-module load Horovod/0.16.2-GPU-Python-3.6.8
-module load Keras/2.2.4-GPU-Python-3.6.8
-
-#module load ParaStationMPI/5.2.2-1
-#module load h5py/2.9.0-Python-3.6.8
-# *** start of job script ***:
-# Note: The current working directory at this point is
-# the directory where sbatch was executed.
-# export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK}
-# *** start of job script ***
-# Note: The current working directory at this point is
-# the directory where sbatch was executed.
-# export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK}
-srun --cpu_bind=none python3.6 kitti_train_horovod.py
diff --git a/workflow_parallel_frame_prediction/Training/keras_utils.py b/workflow_parallel_frame_prediction/Training/keras_utils.py
deleted file mode 100755
index ededcc74..00000000
--- a/workflow_parallel_frame_prediction/Training/keras_utils.py
+++ /dev/null
@@ -1,58 +0,0 @@
-import os
-import numpy as np
-
-from keras import backend as K
-from keras.legacy.interfaces import generate_legacy_interface, recurrent_args_preprocessor
-from keras.models import model_from_json
-
-legacy_prednet_support = generate_legacy_interface(
- allowed_positional_args=['stack_sizes', 'R_stack_sizes',
- 'A_filt_sizes', 'Ahat_filt_sizes', 'R_filt_sizes'],
- conversions=[('dim_ordering', 'data_format'),
- ('consume_less', 'implementation')],
- value_conversions={'dim_ordering': {'tf': 'channels_last',
- 'th': 'channels_first',
- 'default': None},
- 'consume_less': {'cpu': 0,
- 'mem': 1,
- 'gpu': 2}},
- preprocessor=recurrent_args_preprocessor)
-
-# Convert old Keras (1.2) json models and weights to Keras 2.0
-def convert_model_to_keras2(old_json_file, old_weights_file, new_json_file, new_weights_file):
- from prednet import PredNet
- # If using tensorflow, it doesn't allow you to load the old weights.
- if K.backend() != 'theano':
- os.environ['KERAS_BACKEND'] = backend
- reload(K)
-
- f = open(old_json_file, 'r')
- json_string = f.read()
- f.close()
- model = model_from_json(json_string, custom_objects = {'PredNet': PredNet})
- model.load_weights(old_weights_file)
-
- weights = model.layers[1].get_weights()
- if weights[0].shape[0] == model.layers[1].stack_sizes[1]:
- for i, w in enumerate(weights):
- if w.ndim == 4:
- weights[i] = np.transpose(w, (2, 3, 1, 0))
- model.set_weights(weights)
-
- model.save_weights(new_weights_file)
- json_string = model.to_json()
- with open(new_json_file, "w") as f:
- f.write(json_string)
-
-
-if __name__ == '__main__':
- old_dir = './model_data/'
- new_dir = './model_data_keras2/'
- if not os.path.exists(new_dir):
- os.mkdir(new_dir)
- for w_tag in ['', '-Lall', '-extrapfinetuned']:
- m_tag = '' if w_tag == '-Lall' else w_tag
- convert_model_to_keras2(old_dir + 'prednet_kitti_model' + m_tag + '.json',
- old_dir + 'prednet_kitti_weights' + w_tag + '.hdf5',
- new_dir + 'prednet_kitti_model' + m_tag + '.json',
- new_dir + 'prednet_kitti_weights' + w_tag + '.hdf5')
diff --git a/workflow_parallel_frame_prediction/Training/kitti_settings.py b/workflow_parallel_frame_prediction/Training/kitti_settings.py
deleted file mode 100755
index 54767111..00000000
--- a/workflow_parallel_frame_prediction/Training/kitti_settings.py
+++ /dev/null
@@ -1,19 +0,0 @@
-# Where KITTI data will be saved if you run process_kitti.py
-# If you directly download the processed data, change to the path of the data.
-## Changed logic: Now this is the path where the processed data lies: X_train,val,test
-#DATA_DIR = './kitti_data/'
-#data directory for training data 2015 and 2016
-#DATA_DIR = '/p/project/cjjsc42/severin/try3'
-#data directory for moving objects:
-#DATA_DIR = '/p/home/jusers/hussmann1/jureca/movingObjects/se_nw'
-#data directory for featuretesting:
-##DATA_DIR = './testTry2'
-DATA_DIR = '/p/scratch/cjjsc42/bing/PredNet/processData/splits'
-# Where model weights and config will be saved if you run kitti_train.py
-# If you directly download the trained weights, change to appropriate path.
-WEIGHTS_DIR = './model_data_keras2/'
-#WEIGHTS_DIR = '/p/project/cjjsc42/bing/ml-severin/model_data_keras2'
-
-# Where results (prediction plots and evaluation file) will be saved.
-#RESULTS_SAVE_DIR = './kitti_results'
-
diff --git a/workflow_parallel_frame_prediction/Training/kitti_train_horovod.py b/workflow_parallel_frame_prediction/Training/kitti_train_horovod.py
deleted file mode 100755
index 72539927..00000000
--- a/workflow_parallel_frame_prediction/Training/kitti_train_horovod.py
+++ /dev/null
@@ -1,119 +0,0 @@
-'''
-Train PredNet on KITTI sequences. (Geiger et al. 2013, http://www.cvlibs.net/datasets/kitti/)
-'''
-
-import os
-import numpy as np
-np.random.seed(123)
-#from six.moves import cPickle
-
-from keras import backend as K
-from keras.models import Model
-from keras.layers import Input, Dense, Flatten
-from keras.layers import LSTM
-from keras.layers import TimeDistributed
-from keras.callbacks import LearningRateScheduler, ModelCheckpoint
-from keras.optimizers import Adam
-from prednet import PredNet
-from data_utils import SequenceGenerator
-from kitti_settings import *
-import datetime
-import horovod.keras as hvd
-import keras
-import tensorflow as tf
-#Horovod:initialize horovod
-hvd.init()
-#Horovod: pin GPU to be used for process local rank (one GPU per process)
-config = tf.ConfigProto()
-config.gpu_options.allow_growth = True
-config.gpu_options.visible_device_list = str(hvd.local_rank())
-K.set_session(tf.Session(config=config))
-
-print("horovode size", hvd.size())
-
-save_model = True# if weights will be saved
-weights_file = os.path.join(WEIGHTS_DIR, 'prednet_kitti_weights.hdf5') # where weights will be saved
-json_file = os.path.join(WEIGHTS_DIR, 'prednet_kitti_model.json')
-if not os.path.exists(WEIGHTS_DIR): os.mkdir(WEIGHTS_DIR)
-# Data files
-train_file = os.path.join(DATA_DIR, 'X_train.hkl')
-train_sources = os.path.join(DATA_DIR, 'sources_train.hkl')
-val_file = os.path.join(DATA_DIR, 'X_val.hkl')
-val_sources = os.path.join(DATA_DIR, 'sources_val.hkl')
-
-# Training parameters
-nb_epoch = 10 #original: 150; for all tests so far set to 100; t2onlyMax: 150
-batch_size = 15
-samples_per_epoch = 500 #original: 500; for all tests so far set to 300; t2onlyMax: 500
-N_seq_val = 80 # number of sequences to use for validation ##original: 100; for all tests so far set to 65; t2onlyMax: 80
-
-# Model parameters
-n_channels, im_height, im_width = (3, 128, 160)
-input_shape = (n_channels, im_height, im_width) if K.image_data_format() == 'channels_first' else (im_height, im_width, n_channels)
-stack_sizes = (n_channels, 48, 96, 192)
-R_stack_sizes = stack_sizes
-A_filt_sizes = (3, 3, 3)
-Ahat_filt_sizes = (3, 3, 3, 3)
-R_filt_sizes = (3, 3, 3, 3)
-layer_loss_weights = np.array([1., 0., 0., 0.]) # weighting for each layer in final loss; "L_0" model: [1, 0, 0, 0], "L_all": [1, 0.1, 0.1, 0.1]
-layer_loss_weights = np.expand_dims(layer_loss_weights, 1)
-nt = 10 # number of timesteps used for sequences in training
-time_loss_weights = 1./ (nt - 1) * np.ones((nt,1)) # equally weight all timesteps except the first
-time_loss_weights[0] = 0
-
-prednet = PredNet(stack_sizes, R_stack_sizes,
- A_filt_sizes, Ahat_filt_sizes, R_filt_sizes,
- output_mode='error', return_sequences=True)
-inputs = Input(shape=(nt,) + input_shape)
-errors = prednet(inputs) # errors will be (batch_size, nt, nb_layers)
-errors_by_time = TimeDistributed(Dense(1, trainable=False), weights=[layer_loss_weights, np.zeros(1)], trainable=False)(errors) # calculate weighted error by layer
-errors_by_time = Flatten()(errors_by_time) # will be (batch_size, nt)
-final_errors = Dense(1, weights=[time_loss_weights, np.zeros(1)], trainable=False)(errors_by_time) # weight errors by time
-model = Model(inputs=inputs, outputs=final_errors)
-#Horovod:ajust learning rate based on number of GPUs
-opt = keras.optimizers.Adam(0.01 * hvd.size())
-#Horovod: add horovod DistributedOptimizer
-opt = hvd.DistributedOptimizer(opt)
-#Horovode: use hvd.DistributedOptimizer to compute gradients
-model.compile(loss="mean_absolute_error", optimizer=opt, metrics=["accuracy"])
-
-
-
-train_generator = SequenceGenerator(train_file, train_sources, nt, batch_size=batch_size, shuffle=True)
-val_generator = SequenceGenerator(val_file, val_sources, nt, batch_size=batch_size, N_seq=N_seq_val)
-
-#lr_schedule = lambda epoch: 0.001 if epoch < 75 else 0.0001 # start with lr of 0.001 and then drop to 0.0001 after 75 epochs
-callbacks = [hvd.callbacks.BroadcastGlobalVariablesCallback(0),
- #hvd.callbacks.MetricAverageCallback(),
- hvd.callbacks.LearningRateWarmupCallback(warmup_epochs=5,verbose=1)
- ]
-#bing: original save_model is True
-if hvd.rank() == 0:
- if save_model:
- print("===========The model will be saved =======")
- callbacks.append(ModelCheckpoint(filepath=weights_file, monitor='val_loss', save_best_only=True))
-
-#the start training time
-a = datetime.datetime.now()
-
-history = model.fit_generator(generator=train_generator,steps_per_epoch=samples_per_epoch/(batch_size*hvd.size()), epochs=nb_epoch, callbacks=callbacks,
- validation_data=val_generator, validation_steps=N_seq_val/(batch_size*hvd.size()))
-
-
-b = datetime.datetime.now()
-
-#the training time
-t = b-a
-
-stats = list(train_generator.X.shape)
-stats.append(t)
-
-print("training time is",stats)
-
-if save_model:
- json_string = model.to_json()
- with open(json_file, "w") as f:
- f.write(json_string)
-
-
-
diff --git a/workflow_parallel_frame_prediction/Training/minMaxExtractor.py b/workflow_parallel_frame_prediction/Training/minMaxExtractor.py
deleted file mode 100644
index 5c121620..00000000
--- a/workflow_parallel_frame_prediction/Training/minMaxExtractor.py
+++ /dev/null
@@ -1,76 +0,0 @@
-import hickle as hkl
-import numpy as np
-import matplotlib.pyplot as plt
-
-#x_train = hkl.load('/Users/Severin/Desktop/X_train.hkl') #load X_train produces on jureca
-x_train = hkl.load('/p/project/cjjsc42/severin/try3/X_train.hkl') #load X_train produces on jureca
-print('Shape of X:')
-print(x_train.shape)
-print('')
-
-#Print example
-#t2_cutout = x_train[100,:,:,0]
-#printt2cutout = plt.pcolormesh(t2_cutout[::-1,:], shading='bottom', cmap=plt.cm.jet)
-#plt.savefig('t2_cutout')
-#Extract Max min values:
-maxT2 = np.amax(x_train[:,:,:,0]) # numpy.amax() returns the maximum of an array or maximum along an axis.
-print('maxT2: ' + str(maxT2))
-minT2 = np.amin(x_train[:,:,:,0])
-print('minT2: ' + str(minT2))
-meanT2 = np.mean(x_train[:,:,:,0])
-print('meanT2: ' + str(meanT2))
-stdT2 = np.std(x_train[:,:,:,0])
-print('stdT2: ' + str(stdT2))
-highCutT2 = meanT2 + 3 * stdT2
-print('highCutT2: ' + str(highCutT2))
-lowCutT2 = meanT2 - 3 * stdT2
-print('lowCutT2: ' + str(lowCutT2))
-print('')
-
-maxGP = np.amax(x_train[:,:,:,1])
-print('maxGP: ' + str(maxGP))
-minGP = np.amin(x_train[:,:,:,1])
-print('minGP: ' + str(minGP))
-meanGP = np.mean(x_train[:,:,:,1])
-print('meanGP: ' + str(meanGP))
-stdGP = np.std(x_train[:,:,:,1])
-print('stdGP: ' + str(stdGP))
-highCutGP = meanGP + 3 * stdGP
-print('highCutGP: ' + str(highCutGP))
-lowCutGP = meanGP - 3 * stdGP
-print('lowCutGP: ' + str(lowCutGP))
-print('')
-
-maxGPH = np.amax(x_train[:,:,:,2])
-print('maxGPH: ' + str(maxGPH))
-minGPH = np.amin(x_train[:,:,:,2])
-print('minGPH: ' + str(minGPH))
-meanGPH = np.mean(x_train[:,:,:,2])
-print('meanGP: ' + str(meanGPH))
-stdGPH = np.std(x_train[:,:,:,2])
-print('stdGPH: ' + str(stdGPH))
-highCutGPH = meanGPH + 3 * stdGPH
-print('highCutGPH: ' + str(highCutGPH))
-lowCutGPH = meanGPH - 3 * stdGPH
-print('lowCutGPH: ' + str(lowCutGPH))
-print('')
-
-# Formel zum normalisieren: z = (x-min(x))/(max(x)-min(x))
-#x_trainNormalized2 = np.zeros(shape=x_train.shape)
-#print('Empty shape:')
-#print(x_trainNormalized2.shape)
-#x_trainNormalized2[:,:,:,0] = (x_train[:,:,:,0]-minT2)/(maxT2-minT2)
-#x_trainNormalized2[:,:,:,1] = (x_train[:,:,:,1]-minGP)/(maxGP-minGP)
-#x_trainNormalized2[:,:,:,2] = (x_train[:,:,:,2]-minGPH)/(maxGPH-minGPH)
-
-#print('MaxMin values of normalized dataset:')
-#print('T2:')
-#print(np.amax(x_trainNormalized2[:,:,:,0]))
-#print(np.amin(x_trainNormalized2[:,:,:,0]))
-#print('GP:')
-#print(np.amax(x_trainNormalized2[:,:,:,1]))
-#print(np.amin(x_trainNormalized2[:,:,:,1]))
-#print('GPH:')
-#print(np.amax(x_trainNormalized2[:,:,:,2]))
-#print(np.amin(x_trainNormalized2[:,:,:,2]))
-#print(x_trainNormalized2)
\ No newline at end of file
diff --git a/workflow_parallel_frame_prediction/Training/prednet.py b/workflow_parallel_frame_prediction/Training/prednet.py
deleted file mode 100755
index b5a0208a..00000000
--- a/workflow_parallel_frame_prediction/Training/prednet.py
+++ /dev/null
@@ -1,311 +0,0 @@
-import numpy as np
-
-from keras import backend as K
-from keras import activations
-from keras.layers import Recurrent
-from keras.layers import Conv2D, UpSampling2D, MaxPooling2D
-from keras.engine import InputSpec
-from keras_utils import legacy_prednet_support
-
-class PredNet(Recurrent):
- '''PredNet architecture - Lotter 2016.
- Stacked convolutional LSTM inspired by predictive coding principles.
-
- # Arguments
- stack_sizes: number of channels in targets (A) and predictions (Ahat) in each layer of the architecture.
- Length is the number of layers in the architecture.
- First element is the number of channels in the input.
- Ex. (3, 16, 32) would correspond to a 3 layer architecture that takes in RGB images and has 16 and 32
- channels in the second and third layers, respectively.
- R_stack_sizes: number of channels in the representation (R) modules.
- Length must equal length of stack_sizes, but the number of channels per layer can be different.
- A_filt_sizes: filter sizes for the target (A) modules.
- Has length of 1 - len(stack_sizes).
- Ex. (3, 3) would mean that targets for layers 2 and 3 are computed by a 3x3 convolution of the errors (E)
- from the layer below (followed by max-pooling)
- Ahat_filt_sizes: filter sizes for the prediction (Ahat) modules.
- Has length equal to length of stack_sizes.
- Ex. (3, 3, 3) would mean that the predictions for each layer are computed by a 3x3 convolution of the
- representation (R) modules at each layer.
- R_filt_sizes: filter sizes for the representation (R) modules.
- Has length equal to length of stack_sizes.
- Corresponds to the filter sizes for all convolutions in the LSTM.
- pixel_max: the maximum pixel value.
- Used to clip the pixel-layer prediction.
- error_activation: activation function for the error (E) units.
- A_activation: activation function for the target (A) and prediction (A_hat) units.
- LSTM_activation: activation function for the cell and hidden states of the LSTM.
- LSTM_inner_activation: activation function for the gates in the LSTM.
- output_mode: either 'error', 'prediction', 'all' or layer specification (ex. R2, see below).
- Controls what is outputted by the PredNet.
- If 'error', the mean response of the error (E) units of each layer will be outputted.
- That is, the output shape will be (batch_size, nb_layers).
- If 'prediction', the frame prediction will be outputted.
- If 'all', the output will be the frame prediction concatenated with the mean layer errors.
- The frame prediction is flattened before concatenation.
- Nomenclature of 'all' is kept for backwards compatibility, but should not be confused with returning all of the layers of the model
- For returning the features of a particular layer, output_mode should be of the form unit_type + layer_number.
- For instance, to return the features of the LSTM "representational" units in the lowest layer, output_mode should be specificied as 'R0'.
- The possible unit types are 'R', 'Ahat', 'A', and 'E' corresponding to the 'representation', 'prediction', 'target', and 'error' units respectively.
- extrap_start_time: time step for which model will start extrapolating.
- Starting at this time step, the prediction from the previous time step will be treated as the "actual"
- data_format: 'channels_first' or 'channels_last'.
- It defaults to the `image_data_format` value found in your
- Keras config file at `~/.keras/keras.json`.
-
- # References
- - [Deep predictive coding networks for video prediction and unsupervised learning](https://arxiv.org/abs/1605.08104)
- - [Long short-term memory](http://deeplearning.cs.cmu.edu/pdfs/Hochreiter97_lstm.pdf)
- - [Convolutional LSTM network: a machine learning approach for precipitation nowcasting](http://arxiv.org/abs/1506.04214)
- - [Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects](http://www.nature.com/neuro/journal/v2/n1/pdf/nn0199_79.pdf)
- '''
- @legacy_prednet_support
- def __init__(self, stack_sizes, R_stack_sizes,
- A_filt_sizes, Ahat_filt_sizes, R_filt_sizes,
- pixel_max=1., error_activation='relu', A_activation='relu',
- LSTM_activation='tanh', LSTM_inner_activation='hard_sigmoid',
- output_mode='error', extrap_start_time=None,
- data_format=K.image_data_format(), **kwargs):
- self.stack_sizes = stack_sizes
- self.nb_layers = len(stack_sizes)
- assert len(R_stack_sizes) == self.nb_layers, 'len(R_stack_sizes) must equal len(stack_sizes)'
- self.R_stack_sizes = R_stack_sizes
- assert len(A_filt_sizes) == (self.nb_layers - 1), 'len(A_filt_sizes) must equal len(stack_sizes) - 1'
- self.A_filt_sizes = A_filt_sizes
- assert len(Ahat_filt_sizes) == self.nb_layers, 'len(Ahat_filt_sizes) must equal len(stack_sizes)'
- self.Ahat_filt_sizes = Ahat_filt_sizes
- assert len(R_filt_sizes) == (self.nb_layers), 'len(R_filt_sizes) must equal len(stack_sizes)'
- self.R_filt_sizes = R_filt_sizes
-
- self.pixel_max = pixel_max
- self.error_activation = activations.get(error_activation)
- self.A_activation = activations.get(A_activation)
- self.LSTM_activation = activations.get(LSTM_activation)
- self.LSTM_inner_activation = activations.get(LSTM_inner_activation)
-
- default_output_modes = ['prediction', 'error', 'all']
- layer_output_modes = [layer + str(n) for n in range(self.nb_layers) for layer in ['R', 'E', 'A', 'Ahat']]
- assert output_mode in default_output_modes + layer_output_modes, 'Invalid output_mode: ' + str(output_mode)
- self.output_mode = output_mode
- if self.output_mode in layer_output_modes:
- self.output_layer_type = self.output_mode[:-1]
- self.output_layer_num = int(self.output_mode[-1])
- else:
- self.output_layer_type = None
- self.output_layer_num = None
- self.extrap_start_time = extrap_start_time
-
- assert data_format in {'channels_last', 'channels_first'}, 'data_format must be in {channels_last, channels_first}'
- self.data_format = data_format
- self.channel_axis = -3 if data_format == 'channels_first' else -1
- self.row_axis = -2 if data_format == 'channels_first' else -3
- self.column_axis = -1 if data_format == 'channels_first' else -2
- super(PredNet, self).__init__(**kwargs)
- self.input_spec = [InputSpec(ndim=5)]
-
- def compute_output_shape(self, input_shape):
- if self.output_mode == 'prediction':
- out_shape = input_shape[2:]
- elif self.output_mode == 'error':
- out_shape = (self.nb_layers,)
- elif self.output_mode == 'all':
- out_shape = (np.prod(input_shape[2:]) + self.nb_layers,)
- else:
- stack_str = 'R_stack_sizes' if self.output_layer_type == 'R' else 'stack_sizes'
- stack_mult = 2 if self.output_layer_type == 'E' else 1
- out_stack_size = stack_mult * getattr(self, stack_str)[self.output_layer_num]
- out_nb_row = input_shape[self.row_axis] / 2**self.output_layer_num
- out_nb_col = input_shape[self.column_axis] / 2**self.output_layer_num
- if self.data_format == 'channels_first':
- out_shape = (out_stack_size, out_nb_row, out_nb_col)
- else:
- out_shape = (out_nb_row, out_nb_col, out_stack_size)
-
- if self.return_sequences:
- return (input_shape[0], input_shape[1]) + out_shape
- else:
- return (input_shape[0],) + out_shape
-
- def get_initial_state(self, x):
- input_shape = self.input_spec[0].shape
- init_nb_row = input_shape[self.row_axis]
- init_nb_col = input_shape[self.column_axis]
-
- base_initial_state = K.zeros_like(x) # (samples, timesteps) + image_shape
- non_channel_axis = -1 if self.data_format == 'channels_first' else -2
- for _ in range(2):
- base_initial_state = K.sum(base_initial_state, axis=non_channel_axis)
- base_initial_state = K.sum(base_initial_state, axis=1) # (samples, nb_channels)
-
- initial_states = []
- states_to_pass = ['r', 'c', 'e']
- nlayers_to_pass = {u: self.nb_layers for u in states_to_pass}
- if self.extrap_start_time is not None:
- states_to_pass.append('ahat') # pass prediction in states so can use as actual for t+1 when extrapolating
- nlayers_to_pass['ahat'] = 1
- for u in states_to_pass:
- for l in range(nlayers_to_pass[u]):
- ds_factor = 2 ** l
- nb_row = init_nb_row // ds_factor
- nb_col = init_nb_col // ds_factor
- if u in ['r', 'c']:
- stack_size = self.R_stack_sizes[l]
- elif u == 'e':
- stack_size = 2 * self.stack_sizes[l]
- elif u == 'ahat':
- stack_size = self.stack_sizes[l]
- output_size = stack_size * nb_row * nb_col # flattened size
-
- reducer = K.zeros((input_shape[self.channel_axis], output_size)) # (nb_channels, output_size)
- initial_state = K.dot(base_initial_state, reducer) # (samples, output_size)
- if self.data_format == 'channels_first':
- output_shp = (-1, stack_size, nb_row, nb_col)
- else:
- output_shp = (-1, nb_row, nb_col, stack_size)
- initial_state = K.reshape(initial_state, output_shp)
- initial_states += [initial_state]
-
- if K._BACKEND == 'theano':
- from theano import tensor as T
- # There is a known issue in the Theano scan op when dealing with inputs whose shape is 1 along a dimension.
- # In our case, this is a problem when training on grayscale images, and the below line fixes it.
- initial_states = [T.unbroadcast(init_state, 0, 1) for init_state in initial_states]
-
- if self.extrap_start_time is not None:
- initial_states += [K.variable(0, int if K.backend() != 'tensorflow' else 'int32')] # the last state will correspond to the current timestep
- return initial_states
-
- def build(self, input_shape):
- self.input_spec = [InputSpec(shape=input_shape)]
- self.conv_layers = {c: [] for c in ['i', 'f', 'c', 'o', 'a', 'ahat']}
-
- for l in range(self.nb_layers):
- for c in ['i', 'f', 'c', 'o']:
- act = self.LSTM_activation if c == 'c' else self.LSTM_inner_activation
- self.conv_layers[c].append(Conv2D(self.R_stack_sizes[l], self.R_filt_sizes[l], padding='same', activation=act, data_format=self.data_format))
-
- act = 'relu' if l == 0 else self.A_activation
- self.conv_layers['ahat'].append(Conv2D(self.stack_sizes[l], self.Ahat_filt_sizes[l], padding='same', activation=act, data_format=self.data_format))
-
- if l < self.nb_layers - 1:
- self.conv_layers['a'].append(Conv2D(self.stack_sizes[l+1], self.A_filt_sizes[l], padding='same', activation=self.A_activation, data_format=self.data_format))
-
- self.upsample = UpSampling2D(data_format=self.data_format)
- self.pool = MaxPooling2D(data_format=self.data_format)
-
- self.trainable_weights = []
- nb_row, nb_col = (input_shape[-2], input_shape[-1]) if self.data_format == 'channels_first' else (input_shape[-3], input_shape[-2])
- for c in sorted(self.conv_layers.keys()):
- for l in range(len(self.conv_layers[c])):
- ds_factor = 2 ** l
- if c == 'ahat':
- nb_channels = self.R_stack_sizes[l]
- elif c == 'a':
- nb_channels = 2 * self.stack_sizes[l]
- else:
- nb_channels = self.stack_sizes[l] * 2 + self.R_stack_sizes[l]
- if l < self.nb_layers - 1:
- nb_channels += self.R_stack_sizes[l+1]
- in_shape = (input_shape[0], nb_channels, nb_row // ds_factor, nb_col // ds_factor)
- if self.data_format == 'channels_last': in_shape = (in_shape[0], in_shape[2], in_shape[3], in_shape[1])
- with K.name_scope('layer_' + c + '_' + str(l)):
- self.conv_layers[c][l].build(in_shape)
- self.trainable_weights += self.conv_layers[c][l].trainable_weights
-
- self.states = [None] * self.nb_layers*3
-
- if self.extrap_start_time is not None:
- self.t_extrap = K.variable(self.extrap_start_time, int if K.backend() != 'tensorflow' else 'int32')
- self.states += [None] * 2 # [previous frame prediction, timestep]
-
- def step(self, a, states):
- r_tm1 = states[:self.nb_layers]
- c_tm1 = states[self.nb_layers:2*self.nb_layers]
- e_tm1 = states[2*self.nb_layers:3*self.nb_layers]
-
- if self.extrap_start_time is not None:
- t = states[-1]
- a = K.switch(t >= self.t_extrap, states[-2], a) # if past self.extrap_start_time, the previous prediction will be treated as the actual
-
- c = []
- r = []
- e = []
-
- # Update R units starting from the top
- for l in reversed(range(self.nb_layers)):
- inputs = [r_tm1[l], e_tm1[l]]
- if l < self.nb_layers - 1:
- inputs.append(r_up)
-
- inputs = K.concatenate(inputs, axis=self.channel_axis)
- i = self.conv_layers['i'][l].call(inputs)
- f = self.conv_layers['f'][l].call(inputs)
- o = self.conv_layers['o'][l].call(inputs)
- _c = f * c_tm1[l] + i * self.conv_layers['c'][l].call(inputs)
- _r = o * self.LSTM_activation(_c)
- c.insert(0, _c)
- r.insert(0, _r)
-
- if l > 0:
- r_up = self.upsample.call(_r)
-
- # Update feedforward path starting from the bottom
- for l in range(self.nb_layers):
- ahat = self.conv_layers['ahat'][l].call(r[l])
- if l == 0:
- ahat = K.minimum(ahat, self.pixel_max)
- frame_prediction = ahat
-
- # compute errors
- e_up = self.error_activation(ahat - a)
- e_down = self.error_activation(a - ahat)
-
- e.append(K.concatenate((e_up, e_down), axis=self.channel_axis))
-
- if self.output_layer_num == l:
- if self.output_layer_type == 'A':
- output = a
- elif self.output_layer_type == 'Ahat':
- output = ahat
- elif self.output_layer_type == 'R':
- output = r[l]
- elif self.output_layer_type == 'E':
- output = e[l]
-
- if l < self.nb_layers - 1:
- a = self.conv_layers['a'][l].call(e[l])
- a = self.pool.call(a) # target for next layer
-
- if self.output_layer_type is None:
- if self.output_mode == 'prediction':
- output = frame_prediction
- else:
- for l in range(self.nb_layers):
- layer_error = K.mean(K.batch_flatten(e[l]), axis=-1, keepdims=True)
- all_error = layer_error if l == 0 else K.concatenate((all_error, layer_error), axis=-1)
- if self.output_mode == 'error':
- output = all_error
- else:
- output = K.concatenate((K.batch_flatten(frame_prediction), all_error), axis=-1)
-
- states = r + c + e
- if self.extrap_start_time is not None:
- states += [frame_prediction, t + 1]
- return output, states
-
- def get_config(self):
- config = {'stack_sizes': self.stack_sizes,
- 'R_stack_sizes': self.R_stack_sizes,
- 'A_filt_sizes': self.A_filt_sizes,
- 'Ahat_filt_sizes': self.Ahat_filt_sizes,
- 'R_filt_sizes': self.R_filt_sizes,
- 'pixel_max': self.pixel_max,
- 'error_activation': self.error_activation.__name__,
- 'A_activation': self.A_activation.__name__,
- 'LSTM_activation': self.LSTM_activation.__name__,
- 'LSTM_inner_activation': self.LSTM_inner_activation.__name__,
- 'data_format': self.data_format,
- 'extrap_start_time': self.extrap_start_time,
- 'output_mode': self.output_mode}
- base_config = super(PredNet, self).get_config()
- return dict(list(base_config.items()) + list(config.items()))
diff --git a/workflow_parallel_frame_prediction/Training/process_netCDF.py b/workflow_parallel_frame_prediction/Training/process_netCDF.py
deleted file mode 100644
index 2cb822ec..00000000
--- a/workflow_parallel_frame_prediction/Training/process_netCDF.py
+++ /dev/null
@@ -1,169 +0,0 @@
-'''
-Code for processing staged ERA5 data
-'''
-
-import os
-#import requests
-#from bs4 import BeautifulSoup
-#import urllib.request
-import numpy as np
-#from imageio import imread
-#from scipy.misc import imresize
-import hickle as hkl
-from netCDF4 import Dataset
-from kitti_settings import *
-
-#TODO: Not optimal with DATA_DIR and filingPath: In original process_kitti.py
-# there's just DATA_DIR (which is specified in kitti_settings.py) and in there
-# the processed data will be stores. The raw data lies also in there in a subfolder
-
-#Path of ERA5 Data
-DATA_DIR = './testData2'
-print(DATA_DIR)
-#Path where to save processed data
-filingPath = './testTry2'
-
-# ToDo: Define a convenient function to create a list containing all files.
-imageList = list(os.walk(DATA_DIR, topdown=False))[-1][-1]
-imageList = sorted(imageList)
-print('Image List:')
-print(imageList)
-print('Length of Image List: ' + str(len(imageList)))
-#scp hussmann1@jureca.fz-juelich.de:/p/project/cjjsc42/severin/data/era5_extract_481.nc ~/Desktop/netCDFdata
-
-# ToDo: Define properly the train, val and test index
-# Here just for testing and taking weird .DS_Store file into consideration
-# http://www.apfelwiki.de/Main/DSStore
-#train_recordings = imageList[1:6]
-#val_recordings = imageList[7:9]
-#test_recordings = imageList[-2:]
-
-#Train,Val,Test size in percentage
-partition = [0.8, 0.05, 0.15]
-#determine correct indices
-train_begin = 0
-train_end = round(partition[0]*len(imageList))-1
-val_begin = train_end + 1
-val_end = train_end + round(partition[1]*len(imageList))
-test_begin = val_end + 1
-test_end = len(imageList)-1
-print('Indices of Train, Val and test: '+ str(train_begin) + ' ' + str(val_begin) + ' ' + str(test_begin))
-#slightly adapting start and end because starts at the first index given and stops before(!) the last.
-train_recordings = imageList[train_begin:val_begin]
-val_recordings = imageList[val_begin:test_begin]
-test_recordings = imageList[test_begin:test_end]
-
-#adapted for feature testing: just first year (2015); Otherwise would take too long and some weird mistake in some data in 2016
-#in total: 17544
-#half: 8772
-#train: 0-6900
-#val:6901-7000
-#test:7001-8772
-#train_recordings = imageList[0:1000]
-#val_recordings = imageList[6901:7000]
-#test_recordings = imageList[7001:8772]
-
-print('Now everything together:')
-print('Train:')
-print(train_recordings)
-print('Val:')
-print(val_recordings)
-print('Test:')
-print(test_recordings)
-
-desired_im_sz = (128, 160)
-# Create image datasets.
-# Processes images and saves them in train, val, test splits.
-def process_data():
- splits = {s: [] for s in ['train', 'test', 'val']}
- splits['val'] = val_recordings
- splits['test'] = test_recordings
- splits['train'] = train_recordings
- for split in splits:
- source_list = [DATA_DIR] * len(splits[split]) # corresponds to recording that image came from
- print(splits[split])
- print(source_list)
- print((len(splits[split])==(len(source_list))))
- print('The list of ' + split + ' has length: ' + str(len(source_list)))
- print( 'Creating ' + split + ' data: ' + str(len(source_list)) + ' images')
-
- #X = np.zeros((len(splits[split]),) + desired_im_sz + (3,), np.uint8)
- #print(X)
- #print('shape of X' + str(X.shape))
-
- ##### TODO: iterate over split and read every .nc file, cut out array,
- ##### overlay arrays for RGB like style.
- ##### Save everything after for loop.
- EU_stack_list = [0] * (len(splits[split]))
-
- for i, im_file in enumerate(splits[split]):
- im_path = os.path.join(DATA_DIR, im_file)
- print('Open following dataset: ' + im_path)
- im = Dataset(im_path, mode = 'r')
- #print(im)
- t2 = im.variables['T2'][0,:,:]
- msl = im.variables['MSL'][0,:,:]
- gph500 = im.variables['gph500'][0,:,:]
- im.close()
- EU_t2 = t2[74:202, 550:710]
- EU_msl = msl[74:202, 550:710]
- EU_gph500 = gph500[74:202, 550:710]
- print(EU_t2.shape, EU_msl.shape, EU_gph500.shape)
- #Normal stack: T2, MSL & GPH500
- #EU_stack = np.stack([EU_t2, EU_msl, EU_gph500],axis=2)
- #Stack T2 only:
- #EU_stack = np.stack([EU_t2, EU_t2, EU_t2],axis=2)
- #EU_stack_list[i]=EU_stack
- #Stack T2*2 MSL*1:
- #EU_stack = np.stack([EU_t2, EU_t2, EU_msl],axis=2)
- #EU_stack_list[i]=EU_stack
- #EU_stack = np.stack([EU_t2, EU_msl, EU_msl],axis=2)
- #EU_stack_list[i]=EU_stack
- #Stack T2*2 gph500*1:
- #EU_stack = np.stack([EU_t2, EU_t2, EU_gph500],axis=2)
- #EU_stack_list[i]=EU_stack
- #Stack T2*1 gph500*2
- #EU_stack = np.stack([EU_t2, EU_gph500, EU_gph500],axis=2)
- #EU_stack_list[i]=EU_stack
- #print(EU_stack.shape)
- #X[i]=EU_stack #this should be unnecessary
- #t2_1 stack. Stack t2 with two empty arrays
- #empty_image = np.zeros(shape = (128, 160))
- #EU_stack = np.stack([EU_t2, empty_image, empty_image],axis=2)
- #EU_stack_list[i]=EU_stack
- #t2_2 stack. Stack t2 with one empty array
- empty_image = np.zeros(shape = (128, 160))
- EU_stack = np.stack([EU_t2, EU_t2, empty_image],axis=2)
- EU_stack_list[i]=EU_stack
- #print('Does ist work? ')
- #print(EU_stack_list[i][:,:,0]==EU_t2)
- #print(EU_stack[:,:,1]==EU_msl)
- X = np.array(EU_stack_list)
- print('Shape of X: ' + str(X.shape))
- hkl.dump(X, os.path.join(filingPath, 'X_' + split + '.hkl')) #Not optimal!
- hkl.dump(source_list, os.path.join(filingPath, 'sources_' + split + '.hkl'))
-
-
- #for category, folder in splits[split]:
- # im_dir = os.path.join(DATA_DIR, 'raw/', category, folder, folder[:10], folder, 'image_03/data/')
- # files = list(os.walk(im_dir, topdown=False))[-1][-1]
- # im_list += [im_dir + f for f in sorted(files)]
- # multiply path of respective recording with lengths of its files in order to ensure
- # that each entry in X_train.hkl corresponds with an entry of source_list/ sources_train.hkl
- # source_list += [category + '-' + folder] * len(files)
-
- #print( 'Creating ' + split + ' data: ' + str(len(im_list)) + ' images')
- #X = np.zeros((len(im_list),) + desired_im_sz + (3,), np.uint8)
- # enumerate allows us to loop over something and have an automatic counter
- #for i, im_file in enumerate(im_list):
- # im = imread(im_file)
- # X[i] = process_im(im, desired_im_sz)
-
- #hkl.dump(X, os.path.join(DATA_DIR, 'X_' + split + '.hkl'))
- #hkl.dump(source_list, os.path.join(DATA_DIR, 'sources_' + split + '.hkl'))
-
-
-if __name__ == '__main__':
- #download_data()
- #extract_data()
- process_data()
diff --git a/workflow_parallel_frame_prediction/Workflow.png b/workflow_parallel_frame_prediction/Workflow.png
deleted file mode 100644
index 5edc5ed807f496597f52d978d2cd28532e689b2b..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001
literal 195317
zcmeAS@N?(olHy`uVBq!ia0y~yV7<)1z!c2E#=yYvxV}G&fq{XsILO_JVcj{ImkbOH
zoCO|{#S9GMLLkhTKL1h>1A_yDr;B4q1>>8$<q0xRXWD-#PupFlx>jZN=}lcv@8$O7
zZV+5{RDfrpilvpOLhI@GLVaJFgNroX`wBcg_6w*xWGrLfz{@8gE!`uXS2Qu_<g3^y
z&(&{@_dk!Ud@?n{bN4ot)ivtd=Y2b8{k|q`r~SOzFMFI48W~s=7#KMm7?=basLpIm
z6q(Sr@jd6w(#XI*%dn00c}{jVa#BpfF$xQQ-1q<caiOsA9L9crZ1x{p+E5o`ceS!c
zWA%nts)}cA`<)N0Rb*)VtfH{G?eO!L*LLV)b@>4U?h}Xa*zan`=bN5NMwZLR|7ag!
z#^dn=1>6cV?Cu<TA&)0)j_qM+dYSxB^bsSTU`*J-D&QlZm-q#%PoWx<;}{NpG5pW<
zkpWLk7~EmvSjNBI;1?e6H5%$SB>b9D-&(<dCtfA;7+Eg2M@#&5#vSzsW{5i&)QaqP
z%6FI;6PJ48{>6f?)?J$JIT^xN%NE&v@2K5app4ay#?N{NPa1E1{^#;Ig`xDVUGllZ
z?C)m(x&QlrZvIx|ycKfU3+n!MDSf+ck-S<q?sCB0e5J3(6W%)Tt<{@<?{V;w##=ql
ziiN-KbpLF9@CG!1(1N8c&cXgm?4sif1%8}<w``@{{;W+BDc9eKcpne=a<aDNb@)Ps
zU$yH)b*g^`+zy(*NNUI3^yjKM|Nnlql>hWCBSP-})fu1X@`ld$-1suO>;3%S+b<d{
z+n3CdetWy`{lD(pulfBxIc-<=<>art?fWdhpWOHS_@RF;weL25$TM)m?&pQV4hFR^
zblSHp_f_r7yz_9<ni)cq9mCpJeJ=QGw{NxO_n*vb#Xs7XO!{#7?2}Vw>uXoMvlRJT
zQX#_JfBI>#Tkf?Qi`xh1-4zzF%6lAqWX9k8!bNu96W)tGSHzk_k4--?gZDx2`s){d
zXYR;3HjyoyPkwLSw7(VaH>dBv9=~5-CTs4|y?&ks*K&5IAN!p9u3_P_yXkhnt-sbk
z-K~CWvfTZ<Ch50Sf-gHyPnCPO|LKQ0JOB4~9qpZ4_j<d&=UsF2vsNjR|Nq|PJ0HxG
zzy6{`@ABt%_wS#5SiAa#EY=+HAci%-_-_nr`@>(Q)w4b^uewm!IX(KBYADZD-CwV_
zuRJZk=49Lq@oJlMYq(Qx_r6QLd2QRmUHKXIv%O^_rTZ+W?J3z%n)u@5#9MU_9bR8D
z@3+*u5qEE<&6MwN&X;MtHw>zEO*WY~?{`hTx%;cXkL=JA1T>kXt2ZS4y1A40b?EFq
z>t&yLr==X5!k5`Oz3$iB4?lEk_IrvocVCZKBrdV8T<ouh+qbUHUw4#O_w07Wnh$!|
z7^hT9^5!e1@Whxc-f_^hl|R<{vrp&r*s14!?0=g(YoC$&wqM1cg;)RYS9@}&WR30p
z^;$-(cUQlUm{DS6eeez6%T?FYcGS(kcYRLjhJD$SPA_MBJ)u6};>^E$d#Ach_Tv+Y
zome@+I-{8NOZ~m86ZS7(rM>v5XvfCrl69?;hFia{_p!~oQoCu|@8aLlM&?tuJik&_
z@pxtK^`}1zpT_S|*tuuj<@7&SzwMe?y*eiHPJVm%#tVDJK9q&c+4y*$=t}j9+0*&`
zE#K;kZGR%SwtRWe%b8+T_b$CNe)#|CCmDU2oVihsrT3UuX?-@`uJ2*BPf2}S=<Y|I
zw!SyrrD8ix4rYC@-oJB;^81k4<})VNA3yYD(#(}<>f3vl=&$w9l|2_`eSLm=*}nZ>
z{qL`w8+!fa)_)(mdTLAGGG31VX69!d^W5unhv~NZTGvmFKPB&~W|_%ld;h#W@7c}R
zx}yg6y35{WYX|i$yUBB};P1c1J9NL!SHGn`-DLM<<3~Mxsxz<Ms@C(+n*8+L{VPhQ
zmww5;*ePvY+vdLQ>RG+aBK2zKRe4d9c7EK~9n|$%_$>3kgkrmS*0bI(`}QURdzt2A
z!^m>k{id~jteJEEMelXD#Z=~AYTLeZi|yl?e|5F}w^elg-?TgP;Jd8nJa6x8+jOqd
z|6O+G&*QSJ=Y74+$}^8UUw`=P?_+mfahX-+&okeag_N9o#{d5QW|N;5x$1v^zn*&C
zvev8TTHAG#X|3nwGa`3<J#9R1^UM;)w|B(jEGjRqKH8Uh%WnOBt9$!qFuj)kBy{F&
z{jHlNJInfBZ94w<<>F<hE9^DhuCHRfEWccyvu?x6DqYsSyfu03qB|q&H>Cb-D#^al
z@ayX9zV$aFU-De5dH!wBg?$wt6JBr1kDs*dV`J#c<3DcX+vh}t-Fh?g%<VmUsy8Lu
z)i&M!n6@`PE$i1rkKbHzBA@&%qU~%a9Isz<PKWR1BjMSvznSUfzvQ}cIk|W4{@)uG
zZdm<X*5c^w<G02B{VlFMzo+uf#Jzdp!HWK8PRZ_xzi0V<>Z5Cw((mSNvopT;;-b}8
z?XDf)mloBTi(fKa`Flt3o{|-zHrw~_{IWCp(|w<9k~I@0GHXBW+W-7s^xK}@A6H&F
zzrA`&#M-~?qJAH5UEJ2X^!}{x>2D);eEnJRuH5TZu+q=GHfjAn?Ag&r1{{UEuboQM
zIk%}z_v0egzU1Db=cn%6P@d8|w_?S;srR3JQLlS>bd!JX-&1F%AF>VFQMX=egM7Dn
zdD+`lMShFlBtCvpkt}=FJKyMaYJAn2d!JuD+{VPWWckf6zN<ea_4{@imehv4oO7nJ
z=F}^{{|~KAKmSWV`=qwUp3B3}N_LHCxNrCE%{I5gTqVDGO7~fbZQCjGxs`iOs&;V>
zySD7UC7;6l+v3FbajV~MYJ9iHCD{0T;I&8FYU+cpme#qxNaiWJere5&Nvq1@l~@mM
zcB;2|ux(GlH{<Gn_<u_yw<Q-}I=j_xebN%1H?ATlw*F0XcYYYI=cD<w;s30cACFeA
z-YsLTZJl25-;#6LtTPtQ+}}5A9S;-tjj!>2A#|?Hp#Q>a_FK!|yj%TpkCpPxm~%7#
z$%k2_SN&P_^4#;&cBkLFsdpYY#Pd3ExqgDxr3dHM=6zpkw`XhFz4hzoZt{NpeYf-~
z@l5-oHD|WU?l92)kvGX)u3YC{{PS5=|D~s%p8qH2-nD~0TmEe+-SPZc@%?)Nm7MIC
z)FZ#@mg~*gGFj;U)!ySWm;23kHrG#G5Nm&-N@f1<Lz>r~9o#pYxmWZ#w17q{w$<7W
zRPZhf{+26P%jdnQV4I4RX7TmQHCD>g<8s1+zHeI^DXO|Wev4+OP)(+%Sl5lB8QZRF
zFF#)*Z@<@gj<%lp!l@OJ(m{`wzPD9BF8L~cckwIksL1K(ZEGI|of3-=x%1$h=DOmW
zX-gxmoY%ixfB*e@zvpjWJZf_1O9>R)^6~R$;qP_Z4gPkN>Gm!x32a;XO<H`p+;y+U
z3p<%x>us_uJQ)AfUiIVaV{3l-eroxW%iE4H{#<T9e}C(?O3mETpWi;XH649W|Fo}Z
zd)<bWMgQc>s*En>woSYKPUpsv=UbkN-+uMHXu)j-?|r4_PuK0=qp+^?{NFcsZ_N~m
ziFKa+=U&x4<<u>|^WtZGxi?+U!_G?m$;|6M-S(xoi&M7;Yf4K04M^B;V!v+Zi%(~q
zOWrPgGym=R#Xh;VQ_XMwoSM2){bfW!phmUzrF~nnEAGyJn;^HamS@}Fv)#dFfoi`K
z^U7}*tM9I03g=t$nO|uC{<>4Yq@Sr=`}8CH-^Uxh?$UQH-%l?U3%HzMF!%JAJp~(5
zHRG#4&uMvEzeDl4^SxzHPt>lTQa#W1LX}}{`LoOYobo(rS85}_@BeFSKHv7@o0AW|
zhU@!O&fjN!@c*G7)kp28efa1xePUIfo~`kXa^p*ei)FDl(SL}((3@s+Z(hD{;F^fL
zhLJIb)q1v33q6F*JI=I3yG;9=XK4BIv{TSNy(I_!3(Nh!-uyd~v%GW4f#)xaZCfve
zcdkx5BRIX*-c>{_(tXmi^}A<(VqV2Dtzu_tosg)-&nHROyx%>~oBL~%n8$N5{&nX#
z*!(nK=?7}|&I?kzb#>yG)mwRE>!)RHzQ13~)?|X5&0dw$QZmuS;R}}cu%~YR{WE3q
zW%2qY_wGJE=2-Ht<=gH1W>c9?J}>SzO7_~~VD-NK{iQF;=4U;BG;DVk=lk&g)2%mm
z7dDpI?N8ga_u!JrO{sof(KfcpJbSmj|911+)%kA&Dn3Zq+uHt9)}3(f_jX;6H$OK%
zj$VDqtWaTY|K}g;H9EM{er@0N$IR~7>C;DbXMS1SG?n?iee~p=yMJeV-+%qghUuTh
z{)aOCeKFxit;weP-_?2hqSjAtdS6|?*4T8*_B)QRmbPz`U3*^E$}I;+)u+>FIB#a%
zX1_J_rm6jD(-cwgU(}J*B+%%&$b0EqiP;<++_Kv_T{i{UZFW2{DJ~$)DED=XR$<#_
zuiuQfK6f{~5uWr)cvnMkMXBGySGQKiPLKGT(9Pj-*6GD7jj+3FJziTYc6Tt(aGtl(
zFV%e6$Lsyxdm?%^O_My|UL~d#dCAmvXZ0qZw+if&g>|glS8EHzUw-(Ob4k{7<(J{v
z(s^$`X?<c}ZY6yBmd&=b<@akA9i4mRZ4IOA@~!);s$#DG-xvS9{AcLt(ARx;{U*1)
z`}6J87U!7m=xpN~ySAPd=c&;@{f_Ti`?mQrlNRcHeNeL~@ovBQu4KMzr8igh^jv)#
z{$-PkefNb&A20syE9zXf{$Eb}q2_j5^ShS>tZJ&h-)+73@<CTl^!nlub=x1}oF)s`
zMDBi;ve|sz_R>j9Y<`5DReEc&J~Z_0$xq93inguix3#}eR)7BN^)udE=iCaqmRbL{
z<ME->|K4XjGS~CEUHMk0`bcfD{Nk@WD{~@nlzGb>t~4GCJfPBPy5Fqp71!PkH`h2V
zKO?Yelaoh~oGW{AhHY=tvnhhw!3X9%Ubp;ltBf`O&8vLu%k!Q*J}6V0cUs=+Lha1y
z*S=nf=VZOL@=TkT<juy|q?g}qoSwfEiu}m9abHLC(L0ANbq{Wm{JKu&UCII>{l9lx
zjIQmiS$A33plzGw^}6@}-H$$+-W@%?fcMjdy5D){&u=KZv)}0LTKDZ&V?9exudTm%
zb<3}rwP#P~$IjUD_@U`!`**kZy|?q9bW-w4jm_P(H?OMq?f;wmRJQYe)#6W^CcQi9
zy8C`qeJ{JETy8w`&3UWc>z%j0IPdf0*7jb{X?7o(eWo53KK|B*@#+!b8{hUWOSk9R
z**I_Z@84f<Y2IE}zsaOi?ey7SJFbNmT<1<@jrsg{<JOz``lq&R`+mo`tM}rEJApjn
zzBu|q3lB(eFB3oc$jHCxgUMYbwaD;*4K))#c*e}1IluYv!%YDi9eQ(5?yx;xo+eTk
zJ8i<%l2gU|I(`MUcuBr7JSL_1^1jg7#@a`#7F|EGOv!Ji-SUQAVP?&zcYFzdl)4~2
z@XEUn^*xT)wT1u0{Q7&$UTd1o&h<Zq_UZfIuv9-AGd1DbZoVs}sc)7mWZwT+5c|gT
z@}pICpZxA#TKiOXv!(an^Sfv1eb!v{?qc?)?_YA|cgg?JeY|C%waR1LExz}2*Bz|A
z{yEir+fL@fm}{voQ(hO&3b~Sa>EyrV-?zuOd^i2FciH~Gn_p%d+%Eq3B7e?BIrA4!
z7~+NVFKxYFo%reZ!*%Abo*uZ{tv<EYTKwIOg!8_VpXTxJGkd(#efzcZ6`vkIQ#Ccw
z)Vf}@|K-X{=imQ$wPydaqmoVbe_CEwuju1FUd(^>#sz-tCG`yv3G0JjPVbxV_3%}V
z;Dz-I-xSXaFl<j+|8mP+>*Oe_<uksi-(Mudb0#=bV(V?MrI!q4euh5n&a|;S^me|E
zUUk9Z3p=}xM)zCT-*ilIzjw)NPs+FWM@#q5(~ypJ7w%v?U72!Bviej*!SCXi-@h)-
z<=)Y+z5LkQ+Go@LlvmE4E}y$@-r<8KE3ND9*M9i2f6m0ee{z5Rx4HJ>--l`G{_Aw}
zzoz_o>~Xtk-@z~YG_S;bw!WrOZfWC{6S-Oaht6IrzRjP{9ACRLO5yL<-CXlH&Q{ee
z-H|)-u&_bEq0$@rF87<$FNdpan^@ISxku4*O0GGJ@j=nhtF3-|(w@(2d$(jORp#05
zJ)bYWO#Uf<eSF>C#Sgjaw%6aiJw^Vl#g^Z0o9(7QfA{-t&$E3Q+}MV&6b$NE{nT0d
zKe5OBTjBbn=km%qr<ZF~{L2q{|A$M;@T&OqCev;1C0}CroJC`-nqT>QUk{l3ZGG^*
zU!sSyy_a)(S7(K~8>{KfE__kaDZWZqc8<%Y{jX=w`PVW(f8|`ks=B-Lmc{6DSC_47
z-Tb84+vIQ7^^DiY{N^`iZ}|LAwQ^hVt-?5Si@$~oe?Q!|bKURoX=?Q+o?F>^`B&>E
z@7Q_ze`kZ`)$shs-)?JLH49JtWTyWo`JnLcS<_11&+U95pIPGjC|NVCHabIL@71T5
ztAC<Ha~VJo8qF_no?3`yXiC+P7%y+w;mLJcl>h>Yfxbc`9Ue{(b(*o+omDB|mPx
zXlEY(`NGqg?P|xRo4?+bwyLqc&uV8LbNZ&ucZd6DA4;kJYQ;K4Ct=N$tUMzxFWPl!
z*gM5*>>3l~tXooJ&VBS<X*KJPNu<Xo{uZepI<<Nhv-zA)r`<8LUzs$)Zpr_fVc%D>
z-(NNFxZ~q;|1Ii&ld|uo*7#Pf*_ZfMr+RK#?UcV+(^;?E6t3{pudHZSn>5Eyxbo9^
zjz{V5ubr!T=fFLE&%*m6=bq@@-}~>$(fEpoPh79^t?xXq8Nyuqx$|4hJ-@DtmKA5c
zXZgynX@B=^^|Sl$R-Au+(syT=<^0{76#i+f_g?w$Cv$#?-lO6v5A*hZ`uKRh<<(bD
zi|qDJ3z@BG@h~lPxBk<;lke7sY}s{o`v09z)TaAyEJ&70EVy&4HH2TS;{V&_R-d(Q
z?Roas^xTbIW%CVh?PJ^Ad&OD$dcoTKS=sBCy?0gq{{O+VSK8a>mOZNdH}TO%*{{L-
zKR8u>sQqLZ_sjTEakH&?wchXCk7xbgXP<lYZBI?a!!6gXYfUUS2<-G+Ht#*td;aT1
zX6J1$lo#(f)>((6EbTiWq4`-UctgZY-@-ue_ZMDhH1JKU5j32#(2vpcq`&el@wY3s
zY?ACYQqunD^z~xt{@)uOcG<tl+0nDD*}fy-NW{-mt(x*J_pEQ#Ij?{ER(0kzAug-)
zHzQxpG5tER*^=$lt7k#7wYNXou`Uy?+;q~s&q^#fVOOh~;?@FAF}w1!Jii&^J{Mix
z^5fS3J28(xYajY>;n}R$^<VsY?G4@DTNM4jv;W?9ou@B0v(M6BFv+y+)VBR!R@R;s
zXAcQ|_S|>VtEky?8Tn3pbNP@zBUisVDgW}R2CXmVHlbDFt9<X@@tFDhj$OrOga2oI
z`&X{WDPJO&UF0A6>Bd9-%UgHJi!9qKmmA-F)9saDT8DF0$LZ&PZXeBz-oJn2E8{QW
zyq(wWL#xX7c)q*vTBUw>^u~|zCK0_x)75XSySu}F@$c}Z^73zXO3!5!_GO#>_iKGf
ziv5qWQfu~St8ed}{6khXqNMNqd2gGgchfDdr+>b~Tx;H5yyk^8*3q<t9FCseGbIa)
zJSS|{T4<EC-Z3;e)9K2F#%GPk4m=IvGv86P<Ls@Q%+o6#noVBwYg&HXoK20Jk8O3R
zn&bISXP3=m`Tx6=wobYILCAN1RV3FmJHywh{|pzT|NQ$kc$?0d4KmM}W9?Mim|kvs
zQ;~dk{kpSt^ZF%Ib}rxZWv_Sew*rIN%}=hHJXB%7)N6I}xBh40?Y4W;J{`X^Z~7gV
z&97IV4!U>IQ>j8fYrX%L7tN*jDw9r$=bfyt>$)4^ZO^wt=Ec*Ev6Hf`KRjCd^YzQ3
zP<3zH?=cgFZ~uDms>j@qy?N;`zyB)DIzR2L^%H-elDX};Pu|x}s>bZzv#0LiDQ^~?
z3|U<{n|V*o{52Jmjc4ss+Sq&LutWXr1IZ8NwJO%?R;~O!_lxX)zh(Q+TKsKpx_ROE
zO&^nq8u9+QRbNZbvz~m~e}BgBGcS2cr`8K(zT;Bapg;9Q=D$OaJZ&3G_U*G@>@lx8
zbl%#X2G_UE_d5FHr**)?56d<d*DYGvb?k%C|D(+JlTGZtJKXn`GZw=>)Z3}+P`J}@
z!<xDmuSJ~p%yZjS^`$SYcFv|}_p>9uK00l6G(zO1#m;{}4@bXVXZTvR{3Y|}x_rZV
zoB{3M4n7T)yBSt)%6_^3@9#UMF^``9HA)Xk+jCGfN<U_|+NHXkw$pzX>|p+NrXqR6
zn(k}u+n!z0tj@ms?Pq2hJG;zJi|gmM1sYsWyY!XoPSMPMVO_cCS!W-_%}6Z$^4;d`
zyM}j*z0#XbDrlbGyjIsHG4;ivr%B&J%Wvw>mb$*;>%oIDToT%a&+dJHB4kp%YVJ3_
zc!&DTxR}X->Z@zktUR6n=KajB*D>2B>&m^}(<9-&?e2?f?_@pWFWzu_bntxE_7}BZ
z7CW5hXJy_iT=)H7!-OpdwwRU06`z#i+N539z3w`{a+LJGZ_UByR6K=V^`sQHf3w^A
zdEx7QZWnFJXKOFGA9;1+gHN2XFTSg7`G4`(o#t9|=lKiYZxa$Yul7IX{=0q0H)5Y6
zG5E%M%x$Ll`sbkw%~y%6S=O;S#$wft&{G?xT`g}tzdugnUxHbQ$>ZA*j}~->WSU*(
z@ZOVAs~*8qH*3w7xZCe{_brR(>-f&w)&6bz+=Q~r%a@B;&tIId(|k+sHe36{w+_nZ
z?%!|9eneltBm32R8}@@QFTehIh_(M`)nvyo%gr`h^zE+2E6lxp)@yUgC-pB<_xhHs
z&&W~!$YFl2@gC>?-}@e4`)0X2{x;wCXW8rLUHt!D?aM5XpUv4nuh-hZp1rmGo8&Ir
zsQJ|!-lx4VPQa1jA93{bCRDvXX*JR6Z9~A9l_CMlPiigoSyhs_^m>V%_x=Edutiaq
zvs6Pr-ct)%>ayduMTNwi2IEQV1=2a?*Ofmly1mu-?ZG0gi81C8vmd@H-|*Y7<uhca
z#^s0KV@d4Ay^^qm^udzbImV&W{&mQ`R2KY@zxbrx|EW!{K5NXdRo!%4rZ4Q3C)a}D
zhWJ?Zl)qsM&1bvjxE*OMk(=l@>8dvOqwj%#UmTF_o^+}`Ys%rP?*n@Cb-o)1=BmFc
z?tl7T>lh#QrqLFr<i%;fmh#{8KKMoXQOHxhiIbIE0*`R6H+R)2+IHde>KAu{6pIsL
zJ;hh$NCqiQt-8E2)@h<wX4AziMeS-$=eq6-yZyQL@0@$CtZP+TXv$4{=YzFd7k)Tu
zzG&r4xt9_5^)mhcJ^i*z)v_%3bpHDJn~X2LzQD&MAYsXP#Yf&U`wl0zQnzQzfg2}$
z#G6j<T9H)L7F^P?ILSY7@s15AbbadhS<k<{n<4T-EJ!Q&_+|CBZM<8|E~L6CUfj0o
z)voMy#dk|*)lEz@;r-^xSdir|Y`@B6;xds1Wp;`uzX)F0{^6qLtb*XY#czvtEt&VO
ze8#QmT}A0@HB5gW`J%V_gx4&+mVB*P`;4^(yS}ZNrp@(8>gDE#YxIt}qK%d*AWa1o
z_<>qV>t1O@tc}wv*(L2VXSs*YzxRt5Ssv5f6m{yzZdX;64<>V`A9*7$<+97#n?;9J
zM8toYm6qqB^~(YhY-So=w@jXU>gbA)^w^hr$CUyFVyEsgY1CkUI=7^^+vvh<4VLp;
z&u42qyQjl*&uHVt7*&^xGd<<GyT2$!`hWkV!a2Ec+XqeiB|G-Ie7{-km(MJYeeeN1
z$9iCe2q>K`o2>CVMvS{9DMJ5I?0=EMbFofW--@=j?mjwASLDSqp3jF5X;13bYEj=7
zHse<5qO(hcr<p~1q|N;xtt$Az>(Hacrn#S$sx7r+Ii?3Ly=2&RVP?jCX@Ok5Q=a@k
zT1^*l1ly!8|ITtrbz@p@;Ep4#hlHvnSs$cx?Jp9%S}@-wT%Pk=YdvxpA_XUinQ(`*
zXKsR(>$#Ozym(lzYHp5+b!>d2GmD|5b3*t!u~u#E6&osrr>19$PTJnd^z!VK=zFSs
z_1CMf<-fVLNP_EZbj$I%ntDG2H%$=HQ%n}mOEpT}_B`0lI`#BrL(fwSJ_a0^_wV@s
zc2;a@4;)MkEXU>^kl<SO{l;mL6Z6~(UuRB<S{LMGbyaivx4_P6MysOZ9H0AfG(T_g
zFy7kLvdAy-3#-O#TZIy~&>c0aE>ylaEcs&j^oDk$tG-h<`bEB;vr+GN$)W5EoqYL9
zzIj>Vn+(G>_StaFI=|$1O3=<3kxCB23)sY2&YQZe!CpLo0;?yzaifHf{bxIsRg>2R
zBt#2MS*bTIxFB}nf^vr+sk3~hsPUVAP+7#bs(zgr*U{UmXC}@5_EO@i$Mg)R;9U+c
zc-^wxw_fQr$<$2Ux4KOwae>wARR?{(Udy}`G*eAysp<Bo{thoDHuwqQ7%c%g!AGHS
zqpZ(!aZw+>aNdIf<sXjD^|J~Wz4hS1u2~+rMVdF(1q94dmsy&;r`Bq9%v0@^KHXvl
z(^Z&R&-9i|=_{0erDnotTVh?Yzw>jy`2vr|h6y@uMMsTRv3$r=s+KM2I>WgZWDPV9
zkeGIw2H%=r&aYYaPI;0`;`;<Q%k-(UtC=3vHb1oez2)HTyFXw3{c4rie|_ukp6<Hh
zA93sBP4|2~n%NH;xbTnY3H<h;!rW<tW9{<mCdOAUWnKyVx!izz&Y=Yl1h96w9gZ+P
zHlHD9creoR$(bCz-$sek_P@!S_3GnlwQk3<vwj-`=Kkv3XtQs%;d5*0Rd$n;E^imP
zZ~r}i@yC_wuXaBy-?(^>@<Z+Qb9OD;b@#Y1OSeSxovnXY8(&Yo-J5rhKUQs1@CxQ_
z*rWQKAH%^f8FxZf26>zco^tSCW5AR3!JIR9mQ+TW#PVF5*1WwxXSH_8Z;{K3ucz$)
zHq(-GD|77S8!wG(UM~IQ{r!5$Q(y0GA%WL^ymr1B_c}YcvstZ1G|4l}r}LzM-RJOZ
z>v`8?<yzC$iEiq?cRZtZUFWa7AGLSBXC1v7rEpild!OEXq03LcNFHXrtiI`T#@}~I
zG1v;f1RYL=8FFWDPrAalYTG_<37hEFq&RQW#7!&B&P-srd^?8m@AP=R_?<hq)J`;8
zqBgz$0K>aq&AO8JZ%34tuGSVff8E5Yw(!>0hu1D{yZ-CJCkc0{7|T8NcXfUjzWwnt
z=Kg=TqmPdH?f-gv`jU6EXQ#<to%!kgybS-Un)1tYGAH}^^2biAtkAgY_IS7Csprxw
z)`P~sMEUYR?B6dRJMZJETJ0eJ9E)qygx=|^t8KdM&}PAowE`5_$<Xw2VsxSHS3Sj_
z-bd23ii1{MyI9AZy3)UYS$9G1ee0VOe_Rbeb^TiTugNQNcHc@rb-7@AhyLDxx{KOJ
z*LquKTjey{u`kY*ull*m*7ql4=!+dkPV0W2q~&ye?)tald|UKNgTy|$2CoTSY&&D$
z&MOnQKAe$!a$%J7-vIW@?vHm%u6*y3x!HQ!`_xO`^VaKm+a0`}z3jGh#oh9mC;V!4
zH(+mw6_havvL>Bh_gZm}R}izvmS<K^pO;T`nJ4f|zT5j~eSq7qQm08`-*_}E+@zjI
z>iplmH_wp$QgH6CW%l={{Qi;<D{_5jj=uPIy~kzQ>Brvx`Euq*Kie*~Ti<Gb-zvEy
z_~&4PXPVDT7p;$%t|ywj^M6tveE(tp!*Zw6OLsTfeE6?D`-I=(kJa6V*ro>?H)<G|
z8C_PrD0GI){{NvgFP8JdK?mF)&vaody|p~1C{E?1;GdHBu3z7uNhwaAb#=pE@%iWe
zoyeLpRiCe@bd&G<ne)>$YZF3mwTGYIaQ*)8pr4-_^*XoJX2i$Hfv5iaW34xrcTS(b
zFW7qZy35&rABsx;eszm^+5R`VbINv{n>tM}JLmtet|!c&jdz`1pR?qWbp7qiHoH%+
zmw2}P=#|=B*HafwW*?NUlF$7ue>8XHf18VTJD1#?J4Z0RMCas|YxjyBgJz29)F$}{
zf1BtQzQ{kY`TpIrOJysU-`aEVlUccrkKW9gS!G$ZuDkzzm~#1FQla;|i&f>Xrm!TR
ze*R2-8s8Q6>G_op6JB5GGFxxEV3kpP@$aT3kF|@EqJQq%<v0Ih`0dwe@75NjeynQ`
z-&m8M{FLe5xh~_ff=Hoz&r&z++ZFxwU-^Ra!@BI7-`t%okhA-7NV;3-?(6eXu3G2M
znH#aGY}#JGY;%$K(Saf#545d18n=Jd@9$GrvKI@V)m!`D&)?zuh3Ko?FT#yq9uH4@
zw|QoXt-W8N<dQ{auR892A3tl?|N8CbXRIP#dp*z2Op0lKw#&+X-><Hw9d2v=<M_7*
zXWYNhhIL42h6Piy$_%}?OCG7;6j`IU;A=zak@lZgkFu!ChCbaH#WT0*l}P+_8<TbY
zt!v)c&;I@YP4Hy<zGu5qW*_{1Zm)4%@0VE_x7)u}@3LvTw`h9Ic3lsvtFQ0eJN8sJ
zma)2O&Bo8=@@68Rp7h^+KKIet$GP19|Lxq9e$VFJiMZ{nR|aUif3B;19;4-NoPYT4
z<!kB3q-~?>>$o;u>Sq$)8}(+#guc1|S`TT)SzXbZcl&<KuRFRw^FI~;`F6N1b?Mo8
zCvTO$pZ8{l|NIAedm`V@oH?t`#ajIAEcLp-2baCtDze1<L#g$O8N63##P3tMD`mEz
z?(43{*=1o?>8<ygza8ynkDgxpd(xI({xx3q-~2qQPt7kgi(R)hvzYzSzb}E8Q}@gl
zS)6^f>E*;POTP;4diVaWOV-M-M;l+=o^M_Gv#;yi&0B_X`|FNuobvxu;@z{A??3<B
zrC4$Q&vRA(_t`r;=DuuKzJEAd?fv6@m67!;RA1-+p2b~RUDvks<GZ6LZSQYS|7M<Q
z4;t#8tMX_1lUu8tV}DjJ-#Kl2<@W%CTJG-)b)BCs`Py!ZwI(}oL_}hT*`W->!t%Xd
zTTlJhYWZI&b?K-6_e-~r>bgAeT=;9ANYV9Y_m&^~H!n;sxp3On`$GEtIjeWiVwyEg
z^5nBmsW-1BRb@@x)aqMq7ys?%{x?}sy53Gpc|I>T4nC8r6Lj~TcWp)X+dTnu-{-A}
zb7?Hmi&b8GW~*#PLh7yA;r{*iU9W5XejOA1zJ9UMjN1KaRW%=7gLB%yZ2Iu|t0m_$
zse8}Quab%O7c0%U-<&&r>ki%f_d`l$*(+A2c!Vde%l&^vC3tDP;I6w@&q>y<&wukp
z(qh87<cNv6)-~^2C!5{hWl)$s$MfBiTc3X2f0Lo{MCUd4@`%~t&u;L2lGw?g_5Sj)
zs;?!#8STU}<L_*Hf9d}2&MQH$-$zSpP0wolV!86!Wb?D0Hw?|M3Z0pCzi89q@P@hT
zEvJ89ySzx}^5nB$f?EYrW}MB=y&3k5C+)-Mf734Y#Xa9Pt(@)qXCYrxZ}r<zKddLs
z&sg<n+n3Ybi+0A%lMg?4+-ds4BCfZ>x3Sfm3{1lQ2{|dtcHdrcr+IJ0i?el#ntqBW
zy<dhFKIrvY>te?ETws3`--}|04Q)nOzjL_O2D9vX<?3!Kf1UT^-gD9AG0R-EKW^IQ
z-#L9>^RH(|HD~`zE`640S9cpU$9%|lp28I76Yx3aqR#%Q^)_XXFS?bV`<*(M^-78O
z_UC6_i1TgxU@dBX-t)%4cIhI!C%I95cfNb=e14O8t$4m(?tK5o=lorI=WSe$^QZpL
zsh3Mlc)gP4%(?YC-nNoCwbTCpNs9g(!!_qM@1`fw<-zZMo#uWtYsc#Od!wx<Gab*)
zm~8yiKm9cO=2NxX^?j<|dyA{c&z+q1^2xS6Yi5eozuNZvRNnF;y=}3fe!rLgEIhmZ
zPT+@Y{AQu&{vY=9ciVOL!lzqvS!eHxF8}W>`AYD+XjN?edo?NVb6?hHl+KQ~zqD<}
z?|n5{cjA6=H<WMvU9j!nZ1>yCvVJdoU0iuzds)P3{;Rt`^_%ZfudeA@a`mBo$n~qj
zKKJ8qo3_5&?z{VW<E#h!BBtePKi;2F@$_5McK`kJ*Bp_UyLkQTvbzD6nST>`u{IP8
z3R#c2rB!9Q&FSb~DLS`Qx;5@kU)=TIB1fK`xoM`g-pO>js{UHd2X;A;uVqx8`(&*9
zeAl(}dag*6t;_uKmcYM@Z!D_!zM)f>I6rIg=f`P3U%y*cb8gEsA^q)V{JGME+pBl@
z`}hC%;9m0i=WQ{SZ&RXvUR$-rSH|P{<CyPryb5F<zYUwd|M;t?kEZ!>IYe82__x_~
z=I)<+w#679jyrezUPIOrhSEi4H5%9DLo7b;-0wQmP&G<n?)UxWnZE>2UHS)_B(Lj>
zkDI;M+GN7H=huC<o6jf-`P4q`l9<2FdiLmZ+t&ZP=euQEzFcw22HpRx2mIHaxFczG
zPd{!#%)du(S!X-<fBd#{-}~2Ts?&<?=QU(i{mqEB%HO`5dDo{ff2sV>L08{-FO2$<
zc26(2<oK1+Uq9Q8H-7u_y>Yv8WZNmtN!1H^u@_H2#3ZC=zKs26t?5%2a3n29bT+?u
z_tVq+weRH3-cYK--TY8UYxAb2S1X?fJJ`-<dOg$m*QpTCr8z&cxE9DS{PjXKME<Yx
z!;ozp{rduMv_|Jxy%H9G8s$3uqng!m#=pzkzx;c?adYo+*~>q^#XLIdB)s^W`IMb6
zr?p)u4mq}XcZa3y5+jYh=K{kD4JO}y8r<9%wC^2g()qfTXOLXY>NLUp?Cw>oY$YG6
z&uV6TRrU5xz}o2DH3tuvPhIiz2tS)~u+EgFa&!Dcc+2fit?RRlxSqPc<j0!4{fnM2
zFxkGi=k<#Jx4V`>f>u3h_uI*9e@$C?GVRyGj1Ny^?wea~4>YTkzJ9xY_Y&LR$98s_
zPLZ8|&i+op%V}cM?YCVO?O0y*bMpSe!mN|t?a?pyXDhDg@jL$Na=od+4~^Z?JX3?0
ze0R3SJ~sMMTcGs7`#%9MHfaRR)mh>&ZAZ;Fw}~Gdw`R<_s}(a>ZejP|2Bw#$P8U^!
zUKKN`E)Ql~?IVBH>tLxi>-F8&CmCEStcnYDk9>R3Q*C8TkJCc`Qx?mADGT{4Y5jch
zQ)7X|gCL`~A~wr&m(J|_xP0Eb_X3;NIhAa?(sfufHpzFhqsX~)OU=RqfB(F)tjq3k
z66dtzE6SgKILs6FXz6?A=r30P0u|pRO<L;D@^rP#Jd3<DGf$m=lOHMh&B^YDaM=6p
zu2<A6Hu;_Fdd_`0_gF{M$vb7ymF|zW-&?U~&5~-PE5%ZgTWWc~<Sp8|rmTeh@?)P}
z9rNeTw=e1ZR{MD?_vN*z|7|YrE66YZIgNk&v{DxDJ!ibTS7w#{OxiZB^j!7+J;_|`
zm$!$PT=$+oUoKd#>i)rP?Uyn)R6n%Akw(QOln<8hEm_Sf@+M9-q`o_2-Zt+aci$eG
z|3oml>)Yk*Uq8OAo^-_Z{<5CZ;CTUGCc7v&q&kK~1?QG}8U;VN<HH!4TrgWd;OaYt
zT~3ZAtxV4+6>!Eba9uH_blvn<8prNw-C5_g@$Ylnnb%6Dsywq5Q08)7RB&89@Hcmh
zQf({K`#aCpFi*0Xb8$<?Q^PY=Cv5{dr!SP)?5NJ({Qjd_?C-teYWwBa`7e&+yR)|<
zC~`;f^P)E;a~CIE+iAVPCFK1)b)CJhe{Pb#eTT8j(R<#~$!m`rKjz_^?Dx%n@BXJs
z^G=vtGMwuh&%AebhHh%-zfVnK@>hHMOU^i7%q!iuc24ul)V1?U=Cyovn5!+}v)umE
zimcLKzA`@c0ry4Ec-V(LpLOWF&YAA+SAY4JUJkG^{)K%DgT!qnWhn=1b*<1vYl=0a
zS(cP;xu?lf_3KA@*U|fx{%<OG%>DIj!Gc?|Klc5cppi4h<+zEbh0mrFUXxmGD{`)%
zY}8?_QFNnHZSswT#jjT^*?w^qSFpBdQD>;@9(Td&zRMp|ey!m@xoKadx>&k;hlMoj
z*P~ZOO6L1b?$=x3P{YIeo1e$~)fdZ!bN?RA&s2>wV%Pgv^sU^YICb0I1MmAaLhqJy
zU#@<&c7Eo?|6=>U6i4nzt6UTKuBc+ljZJC(8}6J*_-p&QFngVtNB!S*P3mH)QO%E^
zzmC()i+=Kfh35<F?;{uNCQkBydw$vaMo>ZevheOV-R0hQdA_}IF8x>hSy)@HYPXuh
z(%6#oav3v=HXWUOP_<~?{M7Z!JVSWxZw4(lvu5V>f15q;%(W*g@7S0>6i%&RlR6(@
z#akjjrx|;LT!JZCeTHA=%PGR=W*#=<iivkvc<N=ywx?EIEoNQb@{2rmuSzMte6e_r
z5aZ6E7d&f~&R^2IcK^N5(!hOZvb&R7Oszdn{Hstrdq!l_3Bfu|{}Wk8tmiT}d7Ubo
zu5mjyKw$3i!!OUy=a}nOc|P#UoYToNR<Vzs-irM8pk?7<%RQN%)6Cw@KfTsw|9;ia
zn?DP$bo+aN+c!+=&&1}emfP%kmOpN$%wBl*;m0FEmV)y-O8U!Om#Z(RPKv)g>GE$|
z`vWqWB_DT-Ylgw+whyN)&A$2QZ0wiXSDRgKUr%_i7Ik_L%i`wtD}MHBHtyd<p1j-m
zT*yLSKJWC7nOU>GOMbI_uNJ}cZ;wl`>C}@a|DArfZ0Ve{_DA#9JbnD<WQVEAwUQq}
zbsN{}MSN2?KN#Zu^;`4g-EKvn7nXHBnU}$gy`JC4dMv=e_KE9?rtF-{?xz+lFuQK)
zza-+frdQn3Goe@2ME18b-SJ&?@2ZxbUQt(rzCy};FV)MM{MS}+P2X<)Wvztj!K<5?
z<2sw3ZVIRoS=WELQ*ZgyrcWzFl;^(Q;J@%e4ddJM3w|#R+%HqmyQ7DFt8B@`1*zHL
z4}@!X?d_W9olwB;X=o*%9+=IY`u<fdqpHM%(w|cjq(cK!*X}*4plx<}=B=j}xA5!s
z7vBD}WW}b&tsh-C@5@}b_m}(o5c6wgKPJoHiK$JkTJ!Mg)XI&zci7ptTz{K+>TdXB
zufL~#EpKnF+j-19eR972)tTWJpXo01bm3iTwpd;F+=HvPTz2i-r0_2zv-t3npjGw%
z4oFY!{`b1;Oux2F-08coPhXpp_4U$xqc3qmxov)Sp!&Y@eTw@dRg-I7VcUNi`_J7o
z?_O)``HSLf)PKxNyUc$2|Bk<_er~sZ^?CmWm7O`|y_Z3&^i^-|W2-j57<F%>#?C*_
z-(M{aGPwIq;AXIZb;kM9JNrMUU47-%Za%Z=<^JdCQm;}Yx2^4Ie);9!rJe70uAJue
zR_?QLl2%rqWzd#lZ}sC-GM_6&@^1}R`R*c(qZ-sTFl&Cf_g?3b#wV2`yXJ`7_;9{C
z_@{L0p_wu^(?gdVNW5yjVEuAW^olO`h~6#F*jD*uPrmT-Y>4vQL$9h%nS2lIEln=p
zW;uQ0vs*F;H@ukXQ1-#j`N#R|aVsPKJz988w05`7uk^E}UpVFdU0b!qe5;U#Ov1cx
zCmMgRaB*AzJ^bc7@3|jeEc}05T;<q{TV8TAzFXwG-;UGn->8xN!+Xz%A3xli>UUe@
zzW#0=R=e!vOaDFpR$kiu*te)XPW|WYyIsATr>%9Hc|YFYpf<0{$2E1g^4EyT)u$#M
z=c}K2y7kWjgSTHn^WD$w7Dv8|o>uhbCi7YGEhVgPM0V_-;?*4^6A&7?&~EaYU156W
zPJb_cUs}C0aogTEoqK(?Z~kvv?)Bbl>&w~FbN{jAt>4NS8*i$+@0-2(^&fw_3~aXS
zeD+IzZ}4OPo{QFfckI8fzxd7c#_rB*lVYYui>cK8XrJplJ^LG5<fqKi`)|`?{W)h}
zNSEq*yVdy3Q~!6_`&?dUyZYr!-Bz0!_hu#UahY53tR2}a)|Tl_`Wj&ETI9b4`<^$6
z`%H($73FyIgr>Jm)Ci6^yQHJSE_MBRkx4fFUB$sOB|7cBn|?bzTk~1wS3sN2*7qkh
z<z}{D2=u7mV12Rs)m69I7Mu3Jeo?bt=<dXA&(C~2%&WIaAnEO!M(;f@R5c^ATGj6H
ztlPTr!)JlSf<0-@?~B!Dc}%$-I6Ln8onptLpO4L-*-e)7p1l9k);Xp3?udFkSfDpG
z7(QwJ4>T=)%F`h9vl;u=pLLNR)%bR}?=3L<&At7J{rOb49rF8@$4z6J{^S<lo{H>$
zRj&fBo=msVd;yzr{{fzH=W_TBpK+ISSmBws=;qg}shaO!ym&AFLx%P0*G{q8f}q<O
zzq-EMb;<rYd9~X6|IdzUN+<CY88rVZQHa&`FbcJ)OFrjqlNQ(ZskXR&HTR=gKiuah
z+~4#e*8H+u&HO#}hh?fo{okFsZDaU$`Ld%I`IGjFPhR)i+U5PK>*cjO=2YF6t;W$m
z^-X9wd@TKBbi@~(qsID66w1#<els{SeWeJ;w5oK))sK>`zQs8|{Pq8W=CMVJ`yO_b
ze*dhy)mZLcRC#Zt{!VZC8Q-h9Q}w==7oV#r-@*K=VC$Nlbvx`8R93m4ORcEVEq!aJ
z%z9IFx2rgh-ghfc>zcRh%jbAbQn?~|@);kW+xOhXZ^f_g%rU84S{mopqkP-1zObOr
zd+XusTj6(8_vd8AEdKUJX3kQ1#wiw`XKu2rk?!}O_$|cxt?a&rf0y>`P+RO5_A?@Q
z`hAIWOYX{Es114PV<4%LTDfhjt9;Nl^-22{RhmmVn@p=YdoL?pX6iOYfA8oz`R@-a
z?*F;-)o;J%#YLZ6r>Q>LXtB4;r}SCxNyDo3-}HV|yg2k;``((*?QzyZehZU1?uIcv
zy7^@Oo7=KgJUIIKkDV8U&z$h_qsQStB3i7MQZITwP0X3kv4a2eTK7f2dsX`~Yvy}+
z?I~a|HtCSGvYY2(RAOIJy<yk0nzvVt-^@SZ!uQts_PnWSFQ4>pt6!JR*K~RL^}UDx
zDjfK3D48<V__Xu;lgAISUfRja{#bu`&YKJ0Zk4Ijx_-{vQnz^Hlk5tos~0>&x7JnI
z%DkUFfA@ZQ#~m!aS1ms9t2|ueen|5QD-Kr+JA-B)?rmN*jU)1>{KJoVqQ_>)vu%EQ
zG~?v7{%4Mdwm;Q*cKGYM7`xqN%WPlWdi=0jcIRTfQz`#GzdN?4K`3&e+-k>f_lpyY
zZl37<@aDbAwfkpp+E<o@AJVka--3Nyu_K1z;Fl9?S1k1rp5Dck*kS29^W|ERO}y7;
z^sVX+yc3~*^3_z<kH2|Bx_|Y#dOlrTu_Cek+q;wBo-8_RdcD<F^4S8H{cU%)8N8hS
zcd7tq@ySYSkgXmE4`wb|QYeqDS}^#?BH+{iYHo3-_}18vPzjebGvBh6K|80fu#&ji
zyOZbS<oTy>TdiWZJn$fG*^JxM{kF*kDxL8$msYtS!Lj7Ca{t$I;{(~!mx2qVg8WL@
zur?$Flnl(8UQRmnb$h0g&nZ>A`Q0i{xm^Ek^f+&)!SpCg(KFJ?O4fIJ#w$lh_l<MR
z8RxAiIeho>J)3;y1HITMh7=eY3$;OpI>{{N`C!wv^!eQD<_o(dxaOT-w$W!%sZ{UL
z8npxZbC*0;KmYniK`}O;gJxG5m;@xm7%L;HWMBEI7IthppuKWU*YiEoLMMhWvz`xb
z)p>TYBI%`a;XWJPyk9b+R)){i-tzRx{l}b@gHO;MXkbcKHr^Xo=e*>Et`pys%SYz_
zR|`35@*uePhG*!~168{xERE-!{*d#b=gI}q{gXEKN-*WIU=25r3lr{e_w*)IZ8G3G
z8Lb-fS8d@6t5X5-UaIk%UQRZ9)L5!FQ7>3+*Q%gM@vUt}m&>^$_eV-F`%Mh-p7|Qj
z<fTKx9?qU#BmJhPXL;rA70c74T{qQxI=;ES^2jN@D#@6|dm5Wwzb!fAdazWV^}7Fg
z4l9@SUJL&<{axj^tGzjE>GjLDeizcUr_QO2eBrxa@W0It=M#353w}*%f3fZb8=kpj
zw~3(T4AZAZw`x9{u;bLXbvHsnl)1X=bfa#aQF0UaKd!&T>e=ZNC0eO#p9fs4*cYI}
z&ANA~VnNsnzFol@r#!jt$uIi1qNjUP^bYGXuF1z8zkl!K?f;o+!|B9($0N7qO4?n4
zJL0kU;zvUWH0|$tZt0uPs)yFD6P<4JBR%lypNSf^XB1r$mq*0yh<tHpcWc%HhFu21
z0(aF)gr?R93Ttf06K8bH)^zmfnrU>h*J*0+miz^ug|>#jsIOD}d}P_A?!HP5h4%t%
zwcP<szFY3={Rvtuu#0cuck7k;+dtg8b(GycZ+W?dK=*!OeZ9&IP1YOxHAF0@HT{w)
zIBvwnx`)w~i=`#S@YLPlV|^ba&VT>w{;KfyUB=isk>V07e?P>rwNyYtvvH%KkGP16
zxKex5&7UF+_w_afXSvi&be^Frx3Sio_3HKQa~QQJ9o33oHi6euM9qK6jPELnFDJd&
zC82myjceLvmW3vbD_uHY%lL?QO+QzyIeTvNQ$5As(Jk?rVw3iVv8dGf7iz%H0%*UH
zwnluBh6+RVX1|z-6wyQ3f@-l`tdG)z6y`C#S~T(3;U|mJ<BqJjxhng+r1;-wANI!W
z&#HLceOWGkO6F5}XSW>7``_hPvhV%A^wU#)xvbd2!oIf;R+o$HF%9~Dv=*L;Q2TfW
zLQKi3GtP;baWy;-VENPU^rNpk;-#{p)je_TWw|2OdFvK6{Cb%a!BvwlI`zstd*|Ig
ze|LGR98KrWexP-#Wq;qMr=K%zc8a(>Z*BS&;kNide3`lL*B_yik1Fg=TOMKMR@Kvd
z_x{t+CmF`iEHB?(F}c00uA#W4@_Wi}op*~APTnc$i+xyxqYUY>YTPK|W8b{I>!iqm
zZQKjvwR&E~CVgE~wD4a#@6>{o3ww(A7R?v3t6yypn6`7{;@(%Ys?WvPJea1!_1>t{
zfo*E*MPH#!LjJ<@(k_;jl|P8_T@+*(Su$Vc)_j-7l5;bT8}%?<<y-&q*mAGyyZ){+
zypy}>5%1ac_g(E*@NCG@D)f1L*?UuxmrvV7euwtm|CdeL_T>0Ba~->o=citN&<f-G
zl73uTdSC2}fB(L2epC8T?s>WOtIzSak7g9Exj6Oe)8I3~Z<hV~DQPI$A3OW+Lx+Pu
zj+dK0v)QzE-G#ToKK*f%|K3V}rgEk2+yCfS-~X4He%!;%tats~7WqHxwT<~(vcEjL
zxb4ccbqs&aHLqBUt50jmZcfJDm=cgU&6I3B!*A(&&bfA3Ugcpsybk7dGEMN}m0j-X
zq&Z{V<?Vrj)6E`s<_k|3FXfaEJu~ek$N&E8dscM0G!}M#OI@{c-D2gcRg?1stzwzJ
z9z9sZwJi8UOu_#wxx3w`j(KQqIUAlV$=LXJYwWj4Y5YIi<E{2G{?q%iyP$KbGVeN>
zyn?$rzdzPV^Ih&N_+Fpm{`ZP}-nyOB_PK5<i=OVkQ6{^{|K}y{)XCD<Q})(?mTca#
zjI-agwd_}Y{>t5L>t6n|=CCT;banE{`CVrDk6xGi)-KLpRyA9D!T(REpK6;=-==@*
zS60za`@pUj9*eD4UV7+z`PP^1c9$OTNk4i1?YwLb=f8xBlM`0Wot-B8@y-8#$9f*)
z3Mj2cLn)v4pMR{H;^=OqamK^(@a2!5x;`fh-#0v3Q1w!8Rrk9^yIz>w)bLf2>gjl`
z(c$;?a_Fj)>%2nP=c*e_P1=2UY7~2k$&LUXVZ##M<RjNNy6yaYt<_Wh%ihAcMJxBd
z-+19KY(xCB?~p0_-Xmo@e!pK=wyap8Hfz$#Pro8L*)Ko(61!e(yYKC=YnJ`}_M5i8
z3*uU~{_?T!_fGa5X}`aE_pEcCte(rds;8;FpEWb5EMVKuN#9@0vb}ol<^QkeZ*^T*
zQ@HQh>h)9BuYLbT{iFHx^;dtr-o5X6lc#%<@22hV3-6Wx5;=VT`_?s&-u773?!RX3
zSIXA>a{9BQa*wy+s=W;Eu^e+ZsJ*cL5M$BFGOJ}Pf+8j+b>DI_p15k2%fYff%Wu*J
zSE7<@>sXR{Ha%AlPnGTX_71%9-qhPD+3UP_=X8Vb`?IoYZvF}J{`+_JrD~(6@8jmw
zWPP3AckIpjV&B}NU&8)dcIf`>OcU?_dvfntW6{tPg=#w=AHBV7-!#d5o@u&Oino7#
zYu#@<>&VTQ@8@Lb|Ija7#&Y_9-umb>jbB1$?|Zp?{;WMOyY;8sJyqKp{VYAg`o-tM
z9mV#yVr<*{KiA8A*O^;l5;L3k_^qITN2+sp`mik?VC3jgZ{!s6d0tYgYi_R6Cs#B1
zuPp!fS*lZaeGJz9y6*DKH^N8tW!&~3ulwcqcFiuo^<Ue=Q{GfR&Zz!X{5v}BWr@?z
zQ<t9noj)rt{{i=X+l%wQ96ESyWyqo<asi>0UQHXHeq;4K9{%LWwr{r!r>O5Z;#Jt3
z%@yzVbze2p_1#)?Ki1#AZBm;#ZR@+fqm#Af-MVfwv+3Q=kgTIS6b=V{?(M$RchhIr
zW#L=j_Gz;3oPXo8&DVmx)~XA%KCKb){%fx1WjiUy_(#=tkwyD|-``}ga#eoxz4)2Z
z)svcE?uothqVH<c%iDgXCF!-#kIP?UdU?lfT593ho*#$%YcEWw+;ej7+3()n#w^BX
zG>i8dp3S=PuiHPpKE?FZDe<38n<L`v?wHNjpFKBA@Adkc)1TRQdGEPk?b11Yv%xV{
z?3pUrnc?7r7xyO5&srU<Teh<=sMONKdud|Gx~+YxE(v$<9eX-w<(uG;%D;`*M86cf
zZb<W6bo9c}gI`u(myO@N`I4pf6CuankB&C4J-x5i^VTu;u%-N-D<@o54f`Ut_sOnY
z$4i=T>q4Au!29<5&35(eFFO+W;oZq^thaWS`<I=mUwgx_{IWQAoX-Y{Ker0wiZA^A
zJ$u!<!b>d+Z+%>wI``4b{@K$5kNw!(ZM1gpsRduk>)+nmm%V?{mn&jk8zpy$Onkr@
zVHrGk`*i<}VV|u^c3P-!o7VKQ$@secy3$+rqSyZ>A3bkpYMHaF^2OHX+~Q-EM_)$0
znm>2;jrZ}ad+Jvi&-fcZL;UvV=A#L<kFk&JGq4<+<>3DzZw>FCKRw~P3-^4fb_svH
z)QQ)|fAW-RbLJQJ^zOE+U+^`FZ%Ogo_dOQQwLOlXOv`@p$aq{oeQ3YZ?Xn;DxYcv_
zl_ow4(@MPLx%KSV{WmWL+<3`Z^N78xtY7H!jc>nJ?yuRu@4?nb+f4Y3_?Io1S`j%X
zl{flPYJ6|D6#FI18~e1bmnqm^6geYdfAP`DUq7o$4}Ma+W&5KJw6j~5{nE<1-Nk3O
z{J5LF?D+d<g_mz6t$bE#z#Nx-{`J9k^0%gKFAv|Du_;>X&id(hj3?{P-4lEB+3T4J
zrH)Z&)Hi*5zC2pyPT%x4zx_L_vorqvz3kq5a81d&@c&D>C;7L%-WuLHz15RrS-bNg
zL#%auW1(WhnJ+8i7C#ZTS{ZpnrI_Ke<eU>Cq1V_&_LMq4x)#5F?k4Z-B^q5jY7;M2
zi_EdLQeSJGaOSy&MJ;%u_x%Sg_(bn5(>=?Cbfzt;zZN52mGd&?^+fme=j!s)R^61W
znzQ}2)%huqJ>>b;t2FQJ`N(~F{@HEPckeYQ>HfYu@%;R(uW5faZ)x7VwtCZU`I_BX
zoyXVDvbeDK-?f$JV+?Le@g!aDXSUw=r>OGb9ux2NAKj|zb8U`G-M#jAYwXWG`xTU*
zEot@I{GqILw|d+4Jt|S92Zi+RR^R<!R4KVG-OoPCLwDVQ1@;Clan={i`u{fvpPP9n
z?dz2v&VR4Yey{m4$^2U3+pmqkj#y$1CPt2))W(gIW}e&V^lQ%hR<B(i|11OzR4y0r
z*l0hVscNw=(BPP{veo_Q%jz@zTiGsE+QdBnJ^gyxdp_PJpFf(0{5!g2a+7fOv~3kh
zKPB{oC!djhx$&19SK0M*9%j?igW7fns<Sqqk`J{?zxN^J<k^QtH~;V2tvTQLuI1}L
z(|&r2-rf84-*o23=kFVRW#wD)S#q&o*m|uF>qED5^XmOI#qZsZzB=`Ew$|z*7w#pW
z?btWR@7nv|t@LxXYajM)b1AA{tGj;x=8Z+HQ;!;6Fxk;4FS4wD#oX*yR>BJX#;>ZT
z&b@f+g2=9e2dw`-Fi5fQ`}^weu1FiXUG2ZtpPtO0_5IcEcbXq3$y=)|t^D<^zZO>j
zi8U&6@8f;V{4zwpmHV(Chvk~@e^Z`UhA90_O#C3Xywf8*V2<I;rayPrrd(B5ITdsK
zl5)Xco!<-R%{7odE`8UK{q?RlQ~BnDb`pzt@UgYpDmp&QJo#*$^Zhx$U0yu3x%ljQ
zobLW7dpDh{<n^n+<eVR>U-9ME8{37sW^&~$>TjP2=|@c~e^GgFnWx+CUu<7DKk7O7
zCQIx2rQo>M+h%1yA6M?RxgB=$^}%T#@790)_?!Rr=WSC<&prL?@7sM}+KN-#SMHXu
zLw@PS^%jY0R!`5A?3=dfc>A0Ef?7+3bv`=TC(L;j%{}k^(bI0LR-W7XvGTXyl!`<)
z9qdK(<A{VDmdnhSc8AQLv+3Z{mCwJ%u9(!C^gOtAQOczVjp`_?;M>un@<AS()ffNR
z(^ov@&YVdX4yV1!*M9Hu=bH4?=~GRr)h-8AKmG1*^ks`h^Y@wOpXbl}z1n&5?A=V9
z?5}nkz1TG`p=!esuf5vszok|j;hz6~_mNp82jyKd9@sv9^zTlcX+*z`?X?S6zt7tD
z`{JGY@TJE8j=#UMZlT>0%_+72f4^Ry!~a(Q<}&WIv!6GrEkA$6Uaw`|@&)_PNdBnN
zIlJn8sos`YliRQPx20Su{iXiVbV}j<{IFle>`_hj5g&F0K2nuR+=63oVK->Syk0lg
zLXWE~cbu*+6xNt<SNMfYV2S6`-?Lu+OINE3(U=_-7Oj~Rqgdkgwxchd=iZt%M{{>V
zW-%Kr^j<rap6?HSDc0_({(Jtt`$ZRDUQ%kdlf3@noTq`gNx7x4hV`tv8y1^0RKs@k
zc-O1Oo!cT;(_Lry@@L?J)AABu_RsWqe_rxi`8xYbTlQC#L6`T}>-`TsHSPV>*m-HD
zUsx`$wr-j{aa&v8jve=|etz?^kS+Yh&dqn@w(r)QF*VQhL~MNQte9Kz_P6~(jkngF
z8Xqss7wp5n@V-aB(UG}%@uR@|npN&={?tvraI(Ec<^KexSB1}Y7u@4^;pK=l;c(Rz
zdUs%T(sTdUM^D|{7U#R+viF0T{n5o+9e%wo7qt<La(XiL(Kc5;{{s&DBfhY#ua3Qw
z-k!CTz5dFE=#se$m*0c#>Q&wHH{o_nR4emmnJ;?zACK`~vzzogSX27tvHjEIqWvWw
zR7L)f|NG@h_+j5Avu0<!yX1N3{LSa@=e{|yYo~wiw{rh2%Rg7F;J@F*<6RJR<=`>%
z3qQNUwpW%<HI`7>ucH60pkB4eZ0Z)(&+)7U@#T7cR(1jNaV)N7X#A*P5T>Redg!68
z;-7ufTkqGe4p@3W*kSIe#1*~D|AazjWrPG-HvHpXc%q+c{&(&}>#Jjb%#L!a(MkT$
z<}MX`@RK6a^x$swJv}d0@?G7Tle%_Zf`X;~!8>bWkEn+6&DHx@^j7Eh1mCWiB8&Oi
zR+-C|pSkdE>Drwk-iE#H-<9|Ne?HH)@aILha=zF_b?XftpOt>5()M?Co5f~51LdPK
za~!@lbEVH)#B%!7=e13zv)yl}ePp{@r|-W(qUzP^^7+00?z`X5pRssq@6P<lSB?Lw
zD!x6Q_Ohl=z0I|A`oAd!GFZoc5^O-57=AUn&w2E>UE4MML2Z|Po}kmR{w=+6$Ktrw
z&0p}KO82Sp_KSx8DV1jh_qg%R;XdhG8!Rcla*~~DN&c#{y1Bgf;sfsq>wRV3xOauy
zo1emJ^1TfvSJODsCWdeQ9i(yDYNGzboENA6#}rLIeckqDddA*QNBc$ppWeT(D(mIM
zs_mh#|D9c&yK=j2^`okM2h*ak@8Z)`M6;hed9S%+X0IvHa_!vQ(^Y4`{BW+Zy;^yG
z-}|YbxHYxoj?b6RSv9Nt?R~FzHJ=_YTUlc_>34m$^HZl;1toIZ_p4klE65D<x!m`u
zujPWjrti5^&nqM6%DdmURy-ALJ2Sp&<)1|l`1u)KxptZCDE@iJ^zx(rPf^i-L`p!*
z_qOzx-FX}O<<`c1jg|N6KSx9t!~{?FnR<Fz#LNB5Kb-#7^-f5C`jcZy+O{@=eNNbq
z12b;i82Icd&n~TRHfJOcy>6brGTyw?=eLdG>#(2&yBnD9{uJ?;@2Q%gksN7jZ24m5
z>&3SxwS=$mvSmHr&*8bSQ{rmTtl}@uAJ2)Mx^cJYT;zd^ylFLC_Q9scEPww*o*Q!&
zO}rnz_3-RJXX^Zq8UKzBD_SX<rCux$dwECg!zH=eMM>4KI<i&6#8P>DJH_AH-PiT8
z{_<$yHPyB29R6%(kJ7LEm+-pCZh7zRtSQa*%Up{7Z|}EE%l^C1-SO-9Q(n!CId3J_
z-(33R-_H7{Z}VfOl$9JdJiaDq)`qWFla6n?-fy9|Ad*>RzY<T)zeSqcobT_j*cu-2
zJ9~YI&hn4z9(-QtebGL9-D3X_7q*8j-TU+Ip)bm}elGvEH6?p#bl#k~7xKfZHXP~R
zy!*)MyWNxfGOs6NU*X2k_)*8;TGPwD_g!Bm%Oz^v($9G`V}k15fEu<$L#dWmTe)R@
zLsq?5v23$R=c`%%LX20PZktrD@M(;=>LTwM;1IN8)AVh&A-k4#H+|4M<&-bR*m|n%
zM5)NS60W%`=Q_QLQ>?BLiTK{VO@g24<=%Mb`?h;kSk*5s@pP^BU3T)V#fQJ3U8(=l
zqwhZTTc7lF)<YY?m!+HE%3s^|`_Z<k<>z?!zUvnExwYwE!f7GCPeo@xUH%?)Ir(ev
zecdxRif$iT;`R35oWj~22d5tXot^$Y{?X9|i@*DIJBrPhnD2StHtp`q^ao{o3OJ{3
z-*55u{dW8KZO<~AtGmQpwo9a3|GN3jZcU9^i%Y+6pV_5%(e8u1(TPhlHwOAv{C{08
zw&#!j`FTG=3inobsXN?{|Ml@;-nFgqGjeq+A8p<lowD=0{LB|m4*s`mUiTt&c3pC(
za{qkOb1LO^xs_f9((}Im);s?;pYvLXd*}3hMmMk@87Xm|$>jQDix{r%9Hj@nqIUaZ
zBYevidW877m<0yh^+*g^Jb_6hVELUmzGd1|0@nm>h~X&zuy}cp;ktEUiPua!B_@0S
zS-e79YGK==n`U#9bh@6+`gm>H>G>9S7;Dq^OsuQm*#CiL4z{hPJmq&cx^4ejvG}%|
z{ZtL_&1=>^eAWEF#`|qg`}S+Ghb7-E<v;VgwtjN@rw)zBcT={%sGSmcZ#ReK+|}9N
zC6jxK8eUIrm$scAS-JnrOp)wKe@}1D)m3Yj;qh`W{$KFdt*Gbp_8i-bmM7&Fe{Q=r
zr~k>VgSRSHmiWZB-=8A0_&Z;U`CD=JeHVOYz67ro1vQD~*4*zbiR0`@pIpC*UrqSs
z(#MflCPZc!FeO_X@7;fW%hHb5MW5B*Eb03j+xFO5Il=fP(;vwd=6nwA)6ySqtI>&3
zn0wyW-^}{w;oE<8r$ox0nbEi8h<E+>hx;x+;`#H~Yd<&pr_7>Yt^b#+JD>gAyS}H!
z-(T_tXi8OSztX2^f124+XMgVZ`m2{eeT#ar!d&S)8#P{5?B^|H!dmV#Fg^BJP@daS
z_^qH~PN(vLn~&OG#~wUsasAwu{)!3TYwp)Cx%cHWdvrz~lj5ewy`|1`CGCUGH~hvi
zeE%Sa<yd;!ubUnZW*O=(n4I$URMU^z5Pn++$;(fb{~A31{ZIMpp3`d4*|+XC)?Poi
z<$H8+?UG|nxQt6sVLg_9_RB<(H^ur_eBLuR#d7#4FkbPQ@1QT;&xJW7bD%}oVdiws
z@25T<z}}Driwdk|XnJ|EYoRUH^>dJcZ66CpmdomQw2w1mZzh5L@34i5W0|{l+c&VW
zpt&H-3c%nYi-6Dkg7q@|Jecl<*yX^$bdgJ8hTK|-TX-x3`OTpA(B2{&13d0_(O%I1
zjcJQ~ghHguyR0wg{vAHBwm^(YK%$>%(xdhA{pVZLo8Y$MaW;0(3|-NG_SN?p)58P*
z)^A_({hqd`g00HJ@9RVq6+&HSorq`M-L%N1BPgV+Rh~&HLcmp1Nn!7G0T0#27L_G}
zm#Xi+`M3Z6?(ci%?)`nLHvfM0?|bjQ*Vv^mKR@^9z3s-u=jT`|v$Qk_2rxN0aBwgx
zDp15xaaf?@xMxwT!(#V@m)n1D)5^ZLFxt6%OYX~gYu~P)zkS=n+Y6s_<FjRgMB}F;
zo?e1Xz0!@-a;<wczMj&{eG&KW%Pe1K>32CV%3l1$>rfT91tx`)f{5{vl0xFHI(EFK
zfdeE+K_Rh<t6_=`;h;HG@|TUU)k_>44t)_}(2OJ;;4LqI@q!(J*Hn+jhL)F_3_+U+
zM<eIuuhN9$dIA$8=jC9AB`Ji{gV3_q<_ZKuMuC+{XxVZGkCTLhTgB&kKH;S3z{R5C
z<Igx@QY*o<>oH^9eS(TO8bvrfX3S$!P{o@(!9JZ}5W62=8U`y8VA2$rU|`GY;3Y;l
zYbEZgV<sFj#3#fBF#_OJv#Hl1G&D5&ilxr&udDX$Q49zT3i=dxZ{9L?P8=Rsz#%Zf
zU@u49q*)p)T>_2W7oSCUZ@8OPWXe?hutz`wyCWUS1RIzhF|PQy|K{4ztxp9S?VTQe
z@LC;zsKnW6oqzz-O_v7?a>ADF`qXzoy0U(eoCdb^vLI8TKw;O-t(hfT1HTmQyUopc
zIo9NfIFpk?G)KYotE<<n`7%v6dfHm6P=7wb7dJL8zP_)vy1J?5Wm`)O)?|9qH6f{N
zb#Shw+k@Qg|AfpH_6umJv4s7vIHQ@NpIe;nqPjG0yU_ZQ6t~vR5i3@mijEBpwN!ts
z<-K>+y417(f{iL~ZR>e`BFRxE{L#BtSz$ZN-bTG&xqRNKq_?ZI!Z~CLj|j%ix2s+A
zYoo;>lgms+C@GOkoN>+m-HS_Wxwp4wMY^zrCG+kxNs3DqW^!84A$VYi@8wm$-`v`I
zdbV9{mEiX!AI)mNyto**I&AF?$5Xeg5^C0MxaVbjcfy3|5{@bPy0uF;<z77+=jj<3
zUlJL_w6M<e<wVJS4ZicjOY;|UTxzyE;<G1UhMpE5i&go%9NxQJ{?BGxhjpWbOJj{<
z0kd35;y$@$&!6x#{%@c4{zT;4RS_H<jZF>=O(j~bzlwBD`H1O61PHthQqB@+cyMR7
zL1L4uPdV$$9-XZsJDnEX5O`7Y`U~&e<!5$CE_3-8{WLUGEc*B63-@+Pm*vj7^mLbG
ztJ;q*zI!&*XJ1=$(#bs3b)8_w<z>BZcYF$bpRnl0<k#t|>X0L=agCzG?cIwWxj#7j
z=zUX_0!xbn3(GO(e*ONbOi3(Tq!hT%KA1Lj>dnc^XC9BqTqZ44)Gfq#Lvz)qI=jys
zd&B!@Z{NM;)uTzPCi%JEoM<9-wzv7xs!7MKd}sbk5&yRTgNB;V^&<fl`s#7ryUNy<
zzCRiL@soxh2S;O@Bf~6h-49RL%ssCe?A5=u{MC%Qi~FohV>m?QnP#Y4Ex7u7!&4LS
z4Mk5+J$mgToNH(*|NHCf^u5ntUWgQ%pv}_au-Hu@O~3Z;uF~FJi;v#2eE!64_Xocn
z@<Pkn1p{|z2)-~#{q-j_UfOc$mf)YPs!LTDcW)9Fl;U;#waYfN`rSLL)Nk&Bo1WWp
zi+$eHesI3Tjv4p&R2JX6y?}jLzr%FB8`D-Qa$e5fFu|Wggo#Ogks!aQlk=6f6C5?N
zmihg2q9wtJ?W=P_y~uN^JJB1HTEG8#z5e=ov#IB#nPja>3R+)U$<8u=@>N0c!UGPU
z4PqWM@;pSQNl%*SVZ3)kTz}MLUzT0>?b5kl{k}VEwY&W5sZ)Fpb!`gF6?-`8(#zea
zuHU&{{fR5CE?|Yr+8KR0bA0xQywYBEd))<>z}M4W?b<1-%+lg8Pw>LUzc$SCT93b+
zQ77gQBI@M8;o_!X=KmrjBjpU2(qg}Mu_M_5$}JAg3J<gu-rd-E`1ZE76OLN*GWsUk
zT)yb@oO_mdUg%;Wvo`DH1(#hGyq*$ye%)-FPlr!+rA_}-bUw*_;gVCg_b%V^>i*qz
zvwu1XesNWc>w2KN)Az~8p!7u%eIJa>V?pUjN`^5@rpn^y<FoVa_aBsA%yz^*;lcvP
zxXo#2XY7!*F|N})$^|O$ikcjJX6U6(oAY4yW-sSMCAmkQ@nnYTuyC<6u6cQTp{est
zrOc(`=c)^%gQB8deS8w&@c!8DV825p&F*=Nofdq4V(Gm~Prm5xF21$<Ure3ytEuDB
zn}n;UR()EvdzbILNF63W&dc3z+OMq)R{x&FJF`M_)A!}i%<@<^+<11><)7gNUTL$6
z_j`hm$HYHoTi{W3ob~$zMdvkJFK>9TZbJh&w>roPe7JtU`rDh2*Vf0+Z$JK0@TQYO
ziB-s<FFA~>t6Cfql@Dy)n?K9@(3e9sfh+aoQ&+3{=$|QH9+t}3IO)FLjn1rpJKm{3
zeim94`g}@->e<Z!x%XBbIJ+pnM053y{VAX3#8y^k{FPTptomzsCYyJm>z<aEm9Fvp
z!c3Zi5{)uf;y#v?*sF%|PHS9Vtr5WQv1}6$=T0XFjwdpVIlGF~&+q@!*7W7nsZ(?8
zgZWSJF+F|ybltPZLODn3Ce57{+ASc!#M;-OC$IZK{K=d<>E4gCj6G+>J<tz4yV{;<
z?)usPZml}0Y_D}Qwk|U>({ui>>oZ<Uttxx4=;H6)aps#>u*{A)vz^6gaR;Bi+})j%
zt0fmXxhs_H*|MUpxV3NXEp?Vn4Gm@*n;DYUt-i1<Wu0KyUnd6+6-NiPxJeDcCp_m^
zD>!Ud%i_)RI3dc^E8lqI>EfVinrEiVy3UXbUaWaWZ$+q+f;C6Ue4F&t>uJVyjZaT`
z-&|>EzV(30yGeI%uda*BY5Kyz`)<Q8wq#SyC(pi5?2eLE6bPO9>Q2zy?(|iDiV7@B
z$_j~9DJ!J(7900%61?P-zaxC1I(Or;4hNqZ>%Nv3?)^8xsoQ6d$o1F3*{u$>0v)k-
zyJ~-b`*mgUCQGLSk+<^?D%<sZm*Bkoc;TWQZk-M%Q`R#@z4|`ADztn-b8C5OQR>yt
zn@=sjf8%qwak%J9?hek{lAnQvdmQ7YU0*jh-?52z9`9QTKQSdm1yHK}f3Cl)i!1-}
zvEJ6pT3eThG4(m5m>#=$bgPgkxRv)rtZ{}x?CbY;nC8B{C7Z~*dYXalZHCo$iUAEA
z4@~Uyx8CXYz08)uxKF`E^W4+Lg>RRiJma0X&wi^`J4eZ`ykphU>ivHz&d*Dmen4-h
z{e-6<3N=@+TeIp`RM^hv_s;E~VQZZF?9e#@;bqxL8gIl?*Dhx8IPT5MyVGgG27v=f
zyG^4^Carz+)P+MljQOgRe506dR7LLXZKftAKKCbpi_vp}96maI51Y&0+<2IKcbDn&
z^H-QRwKr@Q+%3Iz-|W4!LMMaLR1YXkJ(zxYS?Q}QC--t|K0H~=d71l<>#E9~jtdSy
zv3$Kst$Ke{wB@%elf2%ugfE&gTU~8UX<cRI$-8sa&flA0;QmqN$DuE_)!!=iFnEUi
z;Ak;PQen~h6TVE$OqXTHjc?a8FPZ7w5no@prQm*Vlp4z{2BtacHNxGq)Vp5i77K4n
z^^P-tdYMOK`x{6;l`>`8*D3yO!YQLgcN7fLf3m$_+kdTUNB0Hm3J>%AihMba4!6KJ
zOE;@=US9lgWlXOC<BZyag}sve>vvXN*(0?6k5b9T+I_n^>$5bLgxlUMJ$2@I!tuQ?
zuJT@-E_Yy^{l~=$dp8-q^PEw0aP@N+7YC;W9S)ka1u`U9N^UJrEbxE5a$#Y}w@a6u
z{Aac{2&_<NGWr<g&mg_3BIDhJ&gpGcf>rATPS`U|Y<cN@J@;!^{;ij$+S*^gy}NoZ
zJ1*|J)Av79XYcW~F00Osno?c;J?DPIuZ2I^4X+Bv&)R;fI`?0e{q1*MUYwVM8JCqP
zPT*vUU4CZX?qBTH@zJ;TTy>4>G*_BzbxAAuX>N<b-~PMH&iqkWThuz&a@&mQozoB1
z1?Ih;(`u8ubpQKhC-qrc9GaCM-1bh_Hd$x(W4Ff*hIE;y$9`OCk(>%{CZDilimFeS
zzr5$liy0LUKdbnh=YCbyA)vv*QnBd7<D-jD>dF88xIe9;T5ENwp1w`ahVnNj;<ifd
z&$(-s|4DgsQsHEO>9@SC4zVHPjK|mH-aTSr*q8PzP33(5<fHwocJH02vHVf1gR<L#
z+rIKS?~GD^<pxO9OQdCff3j-zg$MKYF0c8%wby5Y-;O1YGgr&|_Pyej)t}qt>2{p6
zsAcOaQ6@>H2mPyKGV*VSRg|Zt-&pPzenxOveMe!4**U4Z(qZi_4vxwX&I|w8cVSt!
z<Njs4dW)a&sd>$a6IVLQ(dgFBaO{=eiz^9bI#WBvg_juT?{WTbP<&vC1Z&avh=_S-
z1UO>tta!73-*T|Ndg@F!bK4%L1*fOX{QvvQsf+T{rcRFy3VKv}QcOwY!t^Os{N;P|
zjv3!vwfVrSPR`4zn>QY~=hO78MeO6I>GK%WUAY_E91fVyRbBg~t%CXDyqoJ{t^36%
zu&D^^LmJ3MeGJb!?^Sr%+hsgH)*I*Q>UwVeO7;|XMy~9NgRu_T3Tu9z05@VYu5kQ_
zdiiGIwePJz=Q&UB-#2;9)Js18ZGD$!sW92*&DdJ-`<m(XsUe}MPs2(%o`z2koqFfS
z(_1flgC_JHwK=n|df(gibFIs(qPJa0@^~0{be>oK+i7#}t?Qa2YooEpPdzj6Ma9aN
zm!2DfCJAIPuxxR>$*^GgZ;xEBW!*DAaWg7TaAiqSi(7JHl55E408w9#DWLY>7B+L0
zCtQuk`W;lNU-rtk9eSIXk!!CT>AiI3JRaGlrjaba?rzH2{ap9=;a98O<O74k%G6jM
zoi$=fes$~;Z@p~E+efmJ)AZYtwiv&ce$ln-@UK6=wfBlYtBZR5>}pWQliOQ!p7Fl9
zb<8?xzFge*ZyTlG86R69^nC8T!<*;L`r5g@=yKV4Kj)8<$NS#Al<j=(Ag%KFSNG0@
zj71aNQVsX}Y~5?9Bp|%ZT{c`=nWOcddhg=ae{EOn6fdk0oM5omBR(ITF_mijv$hGG
zU})UKbvZo$*2A;EQzn1Exa#b^Z+ENrFvi}haIV_&xve#4b@0m%y{y@L-tPWZaky&h
z=PC!j&o{n2y?2WD_TGzstt#Dl-wVFh`dru0?Jr)l>*!<Em2+=z`M_41{kZql<%`+Z
zS4E|4%lw`9ey(fw-Y(nFxPCs}y&L}~s~7&M+V<emm6aa%V*{UleS7DY^u^`QQD2W6
zvCG}+pT93<?)Q(g)@FD|zkjgs?x9GfcG>EeI+H_;V-I~fu|X?2*k!@(ubbndK0n^b
zt>wMP+k5Y%d4YxpSQqb3xcy50ZoS>(t1D8wEaH7$2!|R!n^->G$-5@}j-G$~rhC<2
z?xx<`$<F)5PwKXM-aL(o7iN6Vt4ZoJufBQA>ZW+|g1x_&_V3|aJFEIJd)c?k;W-Bs
z{(P+J|91S_t9yEDrRTr4S#Utv>A3gwB<DoeM)SzC!c4xrjQ>9c_9d^((K0hl;99xA
z&!I*5R*zHG>|07KEe?y_6}J7mwk*4F@9PUM12`{tzy7*%tJ4BIjvr}p72=orWOrWO
zT2uQk>pj2zneSRM$*<N|t^T`Bo_*;deOaR$+KbJUGoBuky(fFZdltjZ6W8X7?h}~z
zkX7KwY=gN@AEM{adww={W2fiV6~VKg-7ELIaa@|m^3-nm^`HO!DBAi}%kJTSy&Lz#
zXKi8fKmYH@@9);$e-&Eq2cMI8{d@7X{I3_DZqBRCihHx^+Vzc^AD)~Mo$xtj8TV_B
z%bG4;Sxw%DO1}G^&}Vt1aN+B7<7wwDi=UjYeiK~g*BM^rF-K?Nib+pHtIO_P{VQH<
zv1E#X%8&$+uN-!DBGE%&ZIdE<4#<_C(0ccr&TZ@u`K^U~c@W=YqtyvTe~pLWj6
z{>Faq?DxNnzi(aQ_WY9AjK1k+yRNU|GZkWqxg0s$Eca8{`*o7b55>;;|1vFKTWp)*
zA<x-iXHV_1n;TV`TX;!5`U~Ths2wx@iFCged%E-=r&6BFwyb+4(<d(UJ^6iureuA)
zMYO)-SMcBi6N`a=!H2J=GM4Ta(sr$QC=ju^a{2bm-+JpyyNlMRw})v;KjOIj^Q5t9
zDr2eulhcA1f(Le-{r!V&lk(26DV0x~Lk%BV%okTXW#jenY{9>U|6*5cowQsxE?hh2
zQ7HGFl@slLPK>;#ej%*i{q^>DpVrluY%Bk~^YLCg+3aJEd_^Mr-rMhZX5|zvaer;>
z*SpRyuiW36TRL;9^!DogjRp7ju9=z}-M($YWz|#ex3+)h)!UG9>9qUIg)i$^H-1-A
z6cB8j$bar({p0xym)f1z{<^QhQFR@!P>-j6c~s=D^OLNfTV3ky;McsDlgWMg?mE7g
z_pVu|WR&L{vpB6!Wd9m0<8Jip`<A8u4CngzY~1tdq1NhS+Q-+uo^BY)b8NE(qfIww
zi~GD&r&p=BM;tql{4C+n$6nPp8Nr*pH(k9E5x3^o#R!2L(y}j4oKv}MR#W{@>ixFm
z_8qs=`*Li8D&_ose0h1r3Y3p*mQ|{ogx)*cxzJ#1Ufi?V_nVdX?5#d3Hus@Uuhkmg
zlWu>ju39Pvyx{0@WBz;4_NGnXi@6y*Z=?5Cwtru{|LTmZe=lk(@11B+KA|IOTh7a^
znU{}Edj3e--CD7&Ek^Oe4#5*Qj+~ynIjkZ-@$UOct3_TM`#j~Z^Mr*GA7>|&?C;yR
z^S0$_^9LezrKh+*+kN$$e?x!ir{2gTOOI{W_IQ73+fMT>?$z-(+oZde+!fdn{QA$C
zzw8yd3-4@ZUY6=n9G=s|d0EqBxzU7+Ck`C(w)#51_<_UzAE$4a+~5s1ayxs<Yoq6b
zN^9*b?YvJ*@<qI!e+s;&o>(OkGS!`fg~P%7@7uHCZ+B;;S6&VOP;Ygh;o$%0-&Res
zt$$hecFnA-Q&i28#F>S~KPlgI%?fyTYwAC~gnW*wKKZ++-8<j)%gvFy?YTtkNn7sj
zLY)<NZZ|!4lVP{slK1TWl3%P%_A(0Q?^k>%SFK#qepod(T~gL+&Fu0o9rNl660UE%
z=jgnva`Uvj4_mb*PdeUJ6Kne^b8KS8pOd1`F7y39p1u9o^qEF;x82PQU&E@_<8q|4
z<?YRFs}F70cDtQ8=Z43c+kFwC$1C=&ocZ+i_Dbz*`QJ@rGQ4%>v{y#HH+$gpJ7tcI
z#YEi$PmUyqon(m-S|)v|sMWzyN#OPZpERq}u@~B^4;oHT+q@&<??=tWdl+}9D+%v<
zeU!D3_2rsu={H*EB6iuCUCX(%?ey=1_V+(;OVD{XqxSO6$9um?%56NeWbIGA#f2vK
zJ6=~hIdE8T$YlIW`P}l<aC(!7-0%C1!i!{<*SGv!=)K@!dSzMYn~&=%E0*Q=ZIqni
zcdqrqqYFI#ol$dpZ!Yysm>zSmPdfkKg*%GE`fNQhiWhzee(>G1VfM?srgcr56fL=Y
zqqnRJyZ1Tq+|d->J=@QnS^uW#z|XHgvy(C(?mBOoz+lE^{oso4*1i|Q#?y{(3Yb>-
zc8TFWg?XQP=i8YWt7SajTOm@$|Nid1YJRm6PmyKPwOg{f^-H#YSh4Tw?EaV6&VO28
zyW`fX#-61+zjvSS|2$RI^1LrA-^)qKk6K-WPkmMK@fS~gn*WJo%kF=BuZE^qnXi+!
zdspI-+rM_zny4##ik`=UrzAL=4$Fqs%XpRtb+n7>D=dA<kku!>_U@+3zn==f|IL(h
zOm<&+`TN!GyjxT5|2-Zp|9g6hfdQNK`ss>yxAMC=R?pbs?7b-9Z`<+bSG?lb9KTI}
zUv;?On$77)H{bo!w>b}fJ$&P7t(-8++m4-A*p6GBKgRdQaD)9bq1L<ndv@M<@Bemb
znXg_}`Mt`AjE0T2>XYhyV}r8Yr^*{I?|uJ#uGzx)H4oB!a^in~zWq;)K|f?}%a;2#
zpB>-KxP0E;UTijNZt~CA{^C7TZXeZD=i0vJ52UYtKhA2tP-)E5XEFIR${x;me9!&h
z##jH$#VW%c6DH{uKJ>P3aeHfM*DrnH<l`I0**DqhdrD1hD>4OHU#j%&%<&c$TDCve
z;-n~O==up`qsQ!Z56c2Pk3Qo#$1gIerS;{N7tx!xxhbYvidehcbUqN-yg7*T@@Dq6
zSwc*$eGbQFy?j~HGrd=;@bZ@kce_FrJ)znUZ&mLtJK&;ncFB>&hZl5IoZ45L?0F|8
z_LYX#JcCu?K8yF7`q;_!f2dur_vLdi<D}20GdNEA&N#Pei<3iSzjnlXE@6F>-l?8{
zLrh|rEhk>=+>w1=&tdPI7m4>a)t~R+{GscwWv=4mU-eVt^&|sZ(M4Y3Ooaj(l^2=?
zihebRPWP)=^5k5m@5v?J61_K)>{eY+k8!lPBi58Ty<*D>QI>OVdw%{*;@EzAVQ20W
zu9{Qc-nEQ;rdunobhzA(-_!BP_(Dzj)-yA=G2WVT^2DY8AHN-!d%NMdvY*8FJ)b6C
zUt7z~@!9(5rh@P<Z(C1(k63Z{B)_;^>o4I~1?EC)l(Kf(a4=fmE;^W4{qy2(ncS+Z
zQ}_SfX*YDVGHx+VU9;}l+1b@^Z*ATDK<7c==c6ebK|{$I9xQ(fOV7`<z1PMot+ciI
zF}vwO&pjvqiT9uV=KtBiruwH_zO2~nwD<{yff)~9>2A!4*$`L#OQNUns@S;|t^d}S
zPpjBtvSQ_=#j!j19=~W>*46w=7t|Y55V-7f{@SnIj#U>~<fVj`oxYMJ{D*hNX<sqM
zc;njF*Vf)UH`mnZfksBjk&`uA?{~WVbg;?&dbRPS=rh@CoA-u$?N(>J==W=a+*R=w
zBdv+CAJ1Dhl{_nMy~fk-e}MnB<YZj~$*=Few7#rU_>o+8`FUN?;-!9#)|!)lX}&J(
zob|!&kDgRdzKHcv>1qEX_HB5*=I!;xK40HCSWFgP_t$L>OJ!?=fJd)`u=o<;1v3n6
z=O!;ZIpMF$7UxrUy|PwcK4#f@tN+XGi3(huF>58W<`f@TvTa*-|LgVZ^*`PFlhJ!z
z+52#Hd55Xu)MrbU-WF6B%ju52cW*~TbK#!9w_e!4*X`!KY#+08<urxbko$c5Hr(D@
zaQM}u*1B$m9Yw4!*UT5bnv>CRYsn9tzO=&YQ<tU2EQ*}9A=luRcendFm4hj|j~+KJ
zu~59=z~OLtw)v}`hmR{QC$9b}yVcX=WznS_c5?o2BtGhOJvXTPkfCa>;eW^?DS5@y
zpXO$7EYi%5SCokP-O?#H39xVrtUj{B`Sr6KS7eeL-(M`BUisPZ+0HM|uZ#Vf-6C_^
zZ~l$@{=YcBUUqwI6XUW)j;V6nQt?atcUEhgSyx{-Om{U7>)*5c%460QF&Uqo7wP*n
z-)wg*)>)!|<k>3S6`Av1bLPysH09u;tI>Zq&gy=t{QYI%>iu^Z_WoYV`me^QFd+T@
zv&VDw&wUSj;Kw^J@7FiZ!*|zR`Pgozx^!wByP_)xqv8Z@76bi&>|}}8_xDs5-*TE6
zv(@|SGPj2N>V40Y%hzPf_xi@R7I82t2DES_+<DiQ_PhI`kDk13>HWIWGaGdpjqZjT
z^h>L8ZR%aYDYtCe*{qfeG5p6DZ}@qu*I@T+A^+98Tpn)qwpr;_$?dy-#({ZKFCP~k
zD*bm#>g><JgMWo)UOC^(INfSX);qS#-`e%%dxIKVUIqpP$tfzZ2q{k3#+JIMOX9p^
z*{a1S&o<6FH2E)EY2(VKqbCA=ueg{y3Y_KLu|;3+uJXk<Cw&e09}B(h^#6H2=RLz6
zHDBI*u2Y&8GmDNz+;8c=8|Q1CFkOE4_V_C-*SB=vV^0(Dn}6Z{ZT_b#S9&&n&E4Vn
zYwiIx@!UDmA0yrFl<xhMa(~O7@98p)Rb1l56Euz(F63qX=vH%X>+=4utbsN{rAOoM
zAN;u7>;AOg-2yIw7exCtQXU2BWvvQ!5ba+gD9kT<l(VtPvEkLze_UJbZnfDOw(WE=
z4z;<OEZu)6d$Zw`pBa;$2KG2sH=S1IYbe;gbur^%VUzRw|7Xeuhkff2f4=TZnB^>k
zp6!e0$|_bYH~6a=GY^z#A}7w+Ex=S~<ePWX;-pK~{{Is0CfD=VTG}45Hd(_T#&|Zq
zaL&D^ji17Lj=yVDwHL~`)*!ud-&qeC=dHWmUJ>|wVav3c%+>O<{;cenJ*jV6x0xz<
z=tqHNQzygqd;3Hd+I@C0lRC1qnMqb_TFK$I?EiYXuQL2US?<a2kY1EyVxG40ouP2D
z-SMA$7h2vl-jdcErSI$b?%1y#yi;NxarQ>=`<Dj2IP|=&RsYJXm_2FT`}e)OzI5yL
z&9$p<r0L$4?f>DUb>@R~)AaXCon@^`KFl)h=3KZeNp9u7x%zXxri62FG$y$y=&QvY
zI&dU--QyqGS9net1!~xoz2nq;+uF1*iSPB(?QHs-s)6S=b1YAfWj!!4z)U4IXn&-_
z2kis%yG~_wvxxa!f9ke;^EUs?N3CDIXUx+(bxNP5rJ<jpEJAbYhu+M!@%!btpPpee
z$cp=a|FitB<5M&m%JiAGo;K3ZkZfSJYCW=~JaJoVlc3OP=|GO$G>&6VE9%U)T!?+^
zbNag&%Qx=@rIlZwyH$#Oxe&rw?$^8Gt%O<pf#{Vd^$H)%=WyK5dAV}6nW=JS(=pq!
zqmM(}ekV-7*77n>t~~g`jfs0}9IZ}&c-$~+>av;pg=S6u*l~P+-Itvc<5tB+W-dGK
z>|f}#AVJ`RxyH>|zc@TK{*-m9#jW_8*!1H2#sy0!95nIValq8k>h#;@U_+4uyIq3A
zKHJ}YlK-sL_n1m_kNNSr{3>xBvwAGpifqgk|NqU`>EpcQ`~ApYh7G&k6g}JjxQ;7)
zg5TuC+ZWltb$8dy^PANW?3eiaz&cCmYnfjo|9-r4!9wfomkI?<(A-&!qQGqRO{H2U
zciVz4KJI(|yxQ1cUHkFWc}M<8y}9~{W%9+p67w_jCYC%*Y_9#>EwXI#jxuxO)R$L&
z|N3+F-uuKy6OLFu5>M~@-1@VQ<zK;@|IhB<>%9B4@ZiGcy^ouCubPXk`Sz|hHSvth
ztf>#@oTzDkShvie@bTB{{^!2h75F87yr*23?Xq-&Q}>&t?yP=4TO2+rKbT#SXl``T
z$A9_nxwk*4WNr<TH~)WVOP+V{B(LiiSg$<XBev{q>GTNq?PaE2*6Yk{K1c3dT+q0Z
z^Kzzn?W;qk8?)FRs;tdzN(-uwSY5|ErRt8^(c2y!wl;moElrPH$t>J-rL$|BXyEc0
zO5l|kacAO^YsEg>{Q9hwm-J7-y)MSlYU^SDwv|lES<TWH-<&k!IbqUhP#xJT_Hf6)
zxw_r&FWtM&?=ST*aJ7xkmW5xxn~SwC4sP52_*m8~fr__>L+?G_p;J;-;JsCJOSk{s
zevW76QGKU<dwo_MKH2aryJttF18-UKtWAbow?#JpFkbL;rq_gXd-?=A%p{*hR)QvI
zL|B-9M{66&hHE-hFsjAIELvt|{;y}oaTdv*V)M_xO7q@w&sjP5-m#tQn|{U}jQBfu
zzyIN<uJXs<ElKz@^EdB@3dxDydlo<X{QK+obCv}Q<j-<<ohn@PoaM;prYF1gtq%OY
zsHvQMOZ$vVuKnWgb58rO+*tQMobQvrOr({+?nayAfpt0?a^2#sDkgkqk9kvbx>7hv
zdg+_b#s^m_oPNGbc0uZiU>6hJX<c&)r$nmy__NNq)Y>3$LX9ahvOfKmLR>_-sUL4)
zvx$yJ`rMf5|6jykw&gPGsn~Vt#rpO7>rDUO*kyUQ>MW1)>s0GQ(%ZNTIp@cAowd<z
z{@LSYw!iw@VXLN+XAjjr%kFruv8Kb>Rq~$eBd$v`=SyF&Zwj!}YJJ8xH>-<l&L_^s
zNA3#wvtF)iQaGy;RF>!>yfvi9&?vH{Rd{Ej>Z}hH*18r;&wq+}efviamr>aJ$HuG7
zLnX^D8?C%QFR`?^_<Bc?QgB&{Osd;T4Hcj1X;oSWrs%P>IP4YZ$UJt%z1b*^n@{rA
z*~JD&-aIhk_uczJW1pb!rjU@bp1W>mf7B+nYy0dG__A#C_pfm?ey`MLxb=!-ZG3as
z>7N%IIOAUku4_(|3;g@}HTT}>&(?nOU3cX4<6}E#y}hIH!_ZFWmy^fQbg97G^R`qK
z_uegS)4WmC8r@@DWbdKVr)T;mSWsUs_o7nT_Zi2%v-jl%iH7YG5McW0a6mr1{LPJr
zADb4faN&IuUeGgXdGd;{8yV7qdo&dC;^LlV&kp{4_{_|Jh`se2JYUZHASpiM)T130
zCnq}Zxo`K`ac!?(T*ViToF$vj@cT(c=5}#^4dBRMp7-O8a02t)Gbx6DTzg(V@jIuv
zVB&>o&p)4e(v^AP!;AgBSML^@+!woZ{9w=3!<VORvWkB_;i;#Nf1>3|@s@}4B`@Fp
z{#iLu<{9g(6^bA8x+;#C#j;O*w8js17Qp4GyxWhSZnkh&*d*{cqjy$s>I7}jvXK*v
zOadGMoVVwzU)s6EDlj-B$zV;vq>Q<lrGJDEWrPRt?d;soX1aUF8TZLC|L5*IW4<!u
z=^0-6v*wcSyS{!(vc2~)@$J%;2H!8)#O-@4WU4ZK-R)O9(?n<bO`c?}{q2g8twXBh
zIdioK3j@*&&j>GBRQGmQwXs-8h5X|!;Wo2&rHGx`6Y_fFo<Hv%aUb{F^>a_fPZ4#N
z7Kg`936H1A9xu(feJTF-^6H-%yF7NNKiV1jlRZZuuBEzf#Y*37hJ!^n*H2SRUo7DN
z_Jhd8xdl%>mG8YZ6MC#Qm7#L|Y{9AJXJWXfa{sJ#So_7`=B;@jXI{B4mTZ3Z)-Bb=
zIl{iZAM<waj*AKKeyRTFO+}*njD2En<7M>yE!J(@S7#BL2wGR+ki@b{$>+I(1=BMB
z6{0t!n@jee%dJ^h{9~ht%Wh`1x65`<x7glvj8pvD$Bwk{iY=3!re@2}6y43yyhxny
zz`G^F>unxqI6ato(^S}XQH_{?^c$HozS?cw1?7r<3$LX!H~O=-1P5lyKVSB0MXJtx
zWrn0FCP_RJJDnCV2sBJ+S@n3|-&H<l8Y_Lp)n@5f{=eW@XQKah$;`>(8x=hk8O-I*
z=319qc&1|338P!}1>rXg^}oM6w#Xz$crMfL>cn49UmyG{6s9)uHP@<@s?LerrYU!l
zXWS7{*eSg-yRfQrdAY`&^0zvlD)wH?y0O<ZtFe3Py_f*y`}ZE^EKsw$dH8Jf_S&#;
zo9(`{&Eif^+j!Ye%Kb5@wXVQY)Ox@m<PHBN6VVAPJ6*Hq<fcSfO!dyS-|ka?LvsJC
zCxQFKlydw3Gkq7j>8A1i$HSK@TsDsc{f!F`Pu*5-ymtD1{$+|MC;RSJ5PAO0@Z?>U
zWuLs%XW0GF-ts7RmxOqPyzQdp`z3#?pGgndqgYt{t$wrJ$8**h3*_tMzBlb~xvaOx
zVzzko|MFj-B7)kV<?i*r9^RO}<x_I!hnfvF8>MZo?r*)_J29REG-KrOoNuiy*Co%H
zYu;aGZQ<(JqkPEBc6pL<!K+2P+BGWA$H?E=E;iN9dRc$@cW>@LZ|;cvO?{DPwfgyz
zjaRvJ#g<*L(eZh1Wv{dS<dV93&F-x7LgsvDzD&2ZUKJe|_ik&x{A6WUqmPGnsrvMH
zZTQ5=s3_pf;3Pk*W7YwVtbn|mLh-lt!xTExHQsc_w;at8dGsVfN>%>gzFpau%X!#d
zg>1{qbnos?3qL&fmd47C?$7FlmGdq--1xER@0WWwIA$=P6`viUGxuG0#Dv)jcYBx5
zH#@ZAW2U^syF#61^?zq1?%SsJcPqy&dy}aXG(Bg`(=5_bXZh6R5N4*acBxFbo@Elp
zv6Zu0jgAKhFAH{_`}l?W4vTXgOs_sXns)2H`8?Zu)n(_@?>S%KD)$ZA%eU}rq=42A
zGp@DkD)io}{o|jdT=w5}k4?|xr^`k1K7R`2GR^yUKj!@c#n``-9m9_OWmxc1UpX)8
z-?y9mIuWPuGA*pJDe@OO>RtZW$t+*hr~Du|3SyKeT&+~Pd+c09)a@<Y@AWq={>Hp}
z+gCrC+sCU7EL=|o``lS^cXxPfqSaK-ilCDnvmUIA^1M=2x#?ufOOYiX>J;aBEZsTn
z=ZyP3D}0<c?fJsrUc|MZ)8pSU<6lnJZ?<KgKFg>)&%jjb+U8gv=6`pX*?-;Y_jmk|
zXHw&^vi#JMbz(g`ZL7b1>9hNl!EMiQ_IR<b0;v6atefG9n)<@>|Ec`Nx%zX@FZguE
zGGg}qK5JD2onxwX%5$W1#EzX6Q2%rx@ovGT;-%vIvQL~nwXwF~`>frMW*igVRF>ZM
z^zQEP-BD?BlJh<P9#7zTUe2pOi%-~pRUOmj6qU<U_@8eqSnfCXj?F}$JsWgblR#?>
zS{!z9^tdqAFFA9ZZ?)+;A0dM?I}hG{Dr3Ij*$g3H)`APm1Z;fP-8EQt^o)Q*K<<-U
z8tHc~ZQXJ{Z{@u5mlK_96U|R=%NLE`;cIN-cHg3GM*i*Zzt#Ib<o<{&KNit1FY6|2
zxG?cv_eG-{q00?c@?CeHaPE{UOP8Zgzkr4dOVYPXmu15X#W!s?_j}fJv{OPaWo7R7
z)5XWGUouK>cTaM>&E;;|TXE?4)q9Vd%<nBz>Aw{0@yVOxYUP>Bs(t-|9q#`sub50#
zxckZ~tAhXYFD9W|OH0K+yqha^Tc&!#MyEoXz&8tzYldlSZPAPjzq&juj}_d!i*$Y9
zuv+(WwdtA~wRPFkCKPk(uP*CcZWL>DIW|w`{UeL)|Nrtrbc@?xUdc{7xyEQ#m15wE
zyk+|(((K~$ycBm|+;^^=`LV@$-u7j?CvKgqzBuNx$z$iIebP5icd3+aW%K9i7ix7#
zb~_O6kivdhaM@*+se=By+dGq0-%c*R>u1U2|594kWXbIMtQFq3Z>1&7>{qW^*P7`Z
zuGzZWDw2O@sabjdfA*<Cv;U|s@B7Agob~L(eAaDKr<?Ddz2ur$UQC-Pd*vo|pP5(u
zgJbU77w%Z!ng7*PTl=F%%KvMezkf^*-n{lryb;&hx60g?U+<fm;vYB*v~+lZR3qEP
zo`v5QROg(ml}!G9UFX5`Ltzi&<Kv&$^5%Y;T+sQF<HqMjoeh%SW)=R{;ui`p8TOtk
z)6G)TkqdnKF=TUQ-gM=gpMMw{GTeG9^1wz)Mt76R<j&y5mpaVqRtkDO{2?{v=>dC=
zLnX%A0bHIEXK&A#&^i4{;Y9znwaQs1K4iSTc)0r7ECU_esnJnUuS(xY9Db4f=0e!~
zRNv5mm!OeN?#7<uSEpHA`*-=W+wvTB@l_^|KRt85nig;`cF#R=g9yj+;7gBUrYBz$
z=aYJJ_leBgB@IfVg~^xqUg%xM*mqp8{+3C7**UNCcWY8&MVj`AZZV88XO&BToH8Lh
z`{G{Xrl{teWE-x0mLsPpYn_WM*!=kY(o^e8z%{Fw;E7cRwOmuI52+VhEQ>G8|5JPD
z>g=*(yABnauX%f&Pfa>@k>+}XtxZWCGm|!M@p=80ZK>Eip;*Sry(+u&Pux{`_N$)5
zL^am8_!%#Exy1F2rM|z{zJ0yz+LxJ?KV&CyS{Ac+@1N-@s|8wG7pP=#F=}p~!Qb-d
z@1N%$PKz^MQ=J)X<0JF`-M$|V`Q~~RX>#j6#xDBv*gXEudHbiHlFY5|n|EzpY5Z>!
z&r`+n_>u+3PfvS%HT7!|@4u^y+kSajhh#nZaYNpSXPc^E;hz0(yM_Pm3$UEn^!9^H
zLaOn~`CMJwYZI0HoN_<=-o7wx_e0m348KXW{%(DHOVVaf(^)@TEpb<O8)$gh#Z^J=
z%gOC&Jgt`=E_Q0(y}kA9*I9GJFDotYjn6VxQ@WLHvdOzm=S;V@$Cjg_X3oXOKYw2M
zaObJt9A?e23oSp-X8LU)6L0j;qJO7e{QjPkhKnxzjflxRvUrV6-%;<$XG{*7aeS`-
zcEV>p{~6;gd4X)bbLVb+{-$+b+NKH5KQ7(0>4uBaWtNTRiv0o_5-i`6z0Go)?p}{I
zon3e{PuBm(41>AL6Ji#n?Y`Kf`nqtUfA;-zYmQ#z&0ee`)n6gGYVMy%rnjZvSFF5w
zBenK@?Vfply1yi5x2GMSn`<RByTiut$S;d~yAKtMEeW{Sk(ROOPIKbV%Y~X2k0PWl
z__UNqIDi&3J1I=(I1s{+TlRDJvfJD9zi-Q3ZCm&J)6KAQ*VIqduXF$A{#>ae)cPt|
zf9pBl8GaA>`hV{}RFZYNc&=<%Pv>>L@_C!KEiYalxW`KOoZzzJrt5BN7?s00OzK!_
z60OcYTDE!L^D65-j*@L%OkXw#*v@J1GFWy%+^_hsNXq2J0`C_;cFU<SHVcgGTb_2$
zEKtte?(VP1X|ukq+h+UNXN6zAQBljw)f>0PPtsXGTZ~tCdkZMqI2cdN;A9f}mMI&4
zimgCoS-WP<naeB|4{P!&4AdGtg?}AgR581B4`&N^0pCCNH=&<n;wF9P)BDW9eJOeB
z!Ot@T_ufsAwQqfQ-a+{@=ZwXQ8|%LB(6RT>c(G`%+jq;<3kK&~O7@rtUkP#Jc*4%u
zv()*6O8o@S>qoqI=`KmL>UU`C<bK3=!Ng=2`}B>q{~N#C%v%%L%5Asu=0>LEOMmlq
zw(0Ei?iCT*`ah<Md0YB+hr@+O+j6(|T{G^N-TM5g$9pTajcfiaKi9hG&Xy;|k@l>5
zdrW2>e-~S1dQQ`4qxTaX{?GdxxSxcv-Rc&bzwY#9+dWG(6mq?!gg}GoC)k-xL?7@w
z{0vLkc&_xI>O41*sNLco#`j{^8{OGj>s!|*99a9xwE9KZ_v{(fSu-1t%yBO=kClvj
zK1U&S#*<l<Z^SYd%=Z29N+#}2kUe+!(JL7`%huiAmK**3(W~zYYgVp(7q`75sii?6
zgM}lZ*8i++e6_yp+AnSwzAk#nbp7kwTSwP0x?J;0XHSl2alIUE8p6TR$mYs0nMqY$
z`9{at+2+f)9?s&;XMeaQmc_l^%0$Hdho;W*g5w?A19u;JY&%_kx4GWwXraq_TS~aj
ztYobddh@^N`sI$dJ9v&yd)MN&i?eD*+Ojx3n-YDQ+EuGk`xdaivXcLHutJiZ)lMW}
z=Cx<B2Xn8lS*%~?)qi98?)>7yi?4L+WS9MQ3TE5#*Sq(){9L0`)i36%-g|5uc4gD~
z6IMTTZiyQ=TK_)y{9}{)q-^!G{TzP_O_F9PDF$%Jv$i@gD+^e^ot&L$wbdj;Np+2&
z+`TDJDvtL$$KCK*_T#SKJ&CX_ve{9_Tb{q@^YlL$BRcb&)OyR@9?_3?rRC>8?oWO1
zq<y#V(r(EZU-5IIWeexdetx31>OsnR1sQSO!%vso-BZ7BuSyxCpnmJK)|FRg-QSSA
z{dnQ@Wyu-mH>&$ErFbp$J^7<dz-VpEQ*ecSQm|uY#mTQW6IV#a$_T5^k+*BeExLa0
z$Cjhn*`lV6C2LPuH2%7B`nQwC`kS}@HkykcpLd`#TSretwe-{3O96r>_^U1({{Ikt
zpkwy_;*0W&*BMkhyb>^qJ?!n{f3Upe;^WKHXPYmSo@Kp-&A(%cP>aK2X9Z<p{a(IS
zrgs8cFKM=OYHZqD@@%@n-Gs@Y)!`v58C?gqi%(jacE<Plo7dT5K_Bi|o%J@Y^yw($
z^PO?db)9y_Mn4&gC9B=zco$l4l_@_sFXDWk?q0dK(MMVCO}l=FSz*n#$BUxNLMm*n
zw)D)@TyWrp^?b>9*SE{u>V3OF)ly*AoHvITI`6JD(Y+vKcBAr27U!EuA3rHC*6zE%
z``X6E(er1$_H9kGKDqZyXX#wd=mY6*uHSzz-@kU+!u=+h4rfnq`0!uvhS}}L=W3g6
z({c;MK83ujx&G)W7o%c;07pvA&kVJ-&pGFZ94g^*Tv3@JBDUo7riI>pH%?Fb+xmjB
zd3V)%C+!QWeK)Ug$gtk--8?=2%jB@UI}axc3+i({jqq9Ms`O~D*d3QAsqS}Tp0t+T
z;W3Tclct^XBya12_qWp0*Pbw8x;4dzStnJL$w}ch$B~woGuQ2ZeOuv{T=~m>{rV?=
z9~=?<cTCk^a_4uUuZJw;?26t`EB*U*<@D^w7i+DLT)Vn#&7ohJVrR^1Dlbi09<=h0
zQ<eO8ZCk_So%QwXtRJ`Ae@|TN?fw05XQ<HYbj#fgvtL)wyJ5v~U*iAamzTt!ul@8{
z_4ng72WR?zyEM6Rsfpr+9ReMkm!-3Ww-g?n`S+fMT3p1=mX|l%BTTs%6(1-B<nwxQ
zg<rbrvHkgH4mH!n9*u&+GGSe-XP=yUQl=*zF=CU;vOFeyE_z?Z*TRK&*M;^@+s*lF
zEl>a6TkNd*-wOJted}7R>RF@Vqol5U-*49EJ1&*cTQVMg?8tQ3@_F}N@X9+44%nRE
zudkh~C-S^wH)~zt@^4pjno)OuHTT}RH&Rr#I#+VZ#{6RQ`~F8(sxDZ3?ON4@1@De6
zJiM#&xtzs@X^|4sY+pAp?t2?0Y-P|lU3;?L(UbH4SO0U)f7dnZ<80~0;&S_@mw*2M
z?ANWPGRF&NmcCAID&acd-zt`1>G!B0=c((NJ>gAXjx}A~XE67l@`9Cqz9n2&GVPA3
zUUPhVq?SD-Tcc=By3E$AFPCjH&n&e(RPv6iCgCHfloepAbT}|yT`Te3s)a2^Cc?*r
zXP2o9oXxBK{Vy}l^4P@<zMq2CzkJ;>uh;48+VY(L=L{2f{oeDTQ+uAz-lr=oH|o?^
zCvW_IAX@(Vtj9AvQl;iq7+JA+8Nb_;roK(bcIxXhvS+jsAtP>5{7fe2p9Wv~y7%ey
z+43w-kBi@(*~4S~^O>FQ?CSDo%cHBh1O%8`I}TVEPwSm;vFTSuva`;!`%MZ*uTGF*
zH)d{iOH{OAb**-s-9Ej(W=GNF`h_dD-t3znWg(uIG3mm`OEC>=tXdt~I1ZGE@lNee
zF4WtisnO|lyzdR`!m{!i^H&{lJ^GUK^2&>Mxt2GTg@t6#E0*Enw$+%i@6V3Cg|C(!
zie|HZGP^LUJvkyjhapHpwtq$Azs;AA|4^@}C^Rznt?bn7%nYy3+mgS%=IODw|5nbw
zXVB=nOL}?EojM!i>ZS7M?bZ8UzMX$<BS)T35QEbqPVs4*gRPZ+mwd_leaI#M@weD}
zpGt3E+;96&@bA94(azO-mz~Ywp1)K0R!7;Q6YC95%HImfxxa7k(-~d9p5<DbDnMhz
z9F1#~8!DT)^Y5#jJk~31o43bIWN!HdhkE8KX301EHYm&$kClJKv3T*8t#)7gUdI<#
zZu`AvgF)?~1Ltl;ynS&tVwd8y-r1}4wjTUbaq^+_A8&8ne;fAhu0DTgiLyEWp(8)1
zER(R5Ke{;RPU(}Fn7v%Z#}`3|#}>OB5MP#_CAB?JDJWrXk;!HIqd)fO%zd`>>({HR
zBy&}#&*SjC#wKyT$3A(<=aQ$QOil}42uR%6Jz0ORpL?^>?72(S773biRc_gQVTHNR
zrk*z6N0FyjiNF5y;6dRf1Kzh6*G;>7w4iF+A)#gI+@4pJS&EttL|l1w|I)N~Y)k9C
zYncv~ZJPF}q(A$RYE9$p=xI*h7jE#Bo~kkBePgiclM0vFPZM`PeOR~VW>7NIG255S
zjg!{@%=>ytq?hxOZeLz|*x9{&wOM+>rI*k6Up9EM;z*%Lch2c~MZra9mc-w0HT|3z
z25lDxv#k4f_Hyp>n8H2rd+gNWZn1qADg5iR>9Rwe!jHp~tz;&bCoi7Vlr}G>sIxpT
zru<mM`@ZMTuV1>BCO7N9Rc1{79oCSwCCN|Ud=Z)ddfuu{CA(%`$$o16FexOu{Ig_^
z`B~*Q|5-&m|K-kHvibGS=>7f;rFmu7BKZ|1Z!O+ie)FPpvYOo5x@|w*)%(BXwS`%~
ze_-%vu1CCe+Klf<&OE)hT0X3Pev7|VV1RRDm4oJUzSA0@!G%X|4B26M4c}7qOMYe~
z?mL}!cGk}2-!B_|JS2B@o~-h>jh9wi<z92XU#Xc?2HwIpQ9&T%m`-ZBg=WBpiT;m1
zXY}9ckWv%*A|e$y!^QgLo+&RZGd2Y*zO0&6Q&(1!|Nh2ATj5*z<)5!#e}5~hrvBio
znaO?sQ@7r2{S~(6lam9-73s#c(dBa2*qRuEb{+cSBmeJMlj+7qQ~3Ki^)k9$%{H_z
zub#H%?z8Pp%O~#py1jf)-VUBVX@))ChcaxQHPi%V1REWCx>&WAF}Ht}(9GtO*R3xV
zw>TubCj9wuZ=>wlUmTZL8va}ET_D79=-D%YS;uepEK&G-j+_63|F;6$*ZE~_4No~0
zxvpsT#C4Va42u!DEB`Fs>hTqS8!PVbjB|TV+gaXMvATJr=;gK7cCWoSc5i>EBHF+H
z(`%K97v|N@GFdpk{ML_s|2!|IzTW;*ea4M{8x-!WonJe<$u6Jg*}``Q$|r;UmaIS2
zBl-UQkH^8%uX0p&XBVVQ?G%?c{42=h%fxuoTmQp#nO!v|y2AQzq8@rK<dP{c_2)1t
zWT~m{J)#irBrl$yHRaB?qn-Dbz1>~=nQ>3emVi$2)t`(4KSkZBOrO2e<NdvTuDZp{
zKQs1oHk#$$eK_~~;r)A~FNHbR+Z9|lZgCJ{iJP(Og7wc659_Yg^sJsLu*j#i;Z}x(
z$wwhhSsnj<cZC&8D&IdYz0kAf-iD<oB{q3)xn0edv*mfW6VpVFLnYhRyxKaGzxMf)
znyrE71(}={6e<<`+P^x^^?cRF@~?B%Gd^kk+$noL{L2F?Gts~ca$lbWSv@;(Y323Y
z<SAy+b#=i33Vwd=R$GppRNvq&a^dnC$vIXcv#eEo);CX3YH?WXs-Om%P)XdEeEbNv
z^bGm4Uk<f6trcQ&Qs4)*4!&OOS?O>6_wjFwL-9T?XGJf^oh;vddu#S{)BL#J>#w#(
zi&w`zwf}sV&DNiTqwxrb%%1lR-V<(BUSAh0m#g(`){33mWpA~p+^Vvza_Nt;H+uUk
zlPUYt5<^gnw_;CEiFd)S_H8-U2Yk*XZV;{i@<Fa}tJdYZh$BqfG&|pTE3~XMd?mMW
z&nED`StCb=<wh$ljpn6V`sdu*qB%J$WLcv?!85+N4x_D&vlbrbVpN<^&r(zU(k62K
zY4^|kbyL)>9vg4=Hn!!;I0fq89|L>oYR}0XtXea~nAW+SiTk_zPw6q|`+q&7L^U=G
z2ryYXGjQkJoAs^MP9a`;%9fec{QW<jUt7!0erU<8=y((~Cb^F<dW*^|p%q7tdss!a
zbFMn8v8(j;wYb$`hSDE4Rn8Xd(o$h*acJY{akQ*wxOB_fnMHbq#PeRW+@xuu_m8{f
zyemH=y<&aU8UX>OQt(ct>h0@huW9yUbg`KAa(+Th+{C=eLG$l>ZFhSr&g8TpP>Ero
zzogYOe=h5?Hy^g;t=$-VPs93-hn489!l?o)ChiF@U3^htgMHN_d7ksjC$zX<nmzrv
zWq;g)4aYX$oPB;-+p&usYH^=#I@x-IhPrzWG+x_Yle_EI=O@?r8^P-czcvRiZFNxN
zC~;a?_CR%}$hj>%Pt^VfZnHjp`iS>3jc40r{NHWQy?sq?(zZ@<u_*?3AAnoiS}bmg
z&FkY99<9+@!KHrps*n3>6(9XM;ftqnFe*7D?(+XBvh23;Q#WS?6@iNuQoLqQLCMM*
z6bG}^H!)}I%i>&nuvpV)xl4w<L2KKV$KM1i<aXFyVQpRA-+uMVq1gu`^WS>=WL@IE
zyWL#s#l4tw-H~Pu;thq>YgbHEEa3fBY&yfpf8LpSOy|UZa&^?qXnFaw$!70!i>Q}6
zGe9n#WH48wqoqNhqp87XkBj5VhEG$bO{<c%zSPy?-(66Xxs~+^q}`nnquX(P!XCjs
zhZd)<9<e(o=Pw8Wk3KoEBz<m~5-F4vGJ*5lorB+=2F>^<@XB6~g{k%7Yp24OEb;Eo
zHW<u(YRO;6kXFBrX+4uT^V}XcyJ-s++T_*B&p345l=G9ut{(yxvc6`O&o_Q}9z0{+
zX?N~F--U#h?Vs#f>}{;M#K_5k!^D#%sk<`sn8@+p&s7{7s~j#ZGXd?qIOKTXy3hSh
zsomR)(>H2WiSo_ZnZaqMXq$e}Ryr~3Stqy~X6eqbj7dX4_7wZaM}ct;T3UQ8PQ~o`
zy>n+Tp4RWaH__2b{%CjhYv+IO_54^(ep@ot2v)HDkSM5Kq0RMKy7ksWkq5!2?Zmg_
zPx7k#^eg+p^aFP%KRF$CsXS27T;)Xk*|4Rnmh&;a%WSTZ)}4RgS_04Xw5qw+gCqpc
z-I~XH$|F!sEmgBeZzX6Ws)zy8K7+ZFn_6D7{+Tvs*-4W|@a~!e0tX_j1bjc)-&&Fz
zk)VG()32RRcF(7x;G9V-b=ajUL6g$9OhG9pZI4-2GWoYz6(0}bxUOCNX~6@b&oWo%
zuDUhv#0Qfb;XEEwEgh@2FZdHzz|VBN?~2-q`I#kCx|4QzIutAHRD0yQw%+LM=Z>Q0
z1M-1ymp@2-VE^=)AZPa1)|Yo!)~+Z>{l)oNTk$-HMbV_E67$2;K2>mDHg@z)ZE+9=
zO~>4NAbl>cIOsy}m7^b-Ib`*xrd1X0G26Q7rGkQ@r{JFK^o?h)?Tcmc4HIr{5Xi9M
zIPzv`_T{ZR;<7T{otWA=z0u+685@OS)p-}ieRzM!H-<O%YWVLu$K?9x(-MbjgWoQ?
zg}407cS+0>6Pplyh2u`&wc^9e<$wIE`Kl5y?O`W(=F;cfUrgp-UgXNX-dFm+mcz6&
zCysD)9}CHq5C9i|LJ9|Bi|ar8M!n<|&^nNju6TZN?t8HW;j1c&7p4eyWEN${=-!#%
zXxz7H;;kt<cfyxW<7_<E(O?$a{bg;NNK}!k&wsNAn%oLia_=r$Hgz9*{*=+4`@5sT
zUB-k<A0He(sBbKAAz{bz#+7e<bBnXsi=BTlZpeHo&r`qS?0n}tQ6GdpNGB}T_!@aH
zL~nn$8GDWKor@2+4=f3L@bAG-9u^Y;z2C=6rA5yyYu+hWAuH3Lq5nI@K;PyrUs@bj
z@k-hMkp_l_N&C<7G5H@z-971|go+&7GwZ~;Ip>!?D12}<;o3*78y(N;&arN1H~y#m
zf&bv)1OAOG_vRkivu5^VKEsFB)#7K$_p-_KN4)P$n6{9ilJ7@cLB!?hlTJTASo2(@
zU~}B8s|%%<P3!m}G-Fxit_JaDc_v}^=|T^4B!vDiK5%4iZ7hdcFXQI<(FcxfF#RE7
zaC^JRAC((N1rADlEO&U8KYzv<w(theHSY^Qu&{aTs@Wpd?(};$udx1`;L3%iK5qq?
zoE9huK45t%EZ*)eQ`o}#R7`_u%hem(a-*~Fmb8>df!m_2jScqe!otF;USD6Y-@&$0
zuvD;sMd^=6)&4xbqu>o1oBA2HNNX}I`f$!%ctNGoZ;9ud+|u7YNIv9rAjf8xvDQK{
z)6VQaw+|#9G`U>l$Q|&K|G@tOt@$kfPbd7#tLOixwL?T^nYO@3$q#HFY!kR_Vv{0;
zp5LDMpyXBZPM?nV&zY7dl>hLZQK$E(o@sl*rStrMa$avgpyreM#q`1arn~#%I}ASV
z;QUj*!P{NYL2WMUGdEYGT~eLtY?3?lcCz{$Tz4__Lyi2O)e1=(cYZVdUtD0_^wxRa
z>j2?ppTB>}|1j}^fv)?J%m?L8oNIm?7W|O?^yN@t!tcA^8~<|}wBB%wZ&}WD=G(oD
zi3RM9^O-j1O|NS>u=2rwmWd||<m8|1<Xe|`v@TBQ$Zm<ct{wLq?mpk>TN7X2*n8^1
zPU)IPSCs;5#kxgHJt6z%g_gC~U!N$gomiE!VoD%~NhV9up)cRw-mXpw3{*_$p1CM%
zi<CBtls1#Vq=LUs^}m_t+<5Tz=H}%a%eJog9U2z)>|FW#Cl{XO20mVL479~fnB|+}
zT9KNcw|Zxo7kv^4omI;+%RK+x)@=9I#Q9=L?;>&=b$0#dFAcms{fqbquKUyX+w(sT
zTzx>qX0is0YA^HV+|G{sQ|>bdFSuzFsHUzv_t&J5=_Z-(8l}Mk%P!nuxwufL?!)UJ
zr&;fqHy?3+5OgA1s3o?Y>3PcE)O#T>&pf{7`*6E-npLOMZ}Wreo9>oyEsovf*m}N}
z_l%*sT%g3cULmeI)7d3goc+tS!(E_E_TP`4I}g1-FP5+^S;4|5=Qi(y{)7B1o#h8g
zS#LXTc|YTVo#?r+V{faK{!gBEvvB73@`Ojb^*WXp{B1fg|IEW*_r#YKpSh-%VYEU@
z&-c)mw!?A$Vobh*jWZ1PYL@L2{1SQSi`xgGtF3OW2Qt1a6RlRU-qHFyF7wgSEhfui
zmKAO(h!0&fjn~mB|3lb^>IYs1nt?T?SylC$Bg3ow!`>b1JJ5K|aVC?oO4s41cMb|K
zVhmg{GdHV~OP^nBsl$}06Cw%ECYL=<yz;`vu;Rsn1F5%rX8e<xk-LZIO+Mejr4wcy
z`Z6<tD_L`f>7$CXe|(ge2FI6ne)%u=O61{Yd#2(~%iDGeKJQ<Cux8`&22B&ySf+oH
zGn}p)ANcmjzG1%9tt%n_xMnaV9lOZFdA&YtA5$II=btglmOFpb$-5SCxwk1>=<$A@
zx~YvSDaDU=p4i+!IiK^-`esWuUY5x>p5-^JPhu!Bydk<I_=KH!jlhiAHrpzGG5uja
zsABg%>I2gU(*)sJ_3Z!H56B)k-rT#c{kiXlhY4Jp%Nspc7OXor^<MnHd8ZFt%gZwe
zSpAnjLFjhI7ygPkhImFfL6>#yzqw}YZ~oZx_~d10{?huW9#K&yrn|dJH>c^WJ)gGx
zuW)oKWZY(}z=3sI>gV@g=?YO;&v7PtOV)$g#?$6ho(fcc_u|q*Q}IO2m;2n8)s}y`
zXFFfUe$V;eC(dxjgBq6)6*lZQb#)SQUi9J4guP$3bFXjJj6PCYr*FV2+I`Uapl_?j
zff?^hrpbT#-Yk67I#RiMb7c6FcCF+f7L6kd4x|hGaQyK0!LJMafj=Y+SXHkqi{&~e
zE}OuAY1wMUK$*OZcMA**Zhuh?$!I#`{_sD;o9x4Bza2GSML%4^&-3Tcy9$;X@gJQE
zedl}(jwcA2Tb@3!ys^DG@>qw{-BhuL*?zTGg#6nzO6IYs&2#I1c#|#dPh7&{{6o&U
zsY0!f%YW^A_4BLEkD1%b1>B#Xf6(6a!M0#`vps9@3*W>ykqMV-A9&Zqr`2n;>|?mY
zY9pHy<djscX(1QedE)tj9O09qBE|QX2+Va#-1b%1;67*P-+iaPhjLG1UjM(3&n-cD
zhyIWC30&7akDPsR{lgy}jt|>E*F3n-pJ!iiUco1w{YRDo>xci0!9UJ@clBy5E?xJc
z{(tux@doy$pZ`y76+ZdvF4sTDIv-QL__q14SGGP_-uz=~fq(0r6oWhGdLF+#c5iR>
zWo@k(mK_hL8{XtseasaSz`@aYte2te^|r<B_8E8f>b$wvYx^}^J;BZHbNF|g^1pMA
zuZs~{x;0t-?w{QCtDeXk%l-TozvxLgXjS#6E(aeEqpT|$`SWXY_RhO*_99VJAmeJ+
zeSgD8?5P_sJj{6U{lM|&XqMh*?3%5T_u1{GkGK|A^vpfvdHGtB<Tclak8}*|XT>v$
zOPO+;uBa676Olh9r#Ztj*52*x^W!V$%0H^Hf1-Zy^ns5ZOp|{fWNtnZTCks&uU_Dj
zbpQKQh6>>yE+3o>UJ4nUHsEjd_iy;Qdj->O7sksX2A}`(&dJgBIC!@A!3r10BaRQ&
z26DTcIbJtM_=zn4cm9NLhQf;uJY!p~VEAvT@VTwinRMs>`ttH|YfeG*q&rMb3*LNu
zzNhZ*uS#ZizLMyvZ+A8)8y+uTT=MVz8lP=<xRx)Pbilvym5TbkJ(tfgZ|CuGe5dXg
zqm;kx_2XizCe1fXKTNBeUMI+s^xC8)y!rq6lV99_Sek9(U$XFOUp~9thpIW&@l5;K
zp56GRX~9+Ms@C^eZ=+4?ori{1a*y;_ifv*U)1>RU)Hcp5xGzyJf8+1jElw3Y!c4z(
zW4JDP&ati&H#nRVuKvO7LtMh*`v(qxcZp)Nn<OXj{6yfs^}o(D_sxH-d0lARwio>N
z&y5#P`>^4j_QB`_?_xGGCNB0V-1bTSz~QqK&OX@sVERG#X3x8;WdGYnTT8m`@qCv*
zYsUn&5bIyjDjY4>{W48wcPmc>&GI`yj~#RPC(u#5#Ih{(^Y;>4%~+POgyhP~U9D0Y
z|DRUq**hFQ!hf*vLHdEo2l^WpU)VnJ<ll_4(=sgwmL_nO|8(5Xt>e|l@Nv-v1)swL
zg%tvxjBJJb)j4*bOBb-X*O{=B{a{JC^~bXh%@cF<f1KoQF5yzYQ*I@b$d&qz@6uUG
z-u|tn@!dWT`dn>pPv8p-c*x~(NprTw`vmocvVTNG0tF^>?kwH7hcnD*rS@8*3wb%s
zOoobA9vpnQY)R-`qvxEBn=Xm(jkB%ZRh}F0!p!z`zrEzwoBwZVf7n0s>95Nx*KM9`
zsUwkoEAxTl1G@+N5~iQJ<G@mH_WJSW@138|Y!^s>{Ek8WZ|RS_zB|2II)6WJvUwxl
zc_@@C%`?QaPV|q-jO2IK4>XP#-8sR1IbHDM{s5KRTMu_uAO3hb`%s<vFE(?A{k<Q>
z&fcHmUDM|MMtgqbBbQBQKX$yIlg+Pu&hTBun`ImCZ9gdT<o5fnrcav>?q|Et^z6T|
zrnE))^^m<c9UXUFJpQclv}MZ{rZeBq-{-x(O};|w#y$Hz8-Gr{QLx$diGo4?#aD{*
zI#H!3R=W4^Ip3+r@_zEyOqa@9uGOI4I{X+ehY}VAm4}aP?zzW&@tR_MpW_4b?ud@2
zX(2AYo81p4>Ta)i8g;yz=V)NX2C*7fjktrmI0Az8MYp=M)%Db~)r4snx2{Q)>gdpK
zc+cv-u;Tp}Es;OMH$){|SDyY9GDDsBy6BQ6fjXICTy@|0i_WyY=E|ES)-(CU|7xzY
zYvzi~d)lqB#xwlJjP=`IT3!^a5SQM#dr^A-l*vq4i)VhgF~k0n67zS99U@OY|5pfi
zY1}wbyJZj0M3K5}TaHxiy17pOj?BOL|GEAI8~j~TceU?-`PRqg>#euV`0(9NrH<L=
zgwa7;xf-4yLLc@%2ziq9>RGS(x7{zl<ny?0alIkB&De5J<4xlbwL9VA7tXAClU<P$
zCA{izlU`eo_cqPbg=OY5-ajZ$SiJ5-48I+-kIwbx&5g#ZI=!##GS}j?$q|3v+}vzn
zu6vKMpZ`$F4#5Wx|Nk|#NSWAtV1}o8z2t3u_Bmx|H2%q(NKG=>yCY6j{K4S^G6(FK
zZYRvpkc+&uEa%$PZ-+y{St-DUV@}VuSqmP&|8V<V*`gmMyQi(Kx>URF({nM#b$j<s
zx$Q2J_qJPHKac(Q{<Z3UbADtxZ~tU}{`}4Ny)(V{m7YyKKJWgf)Xlb@w?5ZR+T!H<
zsG>x?f^|mIwbKXPY9Bmwe<#%4@#37rzUYIp2W5WqF-gaAo!390t?QKM#UiG7eEk92
z10qH;9Pil7tc2yb&dF`;P<i)#qFe8J4QZ8+t0a3)oNxXwsBh?WXYq;4np58I_M7qF
z+05N&7L#1~k7GM>F7OAQ{C_~CZco^Gu9b?GGB!*$vU?=v=)Y=Ux1G4Y-E`@5orq0o
zFJ66({w;4(cA)s?yK>jmr-iT0@_sXXy>;VP#kTp4zp5-t>tySSY<_Vor8gcw;(f*`
zM*NE5#g*xW50)SJxIH2K+>ydebqnE#Roh-W8`nsl>VKHWR-B`h&$RnV-_pEzoqo<J
z$$N4xd5zT;Kjc0XI?BkN%6rG~*?hamoE|&JMVT(Ue_q?s$Wpi6_spR}Gtt)Tb`SRj
zop>wsTT1SA^bSdx{oRU-H97e0`#;KL$1i99Qf+Yj{p<rW?7cstXEQx}Z!2vPy*Kb~
z`oxPZFKgzrTd<lRs^2R-ZF{GJz`yiNrMr5!^8QKbZtq_h{?K0Mh3BFlK322N?lP*}
zcS?kxY4OF7I-AcN;^H~{jjq1Jjb<;tr5}8JAod{l!9O2rq~`E#eDiN1Q%ry^b3I4j
zb_VX$^VuJ|D{kF9Cp@3$Pv@h#YbNe_dF`_N);_MQK7a1;?+z$B$lm?de4<c5#@Y|b
zkN!=Olul=><A`N_c3jeaN*za@fmHFW_2tv1Osjjnargay$LCm1E;o&E-t_x{{lZ<*
z2hHR*e%pSaz3F?Z!h*dIW*s>GC}^h$(;exQN9^xUKVWV)uwV9_*<P&WWq@OIA@hMh
z&mwr{vB&Z93ID(TvvD8S8UEnDboWN{2F;Ks5pG_~&%F|w`S)q@j&l!ZcrAQi{b0ie
z*1)feW@^;#4t93`@t@c3R_EjYCm$!g-ak3yv^CSh-~<0xiq4LEz<y%hN`qRnQ+4`Z
zc4!Nz^(`qs@bF8_p5jMG?n~6?uS@7W#Q%>^`(Vv{kxwS<pQF2+9RKwj*!j#?YN|Z^
z*m>F4$iH(RKFsfyC}8f&eOByIZQyUiHC^YF5z}JzyMK)T80gLZ`JvqJRMFLYH$1nm
z`M;n2-jBbj>GCD++g_KwDz3loR}zzdd&kC`oqJfW?l|<ZzbR`!!}S04D*yh^VXx!8
zBP)}7j`REaeP17LEX`m~%3CcQZJo4J_|5YlXK!D(DY#pw>-?j_;P^-BgTe<@&P`Nr
z`hTFo`Qg8!b!uUHp$Z4K*37Wr+i7xy_rUp$OVc&<WMYK765{R49^I)knpJ!E*^<kr
zjr7>2y-7N;<L^Ok9mCFQ^-FRr_3VFY4K^yz<@>km{px>aKJ3pIM7a4?Px%!VFsYd1
z{>;7G%^&{fnvpFYDC~TIZCBFkZ#y$Sd`<8ZxP7p_;qhN)Wh=d}4sY&0`0(J+E(<{$
z#yW{JjjJcD*m`+}MxA<fO8%lM;m{AvDsv>go+M3ppPY8U|1dBAt`4n=uIt+i7Vo?;
zSEw&X>EZkZ^S`}2en>yDO1Sdl!n~rwLf#op4F<N|9C}TQ^$g-1;_70KEf($A!}=$C
zLv6va)7Pgwj&pg=uF8<0{AhX8e(q;CVsw_=ciFgSWyae-GMlEdMfUfEx$L~;@}QCB
zrzXe!>z=nC{4oCb&P~zowZrm)`v>l`y*q#4&w(fIE4)R#igf;K&ZrZ5&d(M5<MxAJ
zAA>$9KbV$qDe{5(ftOcK7R|h59{7Q6gYEpU1^0ex|5%++r~jnsho2pDoqUed#DyQ`
zv#80fe9?a5!~Bq$la}<m-zt2NKIzHHu%LDKEbmKSf4}5rLSDy&Jf`~N-z)FA7HGCS
zdVfLGtVsE>&4#~v)(7@W-Qn^1WizAFW9F6PMQ;y?Y=~tuOXS(Jwd2sNy$89oj%q2e
ztP+fv#wIY|qO!kX{{7t72mZWW&z|nJe38O_=H6|Git_~|k|!3tXRw!V$$eKosaC)K
zy&kVa!{@1iyS_&5{IyH(T>B!UfA51P&%MH1GvWQ}e}^lzecqU#KFd@rqyBYWS<<o;
ziIn=-M>CxJoMa}~u|A7%x7vH_<NqR=Xl}mKyXWZayZ%;P*^ayJs{hJ=hs$%f2+P%7
ze!HRLK<d2fb9NkN{vTanS$-?R-DXi=_Mh5M=Gz2n_-9;^jp19Dps*mtB6{AhUBZEl
zYg6SK=HCyhnLb7O({HaUE`ksGH_lk0DpjpLgFABio8xbfseYc{ab&V^iUrR{@rLIO
z_RM_z1|33{95vGaDw@C4b=WJPFqeP#^MQGDL9Gnmv3Qp6OvR7>aaRaih)-FO|8Db-
zH=21aQgwP$envm}|M2#M=?7Y!0t>daGTuJ({h-kt16N=6M&}DxqbHSok63b-&Cc9Y
z&b;%#aA!yNwD_j>W_`aYD?5*W_r3I=&*Sm$8C=gEP3eB`Rj{nYma9%*qlP)=+Dz_C
z`AXs|%wpz>h)JzA?!Plf`GTnbgNocOymIT#zVd$eJ0*4-+YYfh;h47h7i9!Reyw9H
zmQgZ1$D;f9VLihgp`0r`^GaTC+4OSPe&@#Z$2E_bo|}H4`QV?h4f_tw4!zZTt4?@B
z=$s6hKLU~(cV-Ht*huKz->iLwW0H_ny?d9nd$7P?pF?+UdY`yzz5KxG6JNJ)^4M{`
zf!!kU^QYZU)jn)*>JTgVdE0nZe(jp)c^@Lf5}Bs|=N7r9>oH;7i)FnB{yTqP-^I+h
zf3JM`Blko5EjWsKPyEmRZCm`;TvBFgg!%DViVr?Mn0&x2?eOdV$oZF;i>^=0VP45I
z=S6j6aijNi<pWk#4f1?{3TnE7=09D#{`Dzc3$Z)g`*_am`y&~{r&g#}apZgGe3K~d
z|Gf`%AJ{(N+Awi#^&LCm+m{d299#Y1>4Ozl*d+fmKYO1j@FQfy#qMXG4>!#%KG3H7
zR6ghU=Lf+D-fR}U&;EP2;2*&XeS?>?AACPJCEbUqxMwf-AEOx`!^|JwXQ*dTD`lC_
zyG-Y0`h&oPbM@iP@$8e!!mR(@_x<5*lh5vc@TUHhR{I!d$+>bB(+;VY{kiMUyq{al
z)R5(l<Aw{n&mElozb5{yle6WAB})G;7hihVpy}1q;CrgGIOP4YqJVu$k;^(VV<+?r
zOQ^5B-`LOgylc^y+;UsFI?Ge{c%B`a%XHjnk%mR_A(KZchfaoG*&^^lZsG3d3jY5&
zejbY4Vev4jySse4p3?&XbEkhhIM%r9h*!OH=QR#$IQa21Q$cOV`z<yP7{9L6=-u?o
zcL`%yW7^UWUqlV;PvmJ_Ff(gQ+v=wNjj^A3K6jnqnaaB_<SKp$M#M`Au4*_GD`alp
z8+P!s%@5gbjytQJ63tKW=S=^xH|W0A-CsgKk1IWV?WgqnbM%Ja6VLp7o}%urx989Q
zg3kL6A3Kr_t{>7XT&yHp#WhphinXt`zDfApbo2S|ZEnsfWwbi;e}0KZRxSIYOPl9(
z&2OFm{+8%A!<qkQzvQ?Q;dZ4&;KBCiuPn}Nyr#+}y`5E1#{Ta6TNPaec{exwcXIi|
zTEqK8#Nc?!;;A3*q%iJ?*)3fwH*foap2ZVP<<>esDN}F!ue`o-J^!OiYtOG-6&Lp}
zU#~1{fxuLWdM3MbozK~h-BGX+)MGt(`oQ@H%_VYtX0rq<X5?Lwle2rxXkeaoS(WQz
z<h2i*KGmD~vwRm*_t*Y0pW!`&J;$^1=8NmkGGCutJ>$85!nbm+8T+?t8JyHL7D#-i
z7?<*|?>)nOezmXc@~qE(&*!X@I#bolvwwv}^v++q@)t4vyFW3i={)m)fg0H#ejiLe
z)E~?}ut=??!CFA9YWsU<=AVHRxPNJC{pI_SpYTpW>MiqmOR)*Fg&Ma{Y-Z-Rw~1rT
zXP2IOuvmG=de@Ybx+jBdSFTFDUnkIhb9qj{zF!kFqaSa5zk9_SpAVi79viQXc-ke!
z7`?>4D^7;x9m_rozkatPo!+8H|LgW0ynA3jf9gI3k+2PVOp^tVx@KB`IGCX45xYx_
zQEx_f=a<<U%eIT|U;E$j%fp1aDf>k1yOTCI9Pkfdw-33`e4k&A)9e?!&7A$Oy?-xi
zTHmHMQ@rKpL2GWGg&z4I3?KYG5Tao4ipO7^`$(a<!aMB;(FgMzHwUV(^7OJV3KBoK
z-~I=e!SM^N35yH%>)J)@wVJ;Bej<)j>sk4o`v<k>mT>*LJhwQ%_|)aoU-O&;?Cx8#
zy548_TyA;$We)p~cI$^L?rD7e_&voX@nT^@U#s3d5t*WUaZF;IjX8=Nj6)_`?`xTq
zbnw&+{f+HT4-6mh`c1QbSbq9b`TPu?WJ9i4`334O*N&CE<_?`)IN|($p_qwF*tm|~
z{eS<gV_cu|@uqgxL@SYX`#tzy%uf(kwil{jb6)J+m8W<92S~O4@7fa_{N(wcw|QOn
zn?L+fNxjCNxn}lTL#^ia`{hqQDHMf2kgDV9V=L~-ImghkfAahv`=ne~hgh)ReEqh-
z=Gkv=w<+y0lXaReNy#(Cw1{%gJa#C`WQMi<%_YABZt6bJZ>kqr_2Wap4CVtYA2J^V
zADrV5)%-Xx+T|PLb`GK6hi$oQG6c+)v)jlNoM-$jU#MUqeuRIz@PYY^!46Y%XLWMr
zGsQF7@z{u+Y26s;U9l`e;fdS9hW+8v0c$HZE1Y9X?p`jv)X%)$$o}`Y(-wc60y_3L
z-g_`F;h$Sf((NkNoVO|VcWSPgADjJPR^giU+B1$As82OM@bLR|&-l)}76#&<*qhec
zUK4XUsv&nSdV*KtiQNxmH_j=w{UiV2xa{q7WonB){OA7sm34FNpACjF4@5;dnAR_n
zb#Jtk+sAp2;n|~Cr5n7zNtZod^y??%2VH~ud^hJa%+tF|^tL>2G;h4g#uc}Zd)t5W
zirQTQ_5a!HB+lG_#QUH$flIHCW%8w!>vy^Qs{i@XZ_cxZ>&@{j|7GT6=s5-*+rq1Q
zpP@MK=k#Fi|N8r%wfy{leB<v*gO3S2T@$8Ba+*o8oxgveJ%S~s@B4T5oXW-5o`(qE
zoc+f9qTp5=jykRy$r+z7n0yHP;GCe`ml@tPpRG9dqvDKO3${6YH}ktJO4&aDZ~V+u
z-(ItSd&BOA*iA+!>)Ox6Gwl=k<Nv|wgVcxa2l35^OO;M77yDr?dB)7(|M3U*P3(=a
zUd#Wl4fr7SA>@PG2dxj29|#}Nb3Iuu$7Lh<A<{2Vd~#mL^ZZxuA6$J9`5^k>8hamE
zfsfuFq#uYrxS8<iXt`(F{Z7x|#os)w&Xk}3|KBIWhRc?rPWFu9{2$VPndSB|3mv~^
znD9^k9iy~u(!2w2#kry<{I~xTzeoB{?uVL!+Se|knlB^N+3T(U?44}!Y|oS5tN!zp
z{@M9KX!BdYqq#EsU+X$H<Uao2F8J8uzpLuA^mt1?skw~j6}X?AJp0^wz0>@2?EmHd
z@cmFvm_BK3=-rxU{ZaQ5v>Cp~9{<1T@p<!i=N3QJEn?Z>_w0t_qjJfcFOn-PrcW|!
z-YMqi8spxYsj3vma-Z>;%(;A%bC>l#*1s~^usd(P!;I|d_Klto4c}!Q**M4BxXD57
z>&~w;>U|&FWWD}p+O!i-_ms(0cxo{J{>)TS{~<>GZPky?(iuDCel8N;$?@;>!8OM_
zl7HH8yOhW~*sW8It&(nhe9)}@gY6r(&G&2Cjvkg)J|@q%Py9@cUg6pAr3Dpn56!PT
zCe)c1@R&T>{nz-%GKHq|b}TjGH{Rw?cys>!<mLmj3;r{0yI&~qcD?R7#|Kjtl(f#h
zo-jW*S=!>bglbp%ud35ee)zY{_fZN9Z`{wGHov3!;>u?ewOL*%7?fRF-`;Eccbe9n
zrdpS|NA9>6EO=nAX1~co;-mcG3Y9-EFPKlbdFs;J_abS3^KMOcjLP2e=lIj4%f8F!
z7sj*hz9P!_@3B#>)DIJb|3BBfvG`$`QOFqj@<{Zr#~soPWiRtsbmo>ysdjBI`SzZ7
z!&KAnwnuh4X+HB075}{B$jKA09ybUri|cr>Ah*)OLMVB0i9rQ#>za8Rq8m-@x~KTf
zy{h`?zwcFZZP&)Td>gjDwDjh<vwe<<pU#{qy7@NWzb&}4)^FeDhxG?J?S-D4T(VwO
z<9G6g+iBa9HD?IBhRnGsp8enWL;K4Zml~#RHqYu>b~HQBJp5n3X8OajGwv~y|D2Gh
zezo53!42zAQ>1_Fm+3s_x2y8|jmvpIgO;4RdH##Usx9Arzi$?lKgc>~>SBhYs&y<8
zw{ANgdg?c4>CSim?{GbMzUcBL%VV?VzcWAhb!GqO$rFFen9qCpyz#E##a!m&Pv)Hw
z3D{Jn)Bk$ckJyK5EOmX~rzTWC+u!`-TY<Rc$7y-*@|MU-J$agc>e1t!@s3qiUs<oJ
zmE5s&{ptJW@s|GmlSBO~YClXlR}f$D=x$ws-a~iikfQq+^8+*ZZ>;VAXMS+)LASh_
z75kl^WwY0D$gDek=ugCKhO(9BOu+}I9WM=6-WNH+a_+z4I_18+*k{k#X9ZrV$-95|
zKd<>e@f9WNH|8b3``q#PPwExsd8rBB{!jR}F(sCt$=q*uSU%d_FHg>Ib$-Q}C9~%r
zvG{ZU*R{!sm3^r(hd(Im%jx~G<zaDOD0)I<J)e0V^QRBr#nt}x^)uH?{&6<AZl1T3
zSwO9nefP)kUmHLEPl<ZY>R9?PJ<{fN>FooiCuI48HzdE5Iqcc_YL#E_|LaFDUn_bz
z=TCg;@k#D;d8QxAe;Xllq`YDG#m@VT{~6TJ{Aq9Ky<-+5;PQ`W-Bs`JVjp__V)m5E
zO`1H(y6(RCDam7c((mW=?~u*;8*!$iLFlM)rN)g@7ydFSJ3cgvdvHAL|H&nG-!9Jk
zZ1=}(UhJDA^%1dq|2^YkJYiVZ_VMTCm@D>WmvmnnzKfsMy-WGpO|G9me6QO0{WCh!
z%e1{`^0aP^KJPz%wsLpODj04b__lFwqL%BCdk=&UJU;O4KjV7;m)Fvt#!P=~GxNOH
z^dGsuA`OmTs(8Eo*Nwbw|4lvwKA88w<%%F@OuRhryGa_s8vh(ya%FxdPdN5`F4sHh
zmcxAaEk9-p`gI6D%TKrYQT(9sfyR+K8EJ$2`ASK5lMgO$?q}p@Ij)>^@K?vfjDkJ)
zi$mtz%&(i?q8VuOxs<7RZO#$P8wUSdUD@9G&ozmbZrP;KVfg)Yvfuul7atz(pP#hQ
zvocG=-Y#a*gT^EEbMk-f+g1PLd>MO%;E$IIGn=P9Jni^+eL}gyi^dPh33@XgU+nDo
zWBubnMQ7%b!b>%LXIK+wSbX@T{FC9spMs=mhvqMs_xqqY>(|0hQf$xKWm;dIZ(5v`
zAZ*xmMbYov8)a6-mscW}FXB=9ac26#<E*Kcib9vSANcZ~->gb#UY(f7gQ87pJOBP=
zJ|y#{*Fthx_x!1UrEOmAKe`xp7QpkU1BJg2oqxWj_|c)b!!zGsotqM4fBM7!r&?8y
zE+ib;%VEd=N8jM&-};A3ernD586(xi@PGd|w%h-g&ep3>R5)<dH{yi3HMehZ{r}TV
zzbyRJ77I_P(3z3!G~-ZexgxW8-${+g+LNauZ#h)FW1h|~C}XX`u#wf|c#7VseQf7g
zbsd)KH}>;r-PbN?KP)Tp<FRGVd)^1m2dy8a7Cf3U&#%_+;+DN1>xA=EfA3`cQO^Ew
z_5(S|A1~Li7B4)zvm^PAZi3#Nx8-d4D!<&CV&?F#V-P-UaP+sV(RuxYC$HK5o%m1v
zgT#i5y&rtGIGlKToOfn&xz*qO5<g>QepWwaXSJVmtD$$%v(0}_-f<~D9;i{G@}qTu
zf#ECRAGv-Syoc5$zrWj?{pR2OMa?Nf_pe$$h;Mv5_v7{g-5)gtwVkh}`XaUGbp4Ra
zV~uCZdwRtE>$;xF8~2qSTe&Ih%7LhbufCQqGOgf#()&G{J9J{<hVm8vS~asC{a3x7
zf1m%;#@cw!KY_CU;`XuU$%^dY(0jhRCeV3L@oVXi#Vhx{Hw&!Ix8#3l^ONz1?}tdg
zKh=Hlaf%c4m~IC)$n@`UG`Eg(>U;E~{zK7<ylsiT8t0nk9RF$W*ysN8n)v#@vmcd|
zj}`uP@pFw)_T{$-y~lgvL)I!E!Arrm{n?*GUYA$%>~Fuvd*e~i&NuFR^4AtvUb9tx
zP%umS*)E&V+Q34-{^Q@~xP0`9&!5RuK0)}|uaisX`8`lRuAeXLuwY+z>oJatb_@Li
ze>fDiUYx^X@uBtvOZ&Q_KkKH;KmW#<{PB<blXmOQ$;Wj*Xg|nJSls-;@(`2cE_-*&
z>}u9O{0;x#Ub)KoPkU~G-X8u|t-IPB-y3H0v2}{EhyBf}xU}~2sbxhwU)Wk7sj+vv
zK3lz@USP+!#KQ{W7e$!XBq?xilus;bb@O?guQhp}TvO5WO-$`Ej?UFb+z;N=j63fA
z(C>%&R=&b_#Xs)lvuu`6OuF@;AVolq>H80s%q~41mLGLhmcEtJ|Cah$2){q6@g&cw
zRv_e1>jl3Lxesa&{`o7%TXy_nv!I{bdF4X68S<ZQuF$`^*+Om#yW|g_Z=J;l&o^0E
zi8onWKFq0pZY_H2zvOS<M@fHd3i1og4W4r^W`6MSzxKypoHL52UF4~#llpME;PT4z
zuRN~%ewp17&VOI`_R)RcS?f7-zUWr`;OjWLNnuj*1#_nQUFX6?cCgh{uSlr4!>jsx
z$F27he#AdWnZf;^^Fi{#_0}KnRM=>8@VCCUOIABH`M~pLdzQ)D8ZN*6?^t*8mHqbv
zFFD>ZrQQAB3{K1|@5r7g5dSeT;8^fngL-DSBg<8PM0Gcuu6!<0P<vc{`Kd);@*Gyb
zXZUY$rK)f0%S%ckKX@yQX6)a-@qzHc?t^-F8xA~u5SEu>`*-Jx%eg_^<`WM6SGuGf
z{A8tn(T}@2Jpnh@p4fc)<nf&<vQf=v{>Ym@s;Sz>Ds|%QS8M+(Ygf5&N*;@@)nD|7
zYdK%v?K%B#wr>~yH{a^zC-5~9Ee?m3KWx0*(*2-b?#$K;#Si|on^j%3DVoj3wA6R;
z?1|MU`8UpX+WE84`xnO_yBXZ4;~pG7aQGm9bMH4pf2MM?NBjAnu`AhkXqx<Gt}v+J
z`BCs;VM27t!PoKU9L$BKTAh|#u<mEFx39C~FZ=z|wdHn0vFk(rn|qts0}Cedu`O5r
zpYGhdq##SFLS+W`m6F#k6Fo{eifjcQRnMKIf8hTzVLRSBE@`{RNA3#znBQd0_-Dmr
zH%0#6j%JQ~CcmlH{`fuoLH2?FEa{i7uMRx7f=TY5TvU^LLp<Yp7WadaH-&`e2MR5D
z*Z5o2pjChF!Sjv1KX%(6W-7e%U+Epkhxg8zHR6j7xil1?I`8be+fbxW%7#l)pDozo
z!-C+B$B*lqK5T!;$5hX4BfCfP%>Se62i*Ri6+Zs-|0Tl&>C>;H<L9_v>H5_qyi2U^
z-~Ndf{1P|+FK*rZc3r7f4tIUF!SkK92W6)J<NL?rbMTbNhCoZ3y$5v8|A<UjJo7`0
z_=n}-q}cJu?Wb?c>&U*2j}L?o$XMI)=l%PC^EV&E%2liXTc=jvPv!QGo~E@|vx9f{
z4etQq-k-~7L<IL38qR7ta^5A`=ZyG%k?PK4M{WdcT->3wfWu?rxlOxsQmgfiKmY%G
z?@X**|IC{+Z$?Hukg=%#z4QCOd$Yg4vwpv)_}~)V2lmp2@4tR2ji1A~kz;zr-rWLc
zlB<<xL}Wd(wTo8QS@_lFNu|*9)rUXN{a(0_=fT&>pR5neI``4F;(XzcM|;BO@YhNH
zF_}@ExLd7Y$LHsz(ZN<qd$tAJrOrP8JmB@h$n`Ei<5}y)Ycyx*3!8kDQF+>A+%x0)
z{6D)p${Ww>8zxPNd$d~lV53-5q+r7)!Q48Af2KbyKZqx2f4uW|q2}b=3ck-#wX-zs
z=Va8M*zzqt>EsT%mrEFGb}h{_nf#~s?+vAbcGb7~Crv-t`?A((itGIq<gJjsJY}(R
zyMm3x@jnUvucKcWTqu0_x8r(on^RMRf&vEz6El0Gx{ve2sNW$U&L(`G)t231eZa|G
zkg3K(Ea;VE!-4h<j}JKeHU!K((9U{Eza~eervAqP_tami`%L8D{rKP(8K&jVCVoG3
z8B=}t6;qA-ta@LrN~f)o&-=n-@?g;%7uMs!ZfAZ!_hs4s^HPXT4@3T^rHf@vB3ObN
zoma89H#5qMoC!0Za*(~}YR?ain3=m`RT7()Ud*2u_vG|u<(i8jeyInv_nAypp5N?U
zo5_}&@Y%-4)%Ie%;x~I+&3qM^KQ|t^1pgKnxc=PyxH-R1rRalu1#VLtO6IM65_;tS
zOK)@KJz{$<Pfs%tJ|oT?_KQI>WYLH1TbTZsZU~$6W43WuD;on-jLU}^SylXkpWiz?
zZx@Pd@Ji9P?d|cenYm|OmX~8omzkE%&I=ch)YQAEPWbx$ig@L}Uk1-V{_!<V<#GR0
z6g=y>)q8oBX|*{=7X4R0r!r}Oaa^@W_TI+ESiMI_jP6fg-s}+Zs;F2Zw{pta&y6{%
zpDh!0e=8cDIX;!k)>pS<Mt@^voekTPlMkv7>{&Lu^LoL&_i2x+5;`@TuWwsF>r!P$
z`0PtPzpVMWeM`d>&;6dWOv>dR`~1hJFJ!aaJ8Y`!zQjdc;M1+s*@l{Te&&BSGyeB!
z)*hih)*ntM+&gX38s&eTU+mK?$Db<SJ_`PjP;+SBbFgl{ws3Pvf$fY_{=Po)y(UtV
z&E)xe530z1Y~AgX87Ka%`FT@i-hb&FyL}#QK5yLPPv3aq_94j4;aT<qbM-Cz|0-U-
zWqkel+Xof#2SZxc?)!0&t3u`<)4YsV1=^3^&$x7O(f5TPW+y~%@Nzs=;TN$!+HuwQ
z?2892p8d6Av_GV8^0ZdEP%ipj`3eVprrif2DUro_!R?TYzS)~prBYeq8f@Y=*K^7x
zPCWbiTkwP12gBx0U%E=A--Kn3`>htGIp<j!Z~YbdSvt)~uPUu^Wy*u*j~f$wWA+3Y
z`Y!F5DyG!TU0QF#IJ4%*LLWPUS@LE(HuO#5^5Oki@IgIz)@Osu1uAAUjTcE=_X^P6
z6y)xcSf3!^J9Cdk*o<SRU!+g_lI^jc-GAE?<4S#tOYFhhmM1-tb7Kn7C_d|SJ0<w{
zb>Bj_O5+XZE(R^Vp2C-<GAk>SweQ@c=38bTB;wn`SY8&%{u31ZYyHY)PHx|4(X?Mq
zoNQuD>WLOHDH)5DURPTQF8vUAV&_q%*U}#U1#Hevx*uI6tMy^d_Mj!7))cLI$>i;}
z;eK)byvNQyk2RcP;-^>e@8Yfzv|zjw-nDIy!{d+TeluPN8D=bF3Ql5s6e9IYxt7Pz
z@!`%ot%ADph)-X;Ip^kvW@Nc;&`W>u-bTkD^hRda>x_wg3RjMAJ}@Obr7L<d>$0AJ
zXZ|(&`OALXDm?pHX~vE?#%skqTjgeb@d}eU``W;k@80C^4VLv<dGU<@x&P>HSoSaK
za>h?)!zZU(avz8$Xr)eX{CVj`$Oq;G<F%%$6>I-(){O1_QN66hB5?Uqqj)}D)?!=l
z#mvSpU-w&{kKbdT67~CG#E<GZZ?%8abHualf9h>4!M#uD58n^rf_P1FrgO}nZynhB
z_-2CE#Is+7Rd$_`bbK^B@YwxBM;%{%3HwqVVz8g(&Dy&9sq#I^_u`ZJRi<Cp_AOWV
zWHaOX_P~U7`+v^9ay`VQm4(sKLBN8carQ*#r@HeK?Q9o>82#dIJg)G=^Mhf+y6iUz
z>n?6@T>CmKg;SQ7BSKuu{qP*;naPq`cUkIAhRozI;BVN^YNpNY8`|V{`tOu;j`t2H
zygqc(|7yo3E}ylR1fDY&Ds-jzHKs&<KHKd+?X6Mh3x1wo&T8U&uFpF5R{s?HM*AlD
z?IPEd{`RQq-$^l*`ceH`=3ET7&fH(t>bjwa{Gxt8$Uo^d=d?t;+SK<81DaO~#Yb`S
z>ahHfexC1gKln%V4yirNGMa^rK}>pV`E2|7w`u9GNO{m)zeVHLr!$Ye%+;$pixaQO
zAGv?dWb%X9eHCVh$`300q}F!2MbDG-)_C5sPFjQKtHq1t3xW;;@t?lBUCS1&7hgD`
zvghBg=S*>vr)7p+VV9mAxoz4*{VSjEcO23_5HhV;c424N5B>v>51xGx@Ooj^jVF8;
z9wk1}|1x!M&xD+1_A`XtJ#B89`>*3Y(RiY@>A$h)tLjbXq}C}5w!}x>+A!byK+NmD
zEmI=vm8U(Lmr{S?jbL8aAF)EM#hxD@m>&qq3u|g`{AxPsN923=6I`<^dg8b*b#?u5
zbv*S~^ltZ=?!TffGq#I;z4890;<fkdv)EtH_<sGz-;+;IubsKC$*INuYw*j|C$+*c
z@0a^*0>|(TZjOq4A*&x8AMWixc$|HI`OCrsiVH;Lne!6ZD;@s3)SD*g&DdEH&6>wF
zyR-Z9>@$+JJ>m;&*Or~-KTs=D>F{|6bJ3slf_2${#9dX(=eeHpi{HhNxW(`V|IB?4
z+e9MdJ#K0+pQ(A>%y{~j6KjSHU!CT(b56A#TTQrUz1cJ+%eCL~_8N|&+1yP>{+{Fv
zitXZ4V%l?V55qaruRo=a#G5Zw?zHQ0tLM68z?d3RlM&U<aa*K$J$w6VpN?%dJc;(3
z_C@@X^7^{<Ld-_N&Ww2NEq{J?^jzg^*<JHdUg?V0qP!#j#s3J-*uQ)Ffrmj`p1%B7
zZLMCiH(}vVi<+J;d%l3kXw%|<D-&LSEvjuOdzSP1^dYa}#0eEUSDw*ocHOf)<JRNu
z8L6}DyLZjwJ3QC^_Q&sk6@Q3*=uN2Amu&hGelR3!veVk<YwvE~AG1GB&cQ|dQ$61^
z?nR4xUl}CsYj~Oa_jtC+`Fh`L;sy)v%`UAjK9POn@`LrijVg2Z9uV(0dBH#P-?2>`
zGt*p+>c7YApLjMU_VtX@Ro(4Pxz;ToB<^0HZD9ZOk5(e*4-@Z5jlZwd_PajGnfsyj
z;g;<ty&9Dp`YK(TyEk0?zex6d|G|!I@pFHr4{CYN1UE;(trD3i7MGdtj{fIkIqJyA
zD$jDJd;a-}`-Jzf)@c0@EZ9+^E0y7DB`U^!a7)|6+sDce9B=sTwN#C<=7asoDQ51s
zCv33un0flQ&~s<==UL0<?sKV6xFBzIHkn_{-(69ZH|){j8~z9XpO3n5yktl5-Q<_~
zD^z8BQ}()g>{C6PGEZ~;36be99(EpxH+XY%*}~wTKeB^wNj-hvc&#G%KuDBGqq5?*
z>)#FHoX@j_|7Mx{p8GK8BHi=1%-6m->ymoXtJux{xVzK8ztimIeO;P9<J14f9bpFP
zvpzMuuI0MF?-cj7@c7ebjyHZ^ey@-5{mHO9OudP6Tq{nj`w}|iJ$GqBSjuag`h}IY
zMM_#nn!egvv3(5o?JfKL=vn2*xcLmv_Qqc3?5N=}yUv~ZZ<lR-^!DbTPR%oT87I3{
zHqYC?@Y?H4fy5N4TfQH*)-o$zPdt1eCccJi+3A+W6D#J$u*7jZE1h{g+PJpy$l8Q;
z#%I5n37I#oY<YgGP%=WeKT~YRbH&FCFYpI+ZaSlHQ*m}Tw~b_t=8We>+%t|C?PGap
zI`?nHb)geRW_G=8XYT#760g0>woiBti<$7pTH!zW@oD*+&g7QX7%E+V)uR98&i3Z*
zKYtj;cRXNN_4R3i&K|p;i(gMXS@-3AvG^ltU(sxV=Y~%WFMRSjdhGP(`-+d%6Xb-o
zE-6<!p8MQ--pXd#wT*2}u{)j%?s_?s|D)k~-{Wh4<|NB;8(w@q-|~J{oSV}7z_MqH
zWjQKspNb3o@r~i}W0|~<>7Uw+9o(Eu;AF(X)O3KsrGa~i^2}oiMKN<M4)ix`ADk#4
zZaHP)%jk{^T$4;rO1Jzi-?fr;>Wlq`OLtauv(1!0!L2xfKeVb*g7@d%lOE2k=?8dB
zKQb;kH@$;}S?lbXBkk$Nu1c5h?9=}^`$EL|t<r_(jBl%d&}!MwGP}RCQtw*jk6Em|
zhvsH?y54H__*}kFQ*z?Z2hR^a+qlu@%Ez}Kw_fG{FeUw6^Zunw8<r`_PW-UJ<$+Ak
zuc*@}=DYkV@xG_G<mUsAUy?tzyPa9Rh4bs3Po^(9gQ}A5WlNurx7)$b{`n5ybWnP;
zpQ^h1pmGiC9yuN6_g|Mvo?-20wP$>%EqHzAgNYATK9D|;_VmYTk9rohna}h3?!VFf
z_B%ZOc%%iljo=+tJ?7b|Gaes&vhb$kLf>U^%M||QeptM056=;OW_{M+jiP*lpIslk
zZ(JF&UvN#Pwi2(^MeiNwLgsb-uze7HaF1T_K~s)3FJ<FXn$J0$Fyn8k=kR0t{QaA+
z!TG%f&+EBw1sML-H~9VBv`)Q(V}{F75zQ-A_iokxw?3uv$l{1S+w_LygMWAnWlIgN
zTffLVAP^rksrbO-gMaRRmiQ#IhV{4dwe#%7XEz-UYRu-H=U!H@;<>?k{_uv%d!J^#
zVO!7kxu&pTrAbO0_lzqX=NOJR$0p3}&)|3UPRp+?jQQU1p7pbNG0zRQbu9TTv#-C{
zd(!DE^NZal6y3f4yE@-{Ut#m-q03I2Yl{#5u`PHbn_%qg+;n_O_SLZBM$Lc4N9q}$
zZC2{%eBI`4^Uda(4D&{#V2vBC)n_#S`+cZQ`215;XMSqqF(D0mp~}e@UWAx^xp<W$
z`~}M)`=<IQ-ol%649;6BIywjluz>3+i3ScCzf)fp9_Z(;6%yZ;Hmj9Iyfx-c;YPnl
zcfuMk&t&+R_&P{g?B9F|@5xL{t~dTXaO&u?nLjp2a?MN?n;})h`%!aQ$%;=tzEz84
z7OOq_oc>|!1HX8+sWKr_M;iY>JpJ1HK>O{V4{Sc9d~i+B`e<5cSijp(p@)UpPUf@4
z=fs{8%L>j4o*$+KDn=LAgljqQepZSwUr<!TK7-%P{>n^_`Rw~y>zI5V{XEfr>b>#9
z#Zt14GZ_DgRp`#(m;GhqW}?t)9d5@^$9ZkPyxjcg10j=qSnL_(c%R*<K0MRQxc*3S
zckzsO+E0F#^8S!Ckbfi3oc(C$-&KAd6%k4Wee(nNzfoT=RvEdkW?uHJidcD;d#jBP
zacqxxwXgGw(TAT2T1JPLW}N<f{r}$+f8yg0nzX!`{9V)T%jaGGHmeT$H~jY6b8o`m
zKz5`0EUgC?ejk?Z?=oPV!)JGS`g8po$)}!dy!*iWvz1t<&Ey?*y>{#`%bP#f?zL(9
z!1f`};C<}o1BMTz58giTQ0Tf;_8*fZ`-PV_uQ*}I_s#T&pwpvMpIg3j`MwsYygU7e
zx|7ze$~zuw{LLoIb{0I}&yf49p`P{G`|js2JQI9nn4zinbhXNR#Xa$#{`{8TFf-zt
z#t-%fVZ{dnMb;mRIk`}1{pA$xk9)E=rM1n_bF|$wb4S}}_5RoFd+kiGZ$GwJ(Rj+f
z3O>=FH>}u-twbW9{kqt(y+uPsV4e+Ih{$)hHST>=m}Ikc^+VnqvV1n>?gHx#EsdWd
zbuUCPJ>BeU@rmo^0hc|~vL0=%Z1GcAa;Knf_p)U|qL3sez{1kl+{hqW)6Q+%**H%u
zVT1pSPw5Kh*l*@+s9TZpvEX0lgcS`YmxDd_us?I&yzShjr!^b6Tg?_;lHAu7vZ{vb
z*`t>s2I8T!oTe^+J^OI@#2XyfPOZ)suovGWIfrH0RF<i1M_$W6*e7({kIyAU<*kvZ
zTN9)Bv!@ArIhHbI?DbcBF4^f5|9Jsp|5UrUXI-Z!SV-2$TAU8~;C^I3L-5XpqDuE2
z|L7H%&N$M}Y;60=+ou0RQL$q~g-Y>X;S}2^laCj7A8%Q+@~}83Yx><|Qun%fv~oOb
zrq5^j{HKe3cBY`vuNyNztJeKca9EnHk+Lp5^jgIE?tmv7`ReW8H+_9yr1R*BhvuzM
ze|jG$ro{2v9Sfh)`9UT=hW)OUkfX;_|NJkUS~b#pu7-T%JhGqjnQ7y97A?2x#s3!_
z`rmN&yY{w+TOV^4?D%xy$$|jus-41TUh6*hezdDGXW2d(_RTWKx)1al3n_iH`crIh
ze8orK&0E&G|4IvvGk#<vS8woi=3*Q36%E|w|JK?m{;P|hzqsdmVbF@Nd$#`mS8na1
z!!}>sL8M>z{m$M8^6V>=Z-Of&86B32vqqPLJ?^mRG0Ah!W4QK(O=;ns^QE<%tf0;d
zxa4@Cpu9l-5?|4F!P4ncN!n{?EL3frnGq?h8Tj+Fd~kB#Q=tu^6Z2MG%AUf1GIm+i
zmNsiynX}1<T5segtW|KmvVqA~@L`F;R5g{)7JDog{^w2IxLy3wI=1_Xg3_97=OdoH
zNHJHMc<PGLnM94^Zdc}c#n0Nm@=yH7bEf!%;|4V?k@^L~9Pw<~tSy?>Py92LXKIVT
ze|=_d<fD)a8C-3f*FDyMa=PKWF6*!Im7B{IKMBvB@L_{%!}B1G{gb^e^W1m4^E7+A
zOybOyYmSLLFWqjw=+HN2Z&j_M`V5=-gC?6B{JS>u5%01ceQw@y3+h&!c;ft7Iiddg
zU8N`bEZJNt?$e_Wm#>lZxV-DF@`uL>TH=*|x9?MkYuh7zAml`rQe6$_H}iP5_X~6P
zSv2P+#GOC$IiZHzLZ~w1+4MuU>MIPcB|kE~VUu>3aiaxu%o1ZC#oIxSe`aawUwP5J
z!o%#cxW^o(XSD&r8ppqKhS*8(i8lY#YNI#5p;AsT-S}?*k?jYr6<70rs^Gt-Uvzlp
z|Jsajo^6-BPwZe>vypFW+=89#r_%R7Gpg;k;{N%3qrj2Jf3iQwK5##{i*L~}wv{~4
z4(Jq)3lm=IUS{~=bD;XL`GGW7tLp|Kn^4*%ZU-zIj;F~=$v<0Bz~@$TWBH?oBkToQ
zS<jk&E=a9AU}`@nc;-w0g`IEyzHnRe=z#dlqLmdt7=P@3a5F*6YsCfCkBsv&wFGSw
zULV$#->{z1p5q)-c2w7i<Ok}tl4-6JB6o^-%~29-w%n$E;PQd6M;7nQQrRYDui6y<
zg?Gumeer$am0Ni$qYuV6{byUp7Og47qw!eogUE*(wHlY7sR?qE3ktt>S}1%lH_ukp
z;#6<^T*GR^{{HJx$v?6_uGMP73VL%)KiWJ|nswaTHC-v@dH44+9i@Z)`eA+yo6mb6
ze3pG6YDwk1Y_2xr*DO-<h8DtSK6ml6cqaJ%@!iAu&+NvXeY}~|Sby!EF;(-K#Vih4
zCzgL*`%XT*kiEi&<;>?JRk0FXdrAz>w+E~{y>vynRz**LxzZgTwWW=fXU(*)t#kkN
zJRxrK*`FR#BD$3ae=XefJ<4F}_2h%C331!Y8oLjqaj6}T@jrh4`b3UH`xl?xoDg?j
z=%Dt&J^ybUEN}YxdyAa#z5Vl;f(0Gcm^Xa=YbsQkQOD3)?H3?8k;{**_|KkaKNqW>
zkl{SXSo~*?UuctgQ{_KbC!1~88`PULXJ0SsJMOhm?t1yf4Gd=t{TPMo<Cd>I&KPO?
z#UjbSp)#+R>GQr5)vWdTX<ViM@)rd%&Uk*bz~Fr7K7*AEvo)_&H9nKmm|y!x?hpS4
z<~8g8n=7q*>EEGqg|lV$f$zUBH47eV*j<y!32s1Gh;V$6+QaZi_`~-Dd)CeFe7(^9
zz@B?Qqvy|f?G*!SWeBkF88W50-cpqFT7NhGT6Qn9_MDD2$p=Cvn9S<>-}*zTfN6%;
zy8yXO-xl3F`hdwGH7N0e<qqjTCNsLEKQ~*SzI;$K!7kx<wM9MOPZ@Ukpq(pziNtWK
z?PXdf<ZN$Y^xo`4{=tRMHuHA!f4O{6MtLgR&IB!^pAKt&J6~DgQ2U}NyMou~#KfQ-
z5*y_8jLx(2GY2PL;ELnECgH9me3pMf!LwtJW%!>(oS4;N{88;&$b*kH#+U0SM({kd
zoXUBR@gAqzUAAxKdwQMj&-ZwC>hnjg0;UQr3+6qtdw6q}-7DZTvlD0e%j0wOX~qYO
z0>cXB8T%!*zkaL!sU>xXL9OlC{g%+aCH(v>Yp%=MX3aCV5kI6lZ}+v&)1O=3^jbK_
z{^Eq|(e|+t67~E~60XdilaPB@_i=5}&ZQHjGvkgk3Ep1A5g^fNvwDTcxy~{U#dfFA
z6)t^huD^E_)x2*rbcj>0tXKc;w(!jSPuG51%-9p~n$>DK^V7}#ulE+WWGb{S+rj(s
z<FYS$M}9^N3xO)+2LZ|lo*#*Cs%MpBd-j>}`jMFL(d!FO2!SQR?ZU>!t^*I3`1G?l
z+ug|Jx-h{@tY}9_3tNiuN5|e%W!j|*%`!bvfjdt5w>ordX<7UYTu@USd+N`TIf4pu
zU27P8l~3`BGvDRZ{deM@M$pWEDXo*-vu!uoKXS61(KI1B_U(GbPfnI%6VFZ6muJ#E
zc;+x?TXFv5#Y%ng8WXA)C+rvhH?3LoOVwoEr{L;{gUOwbspg^SRr9&594mXYn4UNO
zEQ-3UBs(zxl)M@ln;5QdF)>g?F1r#t8vYu{Z03mssX)|OFI*qg`P}{U^`OTRf7aQe
zFS94?lhsk-;DC3=AQKKv4p#&OQmd{k5OIXMps7KjV2a{~<0kiuf@2o_?Pl1&%1Z7J
z&y7E+3b98&WbdDSVDrI0Ox_+S)<ME@g<`>y-IbkE0ZJ&c+)VavQTbW7Gc_lt{8!o_
z_GtaL(upk>7?;hjLu#Fat6Gpbzjzq~mwlJsq7;F!qK}EC>bCS#jfMvk{0%4B7V!OC
zkjj8?Flg!o<nAAe3VCsS`;*m_#X!*yG7X$@;btTAEKYEA82wdX*1L=7p+W-%lai?e
zDgI!cW|V~JP+&C{+yv}t(74QR^x)E5#4rj*$U`kT@UR`Em`0Aa;NP$K-R@E4=X>S5
z)U1uF)@z!A!@xqIvFPfZ9e?$om&rsZSz-hwG(<s^yEc>i5~I6zcMT6`ZeDK_#lPxI
zw15E1ueJjm{;&SuUUy}3?PR63h~^SzsQ0O`965V>Vu?wr9zKhpa(|l|G%o*U4HMMD
zOs&u`gHr`6iVCN`D8I}&6$ADbt{8#ji(jmul*PS9DFl(SaA|ws;p8yorTe9eQ+DEt
zIk+nnIKFT*E%JFU9^txzhzOt9(x7p<f5~FaVnM>G^o2N!%glTF8@d9BNO#JdjDgGK
zJ>7$!DG?643;G-uzg<^xAi6G)P-1DcQ!Mz^`#$YvQquD;o*gFOy!1do`N6waw>CWa
z?{~kpqh0YXH>@NHbaa^VGO#hQ1wEWDx*iY~Hols+eZN@jN+n-Vm^U_eH|UyubNlk+
zdh_koU({f({KC)l=-JZx0@0PwM1YTS7hqcCW8cKL%*PdECe+n9s0H-`0_OUEn^${a
z_!F)}L53wK>(ts#EqzPRt*a`QT0NS2S919qz4ljS7E_!nd6n+!O)t{iUSt*HRe5Xb
z<;$z?7|O2Nw$r!vWBarMt0}vnj(rG9iyuBrx&`G!DR)bz_xd5{g2h+8TP(via~~vy
zI6UFFaixj1Y%Axhk5?A*iB5kOkZW3&dcJ7YjCVSB5;qAhbDlVlYiZ@CO_oWaYplZ*
z55KsS_d3My%BKpe3D?)yWUb${b5`iG&2GN`B740`%U3VS*`B3Uy!za<ysf_!HSPwU
zK0UENDvApdOAkJ2W0w7j$_6{DZ~gRITWk4*EhrEcmIlfN2M_6lzLr{9l)HH9aWhV<
zcNet`jrkM%K3Po8C|)<^^0GT}QHz>$&7Kx2yxLfK)WRm9*V(e~|BBX}?Si-7+|@eG
zcDkf|<Exxos~-iIt@^Qr_tlx**H%P4kNf)TZRxYbAKmZ&9=$UAja+ooexsAZEG&(S
zr5R_R-KO<;ovC+!o=8Ed-7UrmKj8@s8jcSY8rVBEuWt8^pZfQ=Y3jo*lVJhxZpg$f
zc5KzAYZjuDXDo@b`ka_zCo}D-azn?ng8?1<PF98XQ+U2wPI>lw;zF1E)000>HQum0
zdO=+JY88#!i4))DHJ#fVwsp7Am6zqmwXV+-`|9YoI*9x3Z<aUR3lp~;&vTB3)KlNw
z4?Hbf{$_r0$ugIjbtQ7;Gq*w_6Iv2DEa9lIF{@rvZ8ytaY9&XeIwW})h$tF--fft*
z{;!1SY>Vj1JJYqo&y+;Wb<WHyKJelW<LRQ74OZocx$XwcbMN52y=x)I_M68X8ttS`
zeR(|L^4eWC1-{QR1Ge>T^?JHzg<nAa>KUz}ubNvQZPIaCu@p2=FS8d^IlZv`W>kz)
zNHlT?ez=jd`Bc9A-i#0Lw}rcFIL(6AbuuqGbRr%Em?pPtJ+<JDde=GQU4+I?b&cJ+
z+=ah#KTg`F^S)bYue+G?2hPp@VQb272UW|6{G2ROvtxzCVWqm=R>=>IzP>voWOm`K
zstBYM<LGP9xZM0$c_}o$VAKIdjvG_X$|d_-v9tQi2WUvbBpRYRc->YDT~RpB646yJ
ze(G)Dg~;>^e}68#7+d}5!gr4qXWf<NFlPVwy6SZ7Qn&jtE4Te;m7d83^+8J$gZX9w
zL`x2un;Q!RH{5=9MR(3Lxjf^Q`mh*KD412e;neAw?Al-UJ6GCDa(&s-Xq@e@cjwd`
z7S{jsuXZ>VU*+81DgUIgxFAKPsX@WufS^F?F9v%DBOBB(ZR}*o;hq~CJ8v`lvll5L
zP|viBFh-XP-C335oH&==wMyogQNtJg4;R<vaQ--<<-c?H=Bd_(&n6_P&IPCH0trQf
z2EAwPf>}l(D5f(yGH~2@`|8N5R^iF4JTR#QdzP3iE;Et6EHb;}|223@NYDH9ue7zO
z%;BLvG+PUBF)i}pXH{LP4Dm6j;s!hIha<zj%gJx=8m9lyoTLLWv2n2^<L=w6tH1bt
zIH1~hHtlslpT+kBvP*h=KxIH<V{3!PW%g?yCtO4=(Ph#&9K;RRKXG>!=G&Zjx&T_`
z{}N@a=00-WDf;cJ>v_MF?kw9i<#zC5tKGYQDY@MOmp=t2pptR%%Z$@8P(9FEcTs<X
zud0eFua)VqCf7xfymTP3+adk;@>Q3=&Io#U_EfTT$EjCgKZ@=zxHFf3WyJG&hsDw*
zcYu|$OENxpeCNMQ7?S=hHVHI1K4aP)wZUQKD_^tF-rlKJSw9Z02)gTQ78;P?(rdre
zR8G0)^rsa+4$4gpdKW04<8trwugyXEKlh0Vfem@E!SO)q=Q;Zd4<0F3x7)w*^9IQ+
zJkZ9%g8u>qVOv&SbDVl==7yk~HhdaaiXVPb73Ipw`B$mYIqlQDtfzsZ9QL8PzE@ve
zp359~vtGk-(mt?<IXW6NE;BnRvx1|9gUL~FhWYuGyPhvu;qqG`bhD-~KZmE*?a51X
zUUpWv9dPq*mN|7z_W9I@FTxJ9KQs;sUtY9A?5S3Mm|S_YRq~&>r#tuh?0gh;g`N9$
z^EdTYu%SE&EN@Otn^WEMScT`~UPc=q$ywk)YC7QPykJH2)U6>aE*`F5W#f5WYxi%S
zAL{mE+0Ujlo3UiOo5U7YL`}K<jZf!UxR(3%VuOxG>skC;3cI(u&cFQJZeMAgYjOCm
z_va<Qos&!s`u;6)fxkoj$=+)}*fz-=)Z4Q7^P5?-7Jk=VZBbd{S95d!I+?GI8Ee_f
z)+!hJwKEk4E|cfFveX}(@c7JG^de@gcy;S>NO{eyuJ=A+deMuDR#!w^Pg#*yy)9^6
z$<vS%o0lfH%JFMYmzZ+!)iJiERwthZ<+1C;edt^ry>;OpSIf7)C)R~)Eff0o#`o2g
zc!!hgqC&SX53RI$d+OfR6?Xq>CvUI)u>b1zUhj=ZReSj3yuDZ%9Upi&DxAK!_1v}s
zhNy<>H|dpPEBKd{$!mI?wt=QZCAR~n=Fz=jkzTs@*W`S@*REP&dF`O~)Vs!lTwe_%
z_|ud3_G~o{yZF~NWZvzN?TM{<&Hu}G-{RAMIeW_Czmhde#aCb2dN%8c=i|_Wdt?oA
zGW)(6OcOUvU89pZ(_!aaz49wMpMHDiZFz1l@RYM`q5Rb4*RT0#u6r*tN9uWlx60CK
zIoUEL@7qN;f1WhA*FP?Bhfawn^YUe-e$449?bkhiy<b#Y@latgv<<`2)4;3sj_GNK
zETi`{%ZT^>$0yb6tV&NRo!5JG=k$^jJcloKmw$by>(b^M{$c9wncs5nJ}fkUxG+h7
z((KSRwaRjOT_HYayp}#&+q%?eeU@f<j!)m^wB3iyg+K!x4>;TxY;k;hKxE^}+Xs|l
zUS4{A%zx%;GoCetCo`bI#mC5GxUu<TsM@tNpSL*&?hQJ-zq0q0k7s-1$(yI%Sul4$
z{up7_-yrCz^eDvS_J+7^%I8m;@#ksBXe*msyH@usL1+EO6Wc0nR<wyYNr!xsn|+%j
z@nVyW#HJf&i%s9neYbSg>X*7-72hq%TI7Cq-rO}g?=)X*7J3>OzHI-Fjdd#SbA__=
zEz@(fes^$voqlH@-?~K`wZAQM-R+<Ew&=J`P3FN#ikTZ_Yo|)S{WdX4JR`61(_-7a
z)qBo=-?C(#PNtm5>kYflOU@3H<9Hoze)YMYexB5wD-ZkT&Y2!j`|rn=O%A1JYacxA
z&#kS@X8i1@<8kZght0OS%?BP&F0fOUoMcq{^4zjB)zhb)Kj&4l&8F7%Vf&Qg6({b`
z;kz2a_r1<_X59{XN6G!*{P(x<z)^dj@ZOWg8$QhA*8Z}21CRWwH)T8W3lB@(m1|}-
z*FC>+mwDKSK#4|MNhhHLl@)9jjDO1xw=MIXTXae;)jE^&SJZl;-S-04-)HvzwxU$3
zwXmB5oSk$yZoEmqZIh>ASEJ9Zdvp@C9=7ljcoFs^>r$EKzN<PR#`S+DR4v}&w{m~T
z>PqhTkh`+^E5gc|!Y*rSnWphv+rRjsnxXuucQ$wIKPB$9<LLS#^5pXIDxoJwWIARo
zbU8kK{yVWXQ?nNawEq^X<~7sPYTi^gebdedei!!zZ!OIY{<g#I>N=~?uD!9fpO%aD
z9BkbBjbqJ=^~a*>t!}Mvtoaepucv$eZj47~Fq`_jO}7`9$G_3vSv&Xd>gZRW@87>^
zGPUU5y?Yy8&Nkj${yj8p%We0r*abUTPxXJ>arV=zTf63n+iv}<wP(wLZjGgOT3Jy)
ze>5HHKT~?IRrII7*z`HEcP`9a{`vpbbzy(cPug_t{+y}%cWh7o{Hg8NpKW>k^Zl<r
zoorm5CHb(;zb>w}=bzQ$9qVeZe(3q}tsvE9b={U{B5Oa&oGWsfS!W?{E4dpSUxDrk
z3uDi|I}`SG<K&GD%Z+Ne#GWm_$Sd}x<gU|Z35^TPk;h$Z<PJY-`>vv%BHb-k!n|9?
zUq+5;l|b<-vGlc8i4VnB?OhYKwvL0J53-hy$A%?Ac!sU^Ql+n(pZ#G>x3vFoVUi%Y
z6;r^XoDjZogTVQQP1Rjqo->#F$!(8T{ad!aYOyfa`}?M=S^pQV&|UpN&XBP?yPW;n
z?yCU{4o^4|$0e?lv3$}-F+V3e(Y7ry?MHWQlz1K`V|OmI_g$gp?zOBxc~`iHck*7?
z9j<U^@tp|k<Ew%~<-(VAiCuD%`m0m0;_~}*fpcS>r`$D_DpGSSk(#r?CDxq9{^iEo
zS-093#h2b%aiwGG#GL$P|0ZR>Q@&Kubh%HTagRv)eevs+e+A=yec5JQey!};ox)oW
zzKi>u)8W469>0OxWV^%9gU2pj-?V$$6#uJTC8vr$Tk{C-ocQtozum7tZPVMcVUxh^
zzi)akbwA_tuPlivv(Y{C_mXt<-uTc1OHbch`O{Nzr`%U{jdSIGH$Kkx?L2Om)IMR6
z>2ha_R>sSvSM+Bvr==MQfYUxp;|IYRm*y;ZBy{fBq!h)lmmB&v9{s9u;qdfp+r^G8
zU%SmQLTYY*;hE!3|Fd*{)vuk?9M|;V)IQGX*TA*q7kQ@QV6RQCGw;>^=t=>b$fwM*
zfMa`n&qu!?zTfSu_Wrds-Ta7mt5ke~UA?8~`haJ;Hx{Y+-C5U`ek<g!-tMfz197rx
zckbo}2poE9b-2ZCcai$R4M*QyeYj<}RcQKZ@1x6hO%*;>k*>W@+V{(fU1I%SMLR>k
zdTjUVE?#~0z0=Eod;AY_FP=Fw^@r4!*C8vO?Pv4;QTU>sqvTEYgH*Ms`Ln;t%zA$P
zDqnlJd;8&KG2y<abzW({NM5r4$+yA?x6j?`OQx?~oe|5l-E6M7?-`C&H~QS_r)*rd
zrZgqF-8wb8l>P5}_P3k$Z?MjaUhpU5?0K!_I}bmt$aKD|lfBRK?!~olLYAl>xp8~l
zoJl($X-_sP58GaH^zWm$myErob{FlwtHmbv%D8!FY-H#qqb{R<>1izAJtkh}_dgN+
ze#6Z>C5DqH3KV-xeR*+K@RJM~OI_EQbx&lxXLf?~P_1*qy+ce-i_W){o_-q;$A9e5
zt@F>9)V@~wbWM(3CvVe+%$NNMlOKnv|9=$x^<nHEnNaX}VIyO^Lz++Yoh?hcu9Pz8
zr5GuIBZj51xvL>jzgu57HzewI-*NHz%^_a*R7#KOZ_}K;SxD@~)m57p%iNp(`pU7{
z3%|U(nR3F)$0Ld%wW95M_^DUzD-@P9*)#~rbbY>PYMAwHhR@Vp0{d=1+!VNTbIA3C
z9g{!$IM`24_&TrEBF4gL%kk8nE4MP^FTKynH<>bV@#aU~n<pnP`WBj5fAa0=n9|!%
zEZSf1`+1n{jj<Or!AnifX#cLac1Fi^_LD0Mzdbv1WJcSfU3`DGeOulVvHSO-<V#`y
zo-8?-t^P#y%Nte$8{217jpgqClHK!U<NFmaPu$#pb>Ed;o2E_adX`r+bM2Ks8(;4J
zHnl~5@z>wY;=K`~b*Eo!>v%S2k@cp-!Lu|j^PhjLwirAhqwtGkN33K-i1;BR)_=0h
zYfdk|pZ!88x2;HiwuPq78QY~C>~EY*zj%83f)-&02{x2goty0b>Pbo*I0tYrJydd7
z{V?-Grk&`oPd{(;zev{DZmzU1^5~80D>cuWiGJS9H+9p6rLz|N@?)HStwQ`sVWgaB
z{I>s(YMzz9{N}pyTg~iOD>kk)?AV@q^TM>a8*f)FGjdZsuD;6Uc)`<G^VpvLEHv<u
z<UHPKtSC9V?_0;`$hFd!7JX~aGrKlzd*6}+JC!GWFaG%H+O}VBP91yKqP;wITEb_K
z^P97?+SY01o|K*^c)jB2lb*W=&DwjG-FfwF-HhuG9@oT#)jYn_G-cQE@bLa0YnGVi
zpIPwZ*8B5k@1=eF$QOV8<NEx}6aP2uvdyjut9dJ#eCgEjwBy^O7$5(b7ag9yW82Q6
zEVq@BY_H9>?Uu>U$vhDqd1sEfiSDw@e>-|?k3V56Gu^a&q4B%3Yk8Zmhi~CM<=dCp
zcxKVWnICTiglTHImtQYe-CFvvm1D(<eag8PmsK6M|54~XIqP#!venG>ni6Ge4u4v@
zd)DR+iD`A-%dMZ?Sm-%5e&(Eg?&lRR++lheoK{$y{%4VIsq^-Cx#v%K6l{yPGTeJ&
z{^=?Ael61pa-V+t+xk+|<A!xNmz?pRE^6AFc1J$4cEx7C-{)FyPy4j&-JDt5-ig$|
zo$-DC*R5^q|FbM#`}UFNOUAW-%dc)zxA_-&R$@nAVUqBbN5;pNyxV%}%Jay&ET7jm
zlP^6yb<+ENPWBnS+&3P3ii*NGr~a3JlW`_$%g%iXDQ+=`U4@Oj+Ls^y^>yt&=32db
zcMMX(W8I_5l6+np?&K`3GH()mX{h_}Lu&uRb0^JhIIBQIetQ{Ij)Ietj48*B+~kcX
zUq~_s_V2rIxZ(EmLrrbhbi2Qwp56j35STd{@64FiGVAto{}aFN&hll>h%gd?W^U(B
zhPx6~Ii3M&^Lbdm%N*IbY-{N%N5g{ekJ^;Z1+NUs{ZPIx>&k{TY){{{txqf8yy^Lc
z=a#W9Kkm5P_{q95SaqevQy=f3T)!X9jgxFbQ=PsEm};b3q&TYPH=obgksGq1FTK`0
z*5aGSrHNPVw^S4->xpc-VG*ji>xRMiPk&uipTEz2dCsS^*yi_v<D$i8A={+8ys!Pd
z`#ber>+<LNy9)Ex#b1}YIOX?`^lJCJzY~vE9)EJ={j53D1HR4If8}@Ku}xmaqvVhc
z!6$$BKJWI6+q3jxMC`E-;_s(9&YZpR&q8l*JHzMe?_Hi&_IdTuaD)6BzdwhK+f+<n
zeL6b(b@BS@%<oHe_xT#ntpCm~`f3uVq_*5oFUGiaZ~Xb!uaWzCy8Qc``!7Eoc{Rzt
zKIZ1vzdLun|MzFh!P)ETcm8;zTYcO8#U^pHD`!=W|9*WL=`Ft_;Ya$eo<xH;I~G1q
z^?z@XAL|<3U-WC^)#|H%<wK{~)L(t!8uEq>R3p@{TPgYNv$^+q^Yh==Yp=Sqeg35H
zzjW34@2xs6=5>G9yZLtOWa6(s^YZT5xcy#X)IIU%HewMo*WHeA-gCe(PT{Ck&vVc0
zjT8T^II+9%-IIfzt~C!IufG&OJ-lz%>%X6hzkaNgspvh~rCs;_)<=C6N6U4$Uwr0S
zS0B|?H|<~f-1^Il?jC!uaPnovJNebC($?wbKh`#Ee)n_1p1%`MbIqG?ueaChx#6F~
z?N1-Siw~=u_U>}bu?k1GJM*rZ`&XQsb-qNS_|tX2Pk)pb#3%gFpL!=w&?5Jj!q>m2
zxvsJn_lh^{y*8bz$-zX>!?$Raw4U6`yj@EIo~9Pns;F>$VP!No(|@E?*D(2bF#F}s
zU5s{KlJmgHG$6X-oTvn}i^J0RSKz}tZWh(;>bySTLN}J5FlJ)Y{qw7+XVXsg2iX@r
zJN#RJAI&sSn#tJu!{?1W%eI5zCK7z{0_OtCzVrOBbC=C2$|_#Wdn>5z(Dh|rjtlcn
zeABu$waHst<1;_g!9&*qCZwxczIxW<vFiKvr@HIK<5!5jl$f$`!;;^;IeL|HYZe>-
z-TIP$XTf(vmaY<K&b7aL3M-~AoNl}MvCn2DS*{y?vF%;2^X>M3cA3j3rg!t2#f#he
zBC&o)jh?-doAYQ*U->Pb(~qZ{T|K+=>D45e^R+J7e{bK{+8d$mJNLta^Oy9ttF!4V
zoLkgd)^qq-^~O!>=k}={k%<uF+5K7We%;2fCB-#2bS?*2<gb4z@$I&1>W@u|Hj7NZ
z&CQ=W@AOZh&4<qH2-27lT~QN0WecCFH}}^3eK#szecrqI=C?n|m+bnF{I4)i3E5im
zSzB=G#0Lks*D$Z#clPO>n7ej8s>}O)ZoYS&u<oq5EBAfv-3`avD_;J|HI+TQzAEpy
zq*X3owbo2y!T&F3t3IDC`cfzF{kJ3czj%Fgm07<`ba(WnhmX?iem~hcdFBg?2j#7w
zQ{tKL3M6V28--5K;Na*;t>RE%UDc%Pc*S>mUyx>}-`2fbBKPjvqZAN&Qk>-%Yol@c
z(-q!+hi-9ismk8FYQ@rnO%oIZ6q(A>b*H@CeQ}rY>6;Nh%48yrsBV)@4*nL@Qq6Yl
zL5c~ur2FBT;B)o7THm>y>5t3yFP)?b8*CF`G{1K5)Vy9(3(@<@8~NIU&O8p3yJdfA
zN=)=|Bi>hmGacW)Y;2lScYbgE3O{GnH|8!@F0U(dN+M3YOXu~xb@GCI=KkyT7Uefq
zFYotT{p;!OrI+VN23H%29^+rXYF>?3;;EqQ-Tx-67QLq4@AL1`F3X$Q({u8rtf~TX
zPJddotcSnc)c={p3ZA+@+RE|sS^nS7&s%wZ=fkIux}~gQR_5nq&S4DKna@`2d-&V;
z+K@t--;ML%FH2viQ8O>>eMrH)|4*)Q?_ZN8&)NEG&HcH?Uq0`UcwhMM^{G{lD=w_;
z*!TZVqJC{|RhZ`S&x@~npIg!-Epan`U)q&y(cX1Ce)i|@+_m!Z+&^oR*>!(j>v^;G
z-p6Bd51P~j%5Iz~-mtA;zO7E?<>cwTHTtUeF8;m!{j{3xr_8S_W%bH$R2`mfevSX>
zjHEZ)*12BqseE-K(UUiKZr<-h-@o}R-7VfXQN-{5|Lu#aN>Yo<!{!{mR``77&+NTI
zhkkRv{$rEpb1KaE(`2tdTmO9*-+Fk5qh_7eJMHDEqCZz3wh=eGZkt{5>UsU{B?&ig
zZJM^_{l1uzK;Lfd$B*xx6)y18dHZ(Gg`(R1&C8<wPgTGFq&f5G^?eZ~uQn}ucJk}n
zbyw1_U*X%I_H04YwY~2TZ;HHgL0q3v>fWz!H`3h>-aUUmJVfl?$FiHMI{P1Q`4hh0
zUhnq%D&s4bfqelP94&trbNIfw7?|yqIBTKD^PKB%-~6m{f5^1rQBtkB<&=kFD`rJ8
z?=z7pNVs^*Qex}Y-67)ZWZZ=>zM9?j`kUYMs6C5<wlXv~I;i_SisBY8lIiDNR{SBL
zzc)UoM{d@dYkAoLtIfI@KQ6SBQ~ZAa+6Fhvoa#`YPZ?{Do_F4TWWsH^DP4a=&Tow<
zEwSglKjD$t!vzU?b2*sW^_fDs)z#Rg|6Y}z^;kNzX;<nw^^>2)eBNv{;l8$P`5D)@
zZW&Ts)jiGEUjDy*IZO$h;z1+8qKuEHB}McGbG{30smhOThn)qmxzGH5RCRon{dC=Z
z>0d77`TtIFvD~(Gm+i(S3yr5IotK;}7BJJ#^>r{e*L?ex21}=v{{P-2IDeg1p5}tm
z_v%N_-cirg{A~Jlddw+#*3_S$!+Wcab9Q^az5c7QzkB)O6JpP|l<uiIJ7fA5{!hgZ
zsviIQY_q=h`O!t|C8s?)t*y>-yQIXswE9x|Qr*Wl&VSupY_~*d?#C}I?%(Cl$?kiy
zM}1peH&>{Z%}R@<m*#MlUHf@$jidKUkMucjr^Q(R+kbxTYRPlH_LZ|PSV_MR>|vfS
z`1}pWlX_c;GXC1Avz4n%G=9bJs`@fJYGM8No6h@ZgwKeYH~Gnnz`D(G8e2B6lZ}tn
zV150PP3w~6^7s$cy0bU8UH-m!e%0Db=jCSaG7E7v{cU>ret4K<@lpRH#Xq0)$UZZY
zZPzb;>?%+*eLDN6=ySW&MOD`Szkc0q@A)5l0#}-H`%8Z^i~SSukoW)Z#qVr#Jv`2@
z{`2aAzIac?(?6Uqr%w6$VJYi-|Eh$;6<S~H^|R0JOj~9&<>k?@vB%n0i1+!&JG)xF
zjx$#;f8IPlB=j(Q&9~)C&;Hy#XGZoOpTwCno#sy7*?Xi}Ss*4`YR}ui_^LxG>y;)u
z)a==J?0_<F*NgNzLvF6GwcqW<wv_I;vaTy^>*_VGwJp2m=UP7K=2`rA;lJXyKUrUW
z3T|hSJ@#5#>k9Mv-+SM&<d$2A$CaHDDE-*k+^ZRKwS3jmZ2rCKJzt(km@Yi9JN?+N
zSl(4Da})M#czAlo{8-jxt~*!L<tHni`f~W?k^f7!1fP|O<Ie9|y|LhVYl6n1l8Eg~
z+Ey69T7T0lS+@K92OHH7+CMFMig#SFIB;vxek<{vkaZppe!e?1BWn8Ve+kO_tDpQ<
zHsAJaMdi!T?^ZE}wq|t)Z@C1e1%K;IwV!KY^Z#D&vBLo>MoW6Zm3&i!!UcDh9~(NV
z=bl}9XCBL~b>GDuZ+fRL=?%L&-S?b{$iaWVcBKUU@>=V0{i#R(a-+FnJ4-|-&6QzJ
zmeT(^Jw_{Rp+%gN>#mt*VcYnUrKAh<u6C8g)}36^^Xc|Iot5izOPA)~zMZFGSvu47
zPeFPx-`&|4N>80F{vKX-PI6v|f2;A+H)~#dq~5-oy0&c*=hwUUXV2X{zi7(79s6r@
z$|i2@zjx{22JMya4UXBTyL<{;>v8?lIho0~HYYAS$E|wCzw_wl*+HRa@6P-m7jFHe
zYvs(k{aSom^Yi$+#Z4*$Y|8IeR{eVwa4cc%;prk9&dHyz(fDEhdb-rLs_avH;zQ5K
z$^BT(HSc!6zD7h+UbUs!oV{;*CvDr7=YQzgs=l1!$PGW1PBse5aJ72<agO{vvGCT<
zi5|Y$<^6@xYx56ZnKNa^-+9gMX-C4RnwzF47e^m{bT2-B;hB^1)+f0^YhIX>FK3^;
z(-~(T_w{US)?>Z(sa0WFZy38uxV$=wS8yr^@z}lpVe|FSCEHgIua{|yuUl2C`0La;
zzy8qG`_=@nKb*8{y-CW1je;cshMhCKRZMmogvOQ??R#}>&i_w`_^w_sGMKWzO7GwX
zpRW7n=~0dREVGOi5ARSe){$Rt9`}sDB~f#pe8`f?zpComtrj-c<+WP%JICdh%nP;*
zwz|sB{XSIv{>RQu*5?k*I2U5}+h>j8_rDx`zZ`6>AJ#qN*7+kJUiQ5Gl%)LnpietJ
z9nL@cwSeE#3OuZ;z{Bz=dCup+zw(JD6IZ!2Fn#<H^mP}vyVQBc^?Lf@(q*QyjGLuy
z1Ph#D|Mqd>MOi17oL9v)M*F#cebN1JxQ^MTBjwBWUv-&*!c$*nIpzJy`|Q?p?Pj`~
z1@G&q;(u%YELnAY%JqV^E7lnccDWv2HM8`WcV5z(y;tX7+Oh6#YQ#F<v)*4`pWkkv
zyl>IAt*WxKS61oXIU{U%wELBH)v5izgD!S1Q&PTl>!sNg1JT~tW8$SpdFQ(Qy<*{B
zY;Zbb^_+y42{#g*{TJ{5c=My6=DELLt?P55Zr*)Az5HU)|7Whd#rJI6`S0`nIWzyi
z^qB9uHBUI}e*NzyReZM(1l8@`cIuju6X(n7KkVzL&EfU!`xx{%d^vyQyY;fQw?Dh`
zoz|N>LGZ_FuAI26uQ$(4H*~(7wECwJN7s8f)gAlm&Q$)ZW7jM!d%XSPotLj}O6cp}
zyLf2h!_~a|p8O0ty&~(S(J^Za%WJErTw&d3aK$ommggq$iWhE&#;)T$zCo6f_dZu0
zT^9ACf^SV^==Fuu4szKtiyc|quCj9B%?iez>)k3*R{bZuL$=$CT$(P=&h`E297FAB
z)uYzD;YJq&md&=V`ueztb${G59p!y#(J5hHlOKF@j{a=_+wWvU@G|zpMK5>lJ!Y8G
zT0MR4k6j507dY+2^<_?f;x%PX^f)H*V*B6XANk9(=kQ;bom0sBr8xV5%$$2{lcw8w
z-i&>G<w)!Q(2cpTLzb^)$#;tjl)EmmqyFGAw+$iX+BqF5(yjfRI~{J^uUTyVbMMsZ
z4|7+Yz03Fd^Ie0ql`YjLdUxztdq~XU_Jv^G9ZplPS%lQx$_)7M`O%lGd?jVg_&q!A
z#Jl$kf6TmpuRbnvhl!A6<E@&vJ7qtYT`x_$TY2k2>b1K!nYT`?(PwIkeUaXjAZiuH
zzFJK$;_|~XHU0F5Zu|M4uxHC$I)BjhL50M<3mt7+cz<hm8twi6QT)tr`Q(=m4O|Z-
z-)=4a%l7o$tz{uTrq}KSEV#8OPp7$8^6hM`+wKBOgSr0X-;>}uxTT`Hf1{6Yc6h(i
z#{Ey&q`22UvfgFu_J8L|zx`YdOCsK<Ezy;YlfToLGS$3t>f;NNmS<+qjCy_hO<cL|
z)84qJ8D49`7iO&$&$E&!o1<>Nd&!wOOPJ&HT5eBYXK!V_@9VOiH{aWS2)r5@pegye
zuDts3mF%<>-R9NehwoResNKI|pIN<o?(u-AZ}%&s?xpYQFE|t(uRqt-JAd2GB(G;h
zb%GDp|Lvarsz==M<;T8bmya!Jd)udAcr0%RZ_eAIWd4>DoYD`ZrY%p)y|csd<I~-*
zt=ZS><x2kj&3~#p-@-2aw!Hl+34i%tZ!%v!VA+3dYwg_NCz~xcMRyt59(ufcztZLX
zp&zQ}hp#>^=B)liIm-P1o9yJPGV@)3uPKy#b>rj9SC`M$)gRv#`QgH!*IaSNZ!YcM
zx5>Noci^&iT@UjGpMNy^IJRvs3ZARPFS6+B*`AFrj}`3rsT+HqN9ET`do#H^4|%>6
zzkaYturi-d+HQMce^6#@^VM&syh?MtH?Eq;(OOk)x@~`<<HlHS)oskDZFKhJZb|5H
zv|(E)%w44S|Bc4?)-OLj%wFFvT)@+68S+VP1FssBoOP^riF4q4ryH#sBulp`?_)pp
zW$TYUiI(@ArX7gd=j<CJu=tbOGtKJog`0N&&fOmS<K65g**n}HM14c#4a`EesP5mo
zRWR(I*VV1N!i;P#z5FZN8~ME^z);=#=FLCnwUz1(3r&*5U+#R!zxs8Y&aBOw*A-tV
z-ORprUrA6}<&sTHW90=7s7LqZ9k%-W(avMyvUW+PTOBQj?{Tayk&<fhioGirdRgM`
z*VwwSJqz!YTjjjE@9h<u8{B_R>q_F+E8WZ1$Sn-)m|+;-ytlfAh0*aqd&BbAyb>pL
zLVZ7mYF)LJ&;H;Tk=_2LODRNV;SZTtw=Y~SetKn1!?eAPPaGfiJZINeJz`O1m-=3R
z=Pv#14))3O3TGToS$llTWxnf{jwgTHur$8=`S?=U@6}x2Ro}htJ8h<;8+#`}j3w0P
zyu_caoTmB`Wfoz!{MmXob6d5Hr&N8A+*j!2=k)gO`|vqIEUsqfyWQ>XR$Iya%lTp`
zpLM_6>}r2feZaEnb=50p#QS|0=$X=EeobE7yx;m^e*CjJc3P#k^Uq0ra`k6*$<X-v
z=GH9X8Rk~n>z*`nx2-jrKKrqD_g=lYWzX5mJ|?Zp?|q$o>~W^<>+Gtkx}=}CZ%d2b
zJ#hy#_>_>fp(lCTe5v5dzaRgZ%fEd+=a<zx<*j=)E-Tw?s#@q4|6+bFr|-#AzS{L)
ze@uUMP5ZdS?8r^8^Uui_HW<x4_WJQdp=Ao@fvQf7=UcyoY505(d4KzRW7x-#bLy%U
z{E0zL&b=~k{!VI1UR1haS-_K*yB@YYZc~+(UVDk@{vG*L9<T1up9PYKH;PQ%zG|T*
zXItvNY3H|Gi!F|udn2~-onywq`)$5<+ox{RI5nSR{)4x%d0Uf`bI&);-*;$vk5_2(
zgP`_2m$L9}EA`9GSMpuqcB!e5G`Dfx`1al}zN1G=wU4=Rv3l=%`S8XK`QM+G?_Sze
zt#d5K!ot(^aJ^-Jq-W^t`@57shB|T8biZdZyEA86g{-w5bE$1saNuMux7`Uxc5S@z
zV%nF9N44&1RU637N$d4CkL$_N$qG4o!25E?9hQWdX;X3=Rx9;g_p(_ZBs4ec_&l-X
z$5)+}sEaI-N{;RK_5a&jYVkzq^`}{7p|Y&5^^6~s{$<+K@y@lV`?^XnuPNz+fyAku
zsdFbUVODssk-zQHFA>q#mXW;jpR_||H;a1Z?3oxJu|{o`(N(6i6^>JbuRga3&HA1Y
zu{u2|G`K_lbHJVBI~DvJrH|?4v`uDlt+M&_@bmKLYi-02-<{CFJ##s0>JOf?ml>8;
zh)!czTJSiCukz`tO{+z?|Kv@Ld><5lZ0D74IyNu$OIFXk?;GMd^YMj$*DpN$%(Usg
zlIDS)L+75UH_U&ZX)2bZF1jROrMziX+S_{<MW#L2vNhIB^YHrn2Ue7tIk2p~d3n;;
zoAai9<TPwcJ$uQc{i^tze;*Qm?wD?N&3{(-j6HwfZEj!tSmK*W$+3sVw=XBG6#o4p
zoa<iSlj4gW&n~fLuXVRut(TMWzE5_J&~2#`&!0}8`quM&ZD4Rj%>9Ye#NOx3pEy6F
z^y!u}lOozby1HFh_cyWCzxw#$^wevoCbe7&nELX5^~M)JzKZ5L=ULgMx8D1qm2o`i
zhBi}$n%%aK-#&H+)o=g#pk~|E*Bwio>Xif@Z&~?xL!5G<tqK3^&GB62?F)sj&k?`#
zuX9n5ZB)ra&dOugH%>WjdOv%<%fE##CaJF3bzx`9c_ys)Tx@DTQJLclC*$ng9h)q<
zL)oU(#A?Y`3$*?#)9>lKf9G1jiC2d&C21c&WM$d2%<ZbJZ!G7ovnS5%nD_0kYWVk5
z-p`YgtzL@H=6cZeOmDxQ^{d-4Zgpvkj92ilKd5=HXifX=`TG`qaL$guR#G<Ux{#cn
z-JfsSYwt%}NtE3=F<JWHC0XxK?-%Plt6m3BFTTrJ_wK#67Wcvsdug%To#)rCo$H)h
zwW4-=X(f+#{)(PCCKZ=9omISZbjifYNwOg=Uv1Qj&i?-!-mTv->GsL=4et8VJ8!&_
zjz6urF1!DJ=yk&xnLGY+{&p3(ZCYCOcUrpYF$ufJGsSg(J-;<8a!RYd*sR%BcS|@^
zG%g)aIVx&jbn}GypR?Lh#a-8a^qW{N^67ujacRB9)M}MH6{&Md;;)mwzIeU+>dJ@!
zYuPf_?;;_7?^=SUmPmctUtGudXTkgW{Kb>LPVLveazF0o$_1|jzsH&jo_aO!ch4*{
zt61BT*CozuQ{RXG`*nH46UnKEe;oEzlmGKDWU(u^$np67)urdzX2hKM_w$)%+oW0l
zj=rzXo4rwSY2+XMb&~fd|GE76_SK{x)%LY%+kX8^ZR7L)`}XPFSHHS@e{TqUzgBSS
zE5Qe4$~8yk+o-NS9@{5&sC~-)zs23ZHdXx378AX*y|%Q+D)W>=>{X8FTf46|>%R+m
zziu5L=V@b~<lPVbvw2hZM?TnoPIlqC1IO0h>h0IlUYi%Zzv|6tW|4FI_t#BYm{Z60
zF7HM9|2a2z?&=iW_c?OvFXn>BKOgn`&6sj?^4>d-^kYx|yI#FiOI$zJZ}!IX4;r3d
z49=grBK@9j;C;*W4!Sd@-+z364%5$ouh0L#dol6siTCS2_kCTty{lJTCH>5qo9pNN
zX8rSKVehKF#rIG5{LY_|bEaS4v99r_yVu5wrPZ&eF8)}3)13X(s~1Hpch9@EXwt(+
z9kb`oI(dI*s^0Vp0hTv<dlnWRdA!ZW$#Twpw;xNb9#6kLWp(xDu1x*V;>owK9NPbB
ztz6mky*Jn1`B+-O!u?&_@5)A1$BVMu=7xJLj%iN6+i9}q%H6ovc~OfdKaA%KGn~7j
z?PWmhwz8*ROtiNzX7u`@)>_&a^`GJFi9PB6_O$wiHq^h1?`zgQB(P;3>(f6fTh==9
zuD_#jxpT>jsl8u+25fkg+VGL{!9V*p50+WCYIdx?x#NKEY5zN!ckdhz@m*9W@PfPS
z{F;?>S?^D-?3G=|veqYl&rZd9*TT~*WnZs7YvVOP)7$@Q<2Peg<=5Aa9zAaN-Suwg
zu0LA>HS0}(_Z@Mb`Po`uWbUVf=Z<Oq5Pb0c#nR>1udHCVU#V4F<oGcA_>*JXU)0nm
z#&5Xv$)XQ52l7xN?p{Q^`l*s$pZiWZ>W^M*I6j|Q^ZXazNn7k>1ZH&Vb7b~7{*uU(
z&r}fRD?PB|NVz%B^2iIjK5qPd?85ch+(nw&FBTlu*pejY7Q9l6<BKp;#jMRYzuBsM
z4c1re+<H6X%0&iq{U1tK9GxXJWD1sFkMAwB-1GYVgcjCyEXrJu760u>+#CJPsc`4}
zh1b+SEuQOpSfcf8$o8A@74;{1ZG@KwJ(fNDp?%)Z%*@vXS6&BR*u)#VJm>E1ydAtJ
zz01tlCL8OVT^-S?WZ4)m;CIUVMto*s<{I<&Id2No*J@{b&YU%0BqTCFBSdlW!o}6~
zs+0Wl@^36Xc6)YqRIFt7si>dF%J-&P1bTu7fYjw4-jDndchfaw#bNsz9rIG&rSF5?
z)!%2mD3Dy2v8MUwrd7EW+)tJsyE*$9pVy7zUn{1}Gu<oQwR!cwy`N>YZeBdcZ2vN!
z<?z3qrKPMfznxewvW6R|UfXx(<+(inH`iW$bNS4zylbuI+V>_R*1uXicKXIYD$R{O
zx9OOkv*6}$?33m0hTNRh<eQy*v3_zT&o#S4Gbe9dyQEoJ;nbH0KYr-W%??Y6TGRB`
zY(q-WP2c!SQ71jQ^RzBsUa(+UQb7K_jJ#uey8T<8x7wb4wKmk?$(P&@8S555{*@%v
zXLHx=%7n0kuggqUtlW^btn`oQa$dtJcXQ4cX4QZ4zIH%%wp;+i!(BIG&ad%EH;&4W
zuYG1<nt$t`SE25tYuC5O+xq?8UmYf`FJ1C)Zvad2<8bYZ=l`q??+HA;Zv`i(_vagH
zZrtxnUe>>KtNgE%|4;KTTQU3JpQCkQ_p8o@C;I(+aVFAo)rv)-f1m$l?hTjyaV6m!
z%llo6bMj~3YOPAUv}u2p)in>XM>qBto-1&zpR{Gs_DerMy!Jf(d%OMGSDKeTt+Sk+
z-NBss=tM~A^Xtb~TFc*EZgM^M!S}WPC0`>C^93vkPKfDo_C37w`s||<<=p496<#ql
zi<$l4$;TSz*ROR8_HAN2^*TIs^V;aW0g<cI4&C)~?)6qp*}Cctr|zqtr(e!qyS_VL
zW)*9hY5pSL-hGx~smYwe)1L+8mU?O#rV3p%dH!hiiOztB^8Ky$CX5?0+!xuuc<TGL
z?Q2C`*Ji2RdxAI0i!KUf5xaAY|KA(K%-f5Vw#;vp`Tk&;_BWF<zNY%`=If3#?U=ut
zTVvn)8%vMej{f+9Lz~b2{PWmKO+#DpbB>;dwm%u!1zVZyOdkB1CF*nKQ2BSC>iYD@
zCb90(E{4AG6R+<NDmlh}U-o~O-uAZIqh6=I`!<&zx6GaUW#P<O4>B(Q6#6F1sKMU1
z{%_{P?3v%+_1yZp^!;KX&cBTf|GK7bvOo6Oi6<}jhgItSb$!jdk0)Bzu&=%HT3hRz
zvCjgdz1(h^7ggRdx-6|Ty7ul*)JA;~-^g0Y_MK}*i;HI-VQqA7YB<~HKkM!d_7(Nl
z)+U}knYv#@``hQKhx<zVd#8EX#R`1aDVlolQoq>V+C@)YeI-qG*YR-eEU2kkpY6WG
ze$L0~{o3C7c8b>{qz^OMltl(^PAp|-P3$+?y#L8Jvzr?uKc-ek&obVxk+1t_%M#VI
ze{u@$U;5NpvfWWq-2V*6u5I=^6}Fzr(=L8|wd?P{)Tfu_;sbsJZrIl^Kkv+Q?MZJN
z`?Z(PdL6v8?zVKbU){e|f%#51-zP2oAyMt_+-{v3?z=qP^MH?hYg4SFTTQ5Mw)e60
zrMk&wSGUz?n0?vt^y{OUH#-g2Zl7;easTi>F*%hd7VarWwl7tGQ+(A}M|s_IRi-yy
zz0xr&YT2vAX8x_8t=E43!;#6{LHlYxbsb85rg)dh{^gHTmnydHPDwh}m7Sjc-e`LC
z<M;1B<<05tjqpD6c-yt)S=0W6|I0ppG-CSyX!&+~oz*isZkx>uO;oYgTKfD?>aO)H
z&eJ6r(qcUFDwe)Dvflsvtts=xF0*Z&HakA-_hzs3`47*mT%T-n>(+0*1@5=E{5HG4
zb6>A~X6f0p?#c^&BMxM->WCgxiO`bjirwWF*>cu?<%^}iCwKcv+}T`tv9kCRPwFrE
z|C?p6z25kWJzg?ZL+oYN(@yhp>7p#@z5n<RPfR!2@@FQC>dLF>zwWiw>cz@cuI1=T
zT3`C>s5)<$zOMFh=LYZCm%)#U-uUl{%-K=(^Y8v`hpu|Kou3mQv1_@zpwx+Uo5G5x
zi;E8LPf>kl&Xs59F?+$ATEVv~UF4Wv|9KxD#$m2m*ky4}Vsc)M?t!XewUjR!moiVD
z(Pw%Q^z25m(G}L>&r_KFw`{55WA#4u+5Naf;QD^$wfp|_KK;JyEpMBv#nkgNH(2=}
z5q)>+a+{RcuYV~9`L^>`-fy1zr%+<H-TrfPC%+D#X;)gZ>Z8DsveV{Y>i%8i6+FMx
z+y9ig+1^y^wI0vYR&{^;;P%ue_V|Qzc4q1Jsk<g-{@N@RpZuXQ|JDnWQpIz=Lbv&w
zU(40!CHeRs?wj{+^Q!!uw?X?`Hr17<uKsxP%k{=w;e8zuKU3#+J-NeGt>B)@thF{b
zdBG03M()cqwe(i<?GZmc@2i$+t?j+2w?Uz=zpvUX_AV@6e+`?<MEj5D_t#FyDu3p1
zD8bYuS>)H_=WNa)Nyl6|WLdgrs_dB1={Rqr#mn=DH?G=}D)Lm`-sY5E+_ft`N0@Fk
zdi#C9YHZZ~tC;WNq}a>1=6a=Or51m&|H+fybN-9qZg)M|pVu}mS~jQn*s`gw>&{Eh
zkNSDN+s%eOd&auvt%7gX-@9uZod03T@kHLlmCgFOQ(rGiHnQ9Mc*~yiKMg+5FWu^#
z7a3C&D7|Lp4~GNoi*~YV*8M&GRaDznPVdD0-9=aSmzkHS#o8$D^?B}nqnzadtEro1
z@cE`C&wuOrs_=xce6S8HuB}$NYFg@4__kCq<Lkq@6K%G+&-^EH?M&^*L^&?8H}|dD
z0&e|WE$kcl?@9aY=lq`wQax%N6<%>vgsd&PU8dc>ocZ<59d(;nIdgBAP0jgoUex-_
zbGDC+*Pdk_6<@YHBs_e6{5<eN=BI~vT~F_f$gZ^J{B^zT`|qdE*R9U@{pQC*&(!|f
zZRKh@6M}Ctz3%Vbm0ExD>L=z_bGi9@(<%!;t7!SH|8>6a^t#z|e_woZr)8J<wrl+L
zzdp~Jd?U}LD(9!_WTUEUGybLcHmLquv(Pg-Z0Z)7%`Z;3JziJ4@x<S=l9BOZX~(wf
z=;xiPD$Bjw;@>s<SpJWTN9t2lj+D#~nh|8JWxQ@mo5&je>(%QvA6xS7-<mb5zvs`t
zer{=MdF}4kwrh|0_Lpx<*lU;J>iPLbwVj53bJ0~hP5qT@5@9V(vSzHY+pGLf@8FxR
z{;o-Vdo8QpKRdsc{lAu9doy>}vNgGHA{PhSAK!B*Xj$^xsw~gi=L@8tJ<;2p6_pkk
znVkN|_Wg9p8HL5?-#n3U&b%sL`&pu9-*VNjGq;+m8=c!)TC`^GWPP?7X2RF*?CvOE
zemXZg{$A99O~Txhs%r(0RNdkE@}+&d&W^0AYp-5SJ9k7q?7i*GpEpiyRSVhPwML%n
z<IDt!KZ~ZDlszrt*W1f4e*2QIzoT2u&$-K=)mQv$`7-Z*tV%Igs!q`H?gov^hi!z7
zf|9C4wrn|*muDQu)~4zDc6(w<UR(IgINtaR-|cpXu5C4%R=3df@6T4*V`-B^#9sGy
zN3@qH$1FVJcytTzzn%Ag_WW8?E7G^8cI~&L@0zuRv<uH||5w^Paj&Fa=(O66*9#Up
zc|TlzP9`g2!g95s=S#0BX&hSlM7=Oab3@a`uJkCo?CWN3tf%GgU9bQA<eA)2i|&_v
z&y_V=S2|pE`G5b*^S{DB7OKA#7w)NlD)M->w%bFa)ZM=l-o`2K%POAxD<!$tsx)fb
z!U?>r-isH`x1WA^YxlDD_}gpe{ZHukliGJ){?Ek_=fwJ3J&v#B*S|CRQ}aFKgX;^f
z)_q@IUV7RaI`!4d`R9#ve*aw5FrWSN(x#bbX5YVbNmJtEk3Y{&>bi;kN~-^%ez|(p
zd#(IH`<jBiivN43hQ6}r?l;|MwBYqZhb=6#Uwx8!8nJT0^3=(8%SvKh#EmX0<@GJy
zU+{xx_LonNQ?{g>jcYEoe>g9C*W8wc3J$E>dhIi`HeX)&<{WFf!_8j#Ilb!f-C5-i
zjc2#=Mzm~CXyp5@8@o2+(#kvQ@-z+=E~^k|dhw}My2`3x<BLP;9e#CDT+@xX^g=%*
z@}(?0P{h7)(I=Vt8!9hdP~u+~`=IRD_6O!HU+>RPwVrMCQ&YWqM@~>x0Qckq_mA)T
z)DCMsn(?OPZj4I%?SC2T?uWnm@ltumiJpbBrygl9OMACFVK(<~MZfNk;pW$+xqo~<
zzI1I>uJ_@>f4g>myYa?1-1qdYL#})E_RBmz(tJm2N0nu3<?8QS(pZC*eHYcftA2R0
z0h`UoFFl82)IRQ0Y|ww@ziLP6sYiS4YXq;P`rogbc&%Ni@nmbJmTBj9|Md@<-?u#!
z`tG*qB3p0pz8e=aB%IDjOip`q|M1og_n6tI8@n<~yqh_>$F$J7*!<hhOuGYL6@Nef
z=OVm8exfvU6#vXypK8xsP?<VWrh98;_eY*d%+0b=kw+HW?T@(AWV-*Vu8%_GwZhNc
zUhhgx<f_Fl>MDP~&ADyktB73(&U}A&J^22erQhD<9{+TJTj6`)qdMn*8d5WUPu8w%
za+ohrux)K%Ui#`!7J3J($~3N>|C=Wvbgg1%S@<@gr|<qOUzN3vWA)nkr%sofOY6lg
z<MF$5bZeM~i|bc!>DPVi-$Udl3PzXReKTj?){6R*Z(kZXJegDYGPJl|J^zSRj*{-X
z_i;gGjXSlzv-~<M>3q-lIcwMdH&YI3ulx7?oM1@H{v8Y7RczX-C-=MB;c(g{?Ydn%
zz6eEG{V#XE8?$Mh|NZ+jIr-MJsv7xA%-oZF`rbFelJcVa?V>*xg|E!_IsdES)R%w_
zzi(aXnZ!OlTr7X}ONqF9SLPNUznwb$#XgB=%Z^mYA2~a7?~%D?ySvvVzl#bgQ>@s3
zE8OpF%u3^5i}O##MP9P$X!GeYzVbR_+1Zf$*JS%kzaQ7nUvd6=yi|;x<>v6zsy&yl
z^;Kt_@yl7$_4BJn>j%BNpZDIhl%M^(Bdq41VdiW3sGa85wfgrf%I$u?C!*x&YR%+%
zSK<R&tkqY?c(S`YOnG_mqve7x6%h}z_TM;N9hf3_;l>m}?){HuWt{bW<TTrE&)Oq3
zd1;=mhUTR_Qy1y>|FzPYzruFg4vA^zk!@4Ivlj0SJGI8${?++I>E%mY!p*+8=c{Sm
z|Eu}HT>Sm+ZL>d0KKz^=pS|PiN3Ba+_nhBurco51>9z3VBN^%9GU<5X8BZ@3%?rCz
zveN1QkHZf>r6!-=w{cURrdj<i>3#QV&vUeO&w3iL?s6!fCiC&zt9-8KwQX5CIq>3n
z`BVG8-^{gDEs;LlE-}yD=J~stl;HjxyA^8>a?cW0PnYDXsk@qR_rlbF_uv2j{%zjI
zAOqbiTa9AhD_^*rD6jlSt7*q2Retk;r)ewi<b=HtP<+s6U*bCdg_dmhJI;Gc<)(A*
znO1pgW2fNMydNd}US-yQ7TnI=aG-Ir#Eu>8ljr>S$0Kf#Q}(3gQOrbMRVzE=rAzr(
zzQ)AYM6t8|)|)!z;`%S|ziw?iWUYJdV#c<xw$S2d@ph|b*=1)uns@sd@4Y_<+@~gs
z)kc={)TkcwKjVAn^VXF!qt8i9-u2)0n8~GQeAAO_U0-{oot56Y|7bO6J#YR!?b#c1
z^P}^$&8|24AH9{`e>yJxU3~YU^bL!rzsr&fymVpWRH5|z+2J!bJ>C|~R=?7G>7K7&
zCrM72{?9(0H`38NC+d62W7cD-!p8d^?9hDv^Xt~1`s>z9vwL^x=rGLvVY8v^-;FD8
zza+2t{qwQ=)a2{aw?ElZx|?@@_XmBu!tz-5zTH97ea_seQxto)>z&$<+CTe6rwZ?1
zlXd>^VaJ@ru$Zsk`_D&Pr@LJGCteWWykBp<%+`wR!xBG!c+Z<KIiT+Ioj3YTKO1@4
zWL?`X+5e8&w0Y-)DTkiiH9!Bg&dq0kv-!TD8l}`fKmR<G&weej%-L3L;TPY^CGmTs
z%{DK-E$sf?F@ENpou(6vV)g6xuKo7+=<|I~zs&Yq#kavSBtIfE^Uj?w>%M=N*<G9O
zo)<Q0^UD@NPgi@(%n#GU?El$UE>8KCSEKi5ZSv-8pA#p&%@qBAz&+Ghjkj<8kEbu4
z&4g~7Z!_yV7|yd-&OeWZyZ#;P)U&afHuW1{hS<)N=#~F{qG!ji_<4p$elLsT|H8rd
zut@iu;CoBi;FuB?|IaJt-RhkDy*Bf|;Gd=LPU?yM3*-;0D!yD8-hE~3^UuzP?|lpT
z=DFALJQjQR^7GrzZW8esQNJTP*QdX4WNf+l;rWx5zf<mg`_5|{_G<sXXNJe>eD(FU
zjeqSpv7PU#(NEX6=jHFE<hZQ=%yiPI|M&6#+vDyshBy2P3Z8t=|D~-&#*EJF3aR5C
zw9h(-<OE;KNRV3*e^Q)9$!&r4Lg%d|Q&+p3cm7f6c=DyDi$`ii|0R>8o3+^;dwXoG
zH}u_a{o8Rbkhie<>Nkt^K}t?LOD=9&yG&M}XG7)V7guB#cKm#H#_@grs;twok&o^k
zXA8IZzT?f0N8;<G{$>@;x!&qofAZ;%1w8N5I8NVdD6W4xD=@$A`_V^g5sUpV$4fkt
zU07XF6>~IRPvTpe0yFz5Q(<1!xSeTVjIZz8@jGVCYW8HtE5&>%YF4lG=g!zTYxcZX
zX7}gJ+$;TYdfn%|o0cEu-b;QP#?k%t_cQ+5@W+RL+0UQ8z3%C|XA`AY-|BtNn^L)<
zGTV8@jgLo<-@Y&R{ZUhT*__Xnk*6irw)7R>u+p13<KIj9bJD*ge{;6ySz4xF^^em(
zdxP0K>n>mXi!$9Ab9O#jXZ(5ph05Bu*2fZGxAD%?HaoWOzM;%Csq2S6pX}NE?y`*9
zaVwrP@%N+RPt>ft+H5=fz~;v8zrLq@b#~Qw9Xr<ge%jgR+(O%Ydp2&b%m21uslAP1
z`FtU^pN;2?A0+ljt`q$(ZJm1X>f*w@W0re9H~pU5vOKI>_bq7k)~|J&+t%bXyq>;j
zSKpH9X4m>7L94p8uFA&*T0T9Q7*}q-@LcC#&encV`|wE4y39hKrRhtSh<?g>pU-^x
zVnL>njd|cjqZa8uJ}si0o!uGh7O&LcZ&#Y={$29g#z4tzHLZwDU*GuKX;pciD-)OI
zo#R&bQ_WRabo8`}#>V0ep4$6GPPBE4`~IK5_kBbDE34b~QQLAV-|szZAbp5Kz)eUY
z<^6>=m3a54OD4Iup3U#m&-}a9_q)gJC1*3-Hgj@vwm5Jox9mMA!L#k-z2IrOH*fv2
ze?N77jjs2ml+y{1tABJ>zu%d@?|YQhbIbFUahdN`nhzw%MX39F$gO#H`R8MgbH0_=
z1UFXCIhelcu}i0{&+_7=$j@)Gb}b9Ie(kcq{NzU*f1WpMt+20bd=lw#sqN51rLD)m
zH=HWV3zjak5ty;=Xp(!ed%SJPyr=;EuV*duWxw3yED>i*+FiVxIr0063;Q}J8JsOR
zcRsYFaI)5>&-?3aCg)n-II;16tF?Js@V*(>`Na-=7u>d<iD!G$RzA0#F4OqTc8V%k
z>|6R`*_Q9Oxe_er-hAQB<1En0>s9%M(Yd|hOQ67kI}Z+&Fw9Cl^h{>;wPSCZ`bsyS
z6XB9QUUKW*$@}eY@vpW$>bt3bcj87z4x`WCU+(zlQ@!e968D1j@7UAatxu`3yO{3}
z+$47IMrh5iqMYY0vcdlU6He_vaoG0g<mSce_l9qK<PaLaSH%Ce-uAc`XKwG$d(>_s
z@ORCQ<f_frWMschJ+tBEo+`P0-*-6v`WDnX@4%O<>!;uGxb3Z}FW=$2aet6C?<ceN
z_}$js$$@M1GiH1KeRTPPSn#h4r7MEc9{l$Fr0k!wdS`3-4X^K=q8m4FDLxjT=kxc+
zK}*A{BF9<(WY})m_eV;x!-r?q-PBo{^6So4)|YKsHQ8UwX4hufr60G|HSD$ga;8Mh
zTY-OB`08HAKi^JV3ae#2P}0Zu^sVv7ZFyPhji00SOU^9&T>AMDAK%l|_h!*DkGHN!
z*m*kC)qmTfU78}MFNMSQh1Hnazc^l;`o&LX&84OJl^W@@C%A5VT63JIYQE88=35m8
ze+#ynP5UR7W!)RPbZuLP)Q;QUX=g5fj`^8&^6Zw{8UEhxd-nI8mYBMy)~500mBN)S
zyY?vAKkh#B=hQ!D?e@?klQ-rt#%=09q56l<BmVbc=I2x1TW?SKa{3-;ypxGs-KXrG
zckV2>fA${tqzAkI{|b52v!>*Z*tC}&M?Bwun0c4a(;-|@Uzul9*UdNI-%qi~+pWL)
zeRN;hPOj%RQ)Z@!)&Ke$Q?_tf_vH8|+)F;$GQOPBmwbHpOy-}5pVdSkS)9RWA9di2
z)UN2rtRLMznZb!~x&w{+f-)8Qxc*37_^{!9|FM42-5OjBRxcMD3vZ6($#7=;_p9F^
z`NQfpn<eZ|<?}FniJ9`y?A71PZ(p(;61u?gtZ?e>mL_Qi!L!l(4HKomoZwHg5v$5A
zi!ioaDx#989IAKKG3xF1r>*-1mX++!o^s6X|4!@2409so8b1Fz5!I_%akS-vTDi4v
z>9Ku=(j5AK)y|vTv;7!x@3+gV8^OzZSvR^`S=wzVj%vzgU(B-N@n<2=X(Fbt4=+Ee
zZ5_k7_sX=ruhsIG6y^j5`~TR|@_?@>!k<MUe?F(@v7G<cTF#%~{M*+(%R<#5Mu%a`
zry!#g{qovk(fwNHi&C#X7R>*g+s`}k<$6EfT=8k{sX-r><yD_*{Gb2-*%AL;37>Z)
zSZkbHuCR0RwfJ+ttm|!We5~+Y|M-r``MzRKjrN?g+zbzO9T|gO#|AHsyUAMJs`+%;
z>rKv4Ey<bt{L0_oT<HGk-pdz;nwONOG9PP_zc>G{Ywh$x{b$X0_L}=`d0McqhiU1p
z@b_o#c$Y5zBUiQDBPi`#@$*TmuN+$<RXRU!-}c%^w>9NzH#Z~~J&WG<apo))^EAiF
zvz<J9r){-poVYJSRc#M{MNY`~zi(62SKbiLE{gEml>G7H^d$BrvVN8EPJV|YjcXlP
zJ1VlPOYIHjN~lYJH{0R$s&tO^hu^mo?h8b>iAmV7#>ur@x*&FOA_vQ}&kJR%W~^(9
zWUzDRj<8&J=~LldzfF2IHnN?T=h_dd#Qi8awAfhU-Me|SE4vIV%l+Q7>%O^QFu`Wk
z59>W!b~fF<=EJSLB)|UoB$YDbFYoRjU)R39{7OOT@7;bI_b5g*Fdya=S-!_?=id|4
zAAgDaFiE$@r2ee>ejoRw*G+5ZI$XVPa&c?(g+t5c<?QPBY^tow@+hqCj|{!D$X|WI
zpO5~l^IFZGoL)C)KZD0phpG+vuQx9&cTWxY@$>oCZcYz%F=m@%@|>Jgq{^q4KAV?)
zDgEH<JxZ2$&b|77wQtu<n+fwoR%OhaCK1BiV8)g6!H3D<;yZ<fix>2FU*TR+!=&-D
zW52)j#%YxbvVNsn>1u4(nWju&$e7Q!&HKgj=cdKn(xvGc2K!aCw}gl)T%Di&>g0j&
z8A_bLYCd?JZ@e3>y3#^+pWj~BZz=i{{B$N;o?I1_p)J*aJ*)V_@(H@nO?NsO%-!ZI
zG{G-kH2BffsYebiZ)Gl;*0OfuXZt#Za`vrPEq835yZ)>7^w;lh%}#%mefOtz_VWpB
zs<ze{`Ef^>JofDVb@!<Jx=WS<>m)lJ^};{hz3CO9mK(h6b&5^!8>I}Z*z0p$&v#2C
z?!R|8VDfua>pAzVe^q?8>700u`DOPd=W|VgSHHQgOykO|4vJPguBrW^cFE^Lv8Kak
z5>H=ve*Spm{`EiJ34LsoZ=WR7ZJwN&-;>mG<YfB?Nf#DorML;s!Ax5$<()rlbt!7Q
zTg5H2{rHCbHGa>#RQ_8#8Qh&+rNL`>al35hqi4!C-nE7^J}s1fI>%+v_NM|9{NDcl
zlEAFJ{(aj0J%)2P^KSip?B~ZG6(P+fiQg3sZ)r|&dGdd6VWV_(-QQm#?fQzJ^uPSt
zmep>;JAKFZ{S!Cs+`#uSX3K#IItixBH@%s_@=etGP@d7B*3#4WS*|vAsGC39A(Q+}
zWP<OawvcV-XXf0fN?OaznWkUXna7;{PKn!~+{pBG;>(qvBi-s0Z%=+-l)lhBG?C}D
zVRiAMsap@p*57-wbZ+Ilvgf@aJ8O2wFwXpA`*5Per8u#w`EtDmj_)gTmv<&>UQYGW
z7kD+N=fL}eCD&RjISfkOZ3LIi=E^Pm^`X#tqpamUmg+t8^qnh8uA7&CoO1p4w%{oL
z(#y5sjKyNwcIysGgiFR8nYj7Nv3rFtUvUbG>P`C6Ewu0cm1n!}-0+@p@@lnPH^(|L
zaSi#m5~AnczcN{7(8aEn6msCD#$0Y^fk!W`j69Bca{Ribo*4OPXStZb!IJlHjx3#J
zuGz%=<wcT)!?F#lI9HY}vpj0rDCyS7aQJ@B>3wG>t?R74cqy^r+rrLN2mLuECu=$X
z>79?g8yUeMz!ueauI<{x$`{(djB9xNeBU=K``ZcJZkL%nNlaP)<)QnFE$lwzl>WOF
z@6I&g=YzhvhYd5Tge{Ht{@k-qM%-^}|JID8BB3?kFA7u~6Vx^PX&vIc>e<Ep^N)!=
z<Jp#z#&bMRDPZoL-9Z(vyQS*GHK&>X(s5{J?Mt+s<Xio_M4R!c(53x{87}E=aKF{@
zf2Z+Z=XooCzwOr&pL9;S@zDEuS^mOIC$7Y8i12tk>r`5|k(1t~iTp>`+R8ZmTDv;w
z+w=7vdP^#pj1Ru)JK4Rtdbhg3pT(Qs%KR=*>e-Zae7kzy5=Z5wwNE{~pUwYnuG1g-
zHT}bdy)!)>mqyFa`k={sr0PHp|B(gRo@pl}E9&gEOFEd<E>s#^+wZw%YOc7|WX5Ij
zKR&;@60Gqk<lFpnHaySWx^^#fn^(pbIsfOTIw@V22iX_CwO`eoUA2ADCb6}d?{9q-
z)fPVV=7IIrR{k67GHTww>-nmZ=p6TaeeKR4ez*Fnmdr}l;b*vPdoxet>*u#0O1?as
z`6H-&Yt;HXVRtNNzlt%p|F%GTTi5b$7i>3Oy?08l?`kjOrnJhk8w_VT{I?b9)kmgz
z9Bli&xjZP^V^!MK*Q)#i!E>L~%$dKrIPT%r?dLy#7COsv%3ND>MapOAowr}g)LPB^
zdHj(3ihG_L9edB;pH?^J(ec;`9_u&Prk~S0d0O2q?P~oU$&!+i2#*?xRY$s4&-){2
z?e<;l%m2^+&(0Q?c>72=Yr_8Bo2L~E`LirY@OV1$yi(Dvcexg!w+`0*D+*liR%Lwb
zhNH%LJ9ecDw=1VFU0&3gp!}urN>An4`*L%Sr3wnQZxZ_bOlW=H+DA`xH80xRYn3;=
zym9H+%eA{#<uA#3w@S~@Z%c!}*o0e)3?<H_cBh2C*}O?@TTWMJV@-qZ0ahQSP$nVf
z8F}4DJ<M4aR2jq-tKN95QZLbw%&>j(<=2G<OO$dnCSG=IKOCI)`ruZ<uMV5<t~$yU
zDtb<O#*w17nN0ed%#1mN0up$ps~fE}6xXgd@sqv8GWnR|a_Kjd+PE+8Exz}nSJr)1
zO#Z8Hro6`vbbXWaJ}_JTc3ME(tYCo)ud*%wdA;|)z4P_#-4*p;9|leR?3if1UBD&k
z%Xa_Xf}IlzR^EEFPHw_%i@gb--hchE-1Ku@k#CJ+<6D#SzkhzabyGa*i{zB@up)!&
z|2AGwFRzu={-`wL-QmfPw(|!~V|Kn`zhuvmu$^uP4Ag$|a2+|(@ix_Z^5XOMj?X7d
ze^l{jUQXF%raqICOTRv|ub=*Wl^@d-`F)nj2iE2*_id5+E7>vCr;*{+{&o6KzdY>}
z_3e58z5eNq%yUaKuig;Xc;U3-@`<}&F4sG)6nyB%ocM9i6`rG}+g`VBud(xQd{$+(
zq+Ua&SvWMEQ*J)@7w0yE(EH&X7v$<RZUwUZ<-M@&k@-gEN4vMD6-j+5`K-5xUuniW
zgUeIPlbVfHI=|-naPS$f{$l8(mhtWF?dc2n!+x$h^iH9~EvcaL{kENqJ2@@>-97b0
z*Jgqrr$Tzsm&>1D?Mx^*ZLfbd#j^S3Z2b~7fzQ(~t(z%$m)~c5#9P0@TVK;GlHPJV
z{r!5@Rkm}F-=oLfkN7`2SS_uqsLNrk{Z(}9`~<h}=l0&5|5@rrO?}M9{hh2L@=E#Y
zW^TFD4X^CfDZgHQYPUs0iBr#}S?rfIbq!53x>pq2ZIYES$$hx>LFd8!X;mQyFI&Gh
z?%e7-tIuh!$Eo)x%l2J+e@CwT`aK)r_*ZWKpItiF8e(uzK33xWgdc_B`&a+i|0Z{q
zPmy*BzmWBo3x*06o2D8R=9MIT{S~J$_tEDAaZ#p<dm6-A6n;*6^U9Lr^XKRLY|j7k
z6Tbc}d-EaLzssNVpZ)Xdap^A4sadw=^YyFe$R57+$mjZk^SeQH?TdGMuT<>>CeJyR
z{f{GFbPFHDrIj)11#{PTmtFX@JJVC>Vzj~B^j7YZ3ctB;TN=zgwQN%jr+)s*H_w?n
z=k%Ao@{!pc7a?Z<f8nOttkz|j6B{^GTyAiP_E^rnZ6IV;zRr1%n67~d_lac=r=I4i
zZEZW`{bY7RUHL+pq&XLVye-{kW#1#Iv+<11OlH61DLv(v8LHHoS&mOkRF_ssJh6R3
z_(8A9z6{UT%w#^7dnu^lh3}axy|kduo?(l#6U67go~;r*v-{KJhy@S%(s*|?p43$r
zFkbTZ>C@a}5Aqh9O4OKXmqlCEy<ge&uw|lFxkYZ#^zZd{Yj(&P8C3qMeRJ!L`Mc+j
z^{-x>zpWTlT~1%s|8Ldvq}datyPI6&kNsZsDg5R?t>^b%20mRcmw$2cmDAO%Z?9ia
z=b5BG;Wp2qU8cvgeEK;ue%#!2L}^vA&X<$%+jho0{i2vSeKl+Hk~qs7*FIOvKAPb7
zBS2PBDDl_q)t4;{4PRL`e%ce`qqpb!=ZCGmPx%z)KALbgxuZl;(P+*mn=rK<J0GpG
zIA(mg^Q4oXhNpFPRhh-?`Qj)09IoFERcN?9)i^LbQ?u_tmfQ@xifwb>THJf2rz!4#
zZrX{HY9|l+oacL7CQ>oULQw44scDCP|JlhGow@s=!sG~h;|uXpFD^I4OSJe%CSBxL
z|FEV+>oG%4!AfRNwu?vH?_PCZFx!0PdCw^mL`{WqS7%?{9KW}2-ruj)!8f<8Daflf
z&fNXFSH|;t;L`Q)rztP4Fgm4^a5>Q^;DFDuzdmvW;!^C#4kX-faemFj*rBTX%j8sT
zZJzvt=P|Q_6t7x0KRse`vV$i_q4#6h+@<fm_n3cPR$R7x8N-^}VoSc3pXyvYUsv|m
zJcA5Jy%5bM%l^Ec#9hHKXZ`M&wa>SGaSXmx@O?{%mEFC|AE$W=E}!#N?e0?bYg{rC
zUk%pIER|k(_ho9sVb683_cryPC_8tcKCWg$^Iw+Q{R-&@cPGECS$qF<-o>&zj`KUW
zmrh>1;O)`5n^(Lm<xb+Sy3Llj%c|G#!&7D5Zq2@km-{0pJ`(czo}!&td10SU=`^J#
z<=awBmnZr$?Y$&m)9zTGP*XE6;u~Amg~alt)7#noJa)<-Rj91*_5Szohox88t#^s>
zo@O`Z2K3Iai=S!h+s*x`MlSGupZIF_p1`(SuXEciTXnbJEDjEldDz5ZZ6Y^!&b{qs
zT6a8OZfm$FeMDyY$F&AuK0j?XT(7@yQRD60d-~qi$0n+TI`cjId12w(+Y9}>8PX4K
zTkf#yLyf|M`=Jt=vv;rZ?h)2~u$%P*-|d`(3G?oscyiFJ?C$@$><Mwl%ASSaKAmN}
zL2l<PPv>(7&kFpzkoqv8FS$pR^SC>!r<j1=8ZVwk<_U7Yc`PH3-jkheW-T0h>P*5$
z=?TlERmx;^@19Y5%iu1%|9OyZ!ymJ23*RS*OIJr;`WaNle4uJ^so0jw#S_yskKIW)
z`1_u<ox9(fO}iho2ggnDR58mxzF<oE2H7Xdp?df8>g8JP-Y#lReU%=j%A5J-`Fb<n
zt7Yq6m4kBKx>FKo3-j+y{+XKE{?Q{XgyCPCTe`x}dG+@uefyYSZmzTb+Kr{2m4`(C
zmE@k@bcpL;q4%r0x;ge@zf7!j-`1}F<z{&A`OZbbZj3rLU;1?VZ1>fb%Rb(oE)=oq
zkeMeJ%QHSzf#z8<tY_y5986X3eQ9p@{;#Yn597BQkA}%7y4-%>H?owEwbP!q|LS`W
z=D4hIt8KAX?F*+ok7-+;J?Zw3`O+4DIFnhVs|$l!XHOJ9x3ApVK))s5QT(IE7Nwv1
zk#iX`)Gqw3*vRbuF@(3=|HkwqEA}0&oNzlJe%a|iPrq8Z@lDRyQ1^Y4ireit%?nGC
z-gbq!S@nE>CSPpf^)<zNx~Re@rX@KqDi2kgq<7aJW~+(dik#G$`R$~S?h~)950ZWn
zS9Z@n$n)yS#6=7DHXmL&UuOQlZ_$-E3Ik5~@^PLJJM~T^$-LFx*+Rm@JMGAZSoJ80
z^~(P5h4`OMKKs*PPI<S!vrV57w};A%#%TQwe?Pq2ZNsmysdtZ1agpY~yqCAtqm46N
z=Wn?nKST2BclS4`t_2(`Q)h5zeo6a%P0-QKAf)JpkVDl<g;~~r>#wWG{JM6mt0Zab
z{+cacck!QF@%)~>@vjs5>-}vG^S5MKzxh@1u+eDkb|J+Japj#B0tNpbJ)HK2pX=)7
z2N%@57H@9}x_j=e)SFuw8@)92mo^{&-RN;Lf%VJ1{$Nd+w^wGbsqOH5Ut7NIbNJtN
zvn+2Ie_Z-5{@sfC*-gb8w|r<gR?7Y8bnI30h&#^Gr^2&%Ojf3Co3H16NAu21g)^_u
zg&i_^cdmHRl(H*aoEugMo-xY!^tQU-yUj-3t#iLz<PRx(n)L9Mh3>3`k6J$pdi!qQ
zZRk8b;nnB8t&5G>QywWrtDV^~Yh%;j#hY7`R@ZD~un${o^H0St;^)CF!O0mmzUORO
zCu#b5w$;A$5s}Pm{4>KZ;@r_&`{qxI7FR85kzl$aA2IWkE?>Z{F1G6u$31Sw+^_?6
z7%%8DJYOW#@UkbiTXy=!2(!Dq1=Z%!UzW_$Z1NQ--{`b)k8n@v_5H2I`xYnM;!y3L
z+-@-0RDj2jwIToajk7$Q;%UaoN?8eO&qkH$@ztyR{2S?3b9nmG(>Jp$j=lL2$T0ck
z)H4d_o1dJvogv|WG)4CL4bH61Um68#?Bx3jro522&O7bx`qjbP!%8Kb@=kWkOwlk3
zddhdrE&jcIdGswgk7rX4>G$TP=2aM<(E1y$R;zn~|7GpQxyxVO`Fb{X&+<)fk8PEV
zeS1Q`{T1jv=C;Up$=Sa1))QXx*hk)%ZLRFgamrbc7T4wV_KiUKnlG7CX5Cty_S2^G
z+o~&ub2@GausjIdkytl}*(cs)?Rou<?Nu+2sHi<KH;Fj-l2=GVlV`qcUdq(H4VFi_
z4_<4}_j~?HD4_29C6(h2K{Mu`JNZ#{V*rOi-MjY7k(%2|-4YxhT>q~t7gTxZ)BpRk
zV)#_Vk6pCPRIRJpAb3teSay?1{K1!khG$pum+SjJyETRL@8Q?ak61-KaY{Sw%<xcn
zV}ir{{_j_lrPrh<%)OzpC3YIa6&Zsc^%do2+H0?FT3QkI=TS?swfWop2M0>J)ZKk%
z&L|MrT`5;4A&@<P#;YwH2iX0$J-KWlZ~kUw@fCj~6VDgtEiH1rx;T@xpC7ZFRNqk2
zv`XgVy?Gau&Ob4+e!2BQ(sPDu6ShC)F_*QhsCv76v!vyfSuMr0jsz;i{;y5H)fQc|
z|Ig$Dt9~7fz4@k2r{==uCHtq^=;(apTQy5z_TkN!+CS%gU0t##J|a~>ZsSh5miVmO
z6O7eXK4IHdk)-smV%dUUD(rQZ(<EVM0VK0UdlbIiUiMDw^&^L0o`t8f7CKAj@9vq&
z{@wME{LdP8?cGYVOUo+ztUoPc%lWFJ|9Sm}GhhGzc76MG!l90KyUH8f?X$xj<gJUJ
zZ&%Mza)@2Ky+C^@*UTl;eUD6k_xoeCe5l^NnmFV6=R3bei3W9^%-MSKBIA1XfZo_=
z4r^G8S4T3G|K7Ih18a39OUI4xtZ$|qpZxYu=>FiTcb+DkKXQB5LuKbppOc>RXqgyA
zT>HBJcJ}hmU$2%+9sW`5vBRf*@r`%$V<rbYdDHOXrT(jB^H;rp$nbo&db@meMP82T
zeE*6sDcXrv<<E^){{6f3chSn<>5ug6<=?-nyl12NiRo)I!`I0F7T;qnWY%is8>@Vs
zvBq)9?tdr#{7~B0^Rep4{9jss3}p)b@ky@#CUN@jv*+APMC)G6$<8?w|5y3a!Tg8W
zns%qwiMU<TF^LOH_L<gvx5epF<NjAND~n#;{xTu^*ekt#6Ef6amFJy&6F!yC+R?u0
zb7=U(pj&rB|0d1JU|g@Okb3!Jw{_#qC*PC1b6Hy?zikLBYdop5G{^6sWtr0a8R37A
zyQ#-5E1Wh{^!{buy`{e+)~>!+$K$(c;|&|>+Zz}?%`WWNQ6J~R5Ho#K=UuG>?}KM3
z7Kw3BN@={?#~SnI$Cix-41c+%-`ufB`MZ4Y_5iEeKT>Y1V)CVC^8S5wTuh6-=lkB1
z*0(P2cs+afkBj%_Z@g#c&sTlC>i)5y->wFCUuLp$Zxfz%wsyaJxc|oHmn^sDt}mLf
zV@}cc{VKKDagU=a+Ml?`-HJK6O-5-^YW$7bO|8k^lTAFcVhpO6On-QHdeQ^QjZf6S
zZkqqnWBw{N>*8dMr<~b4%F5P!xKyaeu3XE1;InA~hgT!p`~P?24$qx!P+5`g&inh~
zUduaQ|KFK#ZlBH7F2+s0a~ThYaI<?TMkp+>6Z+h8=-cAv|9y+N^zQ2Jt<2k3Rq*o3
z`Ugv1o?9B5leB*G^;Ew%b;buuvc%<WZcg%lT%%I<_u%GSzULm?ALd<fmw557@A26~
z8uQP;{QG$BoT}Y39v7>|r76UEr~lr+Z*AAb&EL}(92Mr;viR}r8&4m|{FuuoVn3(+
z^v*klwYpoI92H;Q&aSQO6=dG=r_3VL%h{O0>(->-vy>Nh9-ZKLXJ3(o@^W2P-n)0o
zW~I(dOg&V6Z=Un&zGzL}@KbN&_wALuufN56kHh(Oua|iy>g|8Apz&_S=Or7q$6MW+
z^!3icXQ~V`i@Bw%i#*h&XO%xsZc2@rS>R&T@bz8pZ;iXxxMe@uOz`Vds4uPk)cA9y
zmg!3-y^4h|=3O#lzIk)?1R=HS%}4zsCi1+M4cmPu^u@_C<=Fkp(?30#^yBlrFS2Lq
zRWp|c|L2mbDth^N)%y4IGvj?eUbvdGu<ECYyy!Ody(LS2M(sM+#kHpP=eM4}e{7#G
zR*$w^URrwhT5i#rFY0aDZQEN<xw1y<tx*4XF-7>E*pnHrpPv@m?%yZ>*GFOBU*%Gk
zT*Gy5U(O8Qcz=$aSx{-m)4n~dkCd2XlFu1Gl{$Xz_{3F}7q4WnPl#*TdGh^zy|#Ke
zYr`+u5{q|BS@(uSw?)5u_4A|5gmcrK?G3j4o^ov8*AKd1pMTeHaxnRj{y*sUoc#F}
zBDw}D+!K-mkG&9P5|H-3IU#<cN0XyeXWBxAc%9m)!zvT*ZEh&cORf^@O6G4U;oeq$
z`+&oHha3gB<EJ=&FIl~sy<XK-Xq$z!y|VprmfMRGgj#z_wGYIxOFA!Vu<7Ako%@O7
zitFUhnik6z`7yk_#InWl68HZ2GPbW1-e{%?=_*7yY&np+rGaCawd8(@O|mjh4L6PU
z&n*y=IB2u_kH8aC8^a!HgAUbcc@`0(H48qyx?Oc(P3twQ8(ViX%$odeSFwWc+xrcH
z{}xwomi;PNJnKL;(|S+4)(GiCzdAQs>uUF$70n0_{wvZJbg%W5p4uI~mv`KDspW2-
zlQPx$^@r}{vwwem{<%@DO#H@|%m^QaYc?<EUVb3wV|%Xf#pL+e4-#$(3!Jxh6|%Zw
z@|9`jp<nSo*K}zvWcIXrTl3<T+2-%%*PYnZ+ipL~YO4L=kZRAUzQTSdKi`#&-VTkf
z40dxGRZeotPq&_NR^<K3Y4dJP`+tnp@6pp|t(-bMy9ItYaO=-cUf8gtrsaIkyU%Y;
zX0b6{IJEEc*H7nL?zJCGX|tKu%gVB@?x2&PxMS|CUge*PRq1mNpPa;ZuJaY|=SPMI
zHwr6Vxgq)S*S4%}Z$0YgTnLt)bGo$L*Kd`b?S%tj3+)tld$qGl7P;*T4S1gLe%fxU
zBNge~9;+Fm<XztwJ-43zP)e9r^UJc0U$3YIF06}`FTYjA*8Jc2U&iI%t+^}NHutKT
zZQHu}mo-PJp8S1Hqtbp)y*qB{Z6#i=yBnh=dqN%GxeD-wys|j<d(M*yM*c#&CGQ1K
zG~BT@U6ppR`?-9^>uoM_t1qf8Uh%ne@4VX!JC`f*=6$-Qx8l(2%lg{8Ecnk|NPB<p
z%Y&BlcTTqM&=TNzRF}+}r5|x6s<dBq&yLXi`=_32JvMk`bMsx$mG|#vdMxdHzV~Hy
zj7LLb$tOm+@0(?3w%k*Wdu-sO`I<%W@+6^k1v*FO<X>7{_bKx1+AR@|+=lu0r*AzP
z=^dXD<-bj6+2YTB%|{vMeE$9|G)s2%<;h9ERN{a9K5RKjhfl2CXl0MT+?mSS>qpNm
zQCsUdRaW#y;qfacvXV@Am;ce$)Socn>WhAkW&ID^cL_W&lRu$;$)i2yjlh$`-m|r3
z9ITH1Tr~OI^g_%2J;$oqpBcRVS2gSQMeikCIu*sm>WV*IZ=CbaEl#}qXlF*_!sr8T
zm&!aDw&ZW0-spX+RC8k}<C}GdJH(@BsfMmOe(lT}Gt;nH3+naWCS4S>WfPq7Li^2b
zQN>?Pl5^&GKJ`+Y%v2SgFsWzp6|vJPlS(uWJb84Q;jxLUkk|XU9JikrtjK<GR=l`N
zOaPp@7|i~3ES%5kr65~$etln24oB~~f7Y&N`0Zp`+^hW?OIY8#{ae_!Sp3SHfA{6m
zH79T?x_3R<(Vl<YFT*|AzHa{_&+-rA>qQJFF?x0d8p+D!hkY};thsRg)4lw2D}Gf7
zzpHu0@%ICN!aauQr29WlXBu2coe;F~Yw@X_^53k3N?6YPR^XFZ?%H}T&RST(Tk0#z
zGb?NRMc*dm{{IrEsT^BWWaeU|EB^YyVu9pF|8J%XY+g2Oo4|9zIA65+j@Cn&r9v9f
z{42%7&l%a?ycB$UMIF<Wmy;PH<tMFEw3ICGo3uio|L;2$F_8;#42RD(Iy|anvSvwU
zOjO(QLUW44vE!k)TT?5veuuA*e=x6Gd+(|12lVbY=x#BoeE2lD<*AUvzwf-MZ`SH(
zMR?^1EI!(lWp4Fh-Yo^4%a<*E=ltVz-=m)-&VKUbiVt>MGkfObW-dHpSy8op&gHJ_
zc{&M=VVh!@%C@bQs=vcfyz}6`*+-ORd~+nNtJN<UKG>f4M1SM8MUkm9WUl;L%zFD=
z`Lh7amlGLzg&)|Q44!#p)!x@@H_hJnVrs&|E$ij7PhXk2M#j22Eoee>WR2jg3!z!e
z><R1Vy_og!-pcJKS}w_{7)};knUnYWn(HIp=V{i)LLo9r&Dk%D*=K$JSLS_-b6+pZ
z8|IR^|F6#u>gQmr`?v4IjWwn>1H}s_Fg@JYwcq{ik-c{QhyL)&Jp3}{_R@frX5a4T
zHCJCan5S&;*P*qn$h|++-ZJUw=6#imOxH%Hb5$jICK<{<KYda9c5levdp5%Pul0U?
z++rT{wc0z-{g1PJ*qWu)6Fyf)?{H|6<=FK5`t-ZkQn#$y@Gtw{bKRG-Z89yYlC=}1
z&E|#b@36C8(`F*BIZv$FPXFbZ#l<JiZsB)**70c<x2e=(A^WR$v_3ySD!Z~@-%I0D
zx$L`Fa@8NVJz2JV`I(OYXLmMjy#DRu<)lerl{`WvofE=t_OSkct1#t5W4V?85%&65
zOFuugmRCIeq(f=uL8TK0K_1QWOK*$cF0Z!|>iHzxdgq?_qzc7&kLCJVJ9aYk{#`kL
z%j;`bZ+r>5T{C5NV9vc6+ZeX9w0)Eid#I`-^!Sd5OsU1Ii@LTqa@f)u4+mC=yY8u#
zNq<@DrgeDYoGD&QCHjk41Z7m8Ea^R#U(#p4p|N<mVXofBkBeAcxI4sloB>V9iSr+j
zeDGxJk6jC|m#@0AdE55xM>Ty)2D2Qz(icpMxbcoDx3*;K--(y_m)ewn<#;^zt;!zv
z1oizK3O&n<CowD5mWCe;mjC%>N?Xb8*2ad?Upu%%(hVCj@|&!qt1g=*J@_1*V|&#!
zqPn$o;xX&<tDeg~;@g_zf4)_W^+zkO|I_VE7nb!eH;DLe^R#rId+J4{>E;J7?|$n0
zuW)`^e4SwQpUrH`I8HDqr^&M26O^d`e)gLF{zij^8*&q#&Ew7aFR#^hTc+ql<r=@!
z8;^@GZ;$!1Y!8#ihBZ}dpImbnKO&obIdaPRH|zE{eqwF@$Xiya8GkH#`Rz>~)m1sC
zc=BYYxg-}Zs#lo%+WNj;^@Der6WixU%`34|^hx~lV~y3u4I8Z|2c@~N7HMza?htHb
z>T@xK+4t*q|Lw8DtENcipYC2^#kkz~{JZ=ZCf@fq<aPV(nGN>-pPuypRYY3Kf?10c
zJQx4E+P6#6U)NvHCS+;vuYb!!&9ix~1uTDZC$%MLo$}n*kGJl6v)zAdVQus06Z2jd
zACB5@ro8IGO8r-xGZY`TUGIAT<7ug9^+mJG*~g<k+1ak2xm!c|$sJ+u8xszEIhf<#
zE8Y2PZ>^egpn3j(cl(u55mytR1Z+52=s0!h+qMt;-{x;xzwDb@Q_!rLWkwg4ns0k_
zdYSS*zo+(hf{Lx@h`o)UC>-|v&%$Xlr)DvWos-y|8MGt5tU4gFM&X%d<CnEo#aFuB
zqt-9y-xsiEOF(n0imb$|-u)5NDpC^NG``n;`g!bi;ISu1{KOA_3H-z4bDr<{pRWPJ
z`7KWr_?NXarp)h}yGSHskL3-EPH&4d|64w@d~y|$<l5i#VBREtc`JGE#v1il<GcGV
zcGiS{{I~jtoxsc&!f%fKJwMG$p!Ds=>4%@yRD+iO{Eu!>o_Jx$Y=;7tJumjY-}QKw
z^hEBqr+Z_HtDU;uPiI^lP<@W^W6hQd5pW&SkkP0Z?Z9%G-7vOj&7{lv%TwG9FMQsv
z_eJ4q_jXrqgShqceGF$DIh|n|9+qU#-FBy6zcD2)X5)k!O=q9ii&?|$*;L(@wtSrt
ze=zMrvBZV*hZ$t#&4iC+oNz8+u9fGO%`3`q@rqD<#;&@eUOK{(eP+zflhdBAzA>-n
zeOTljlf>7p=PFL*7C3yG^^#e(W!I@TU%h9SzuNBP&tG#&^j^HR_Nxg2R;m6w=YF))
zRovkuasIht*2I%p!F$3l)bZMN8{S{|RVOdXB}k_~RyO>|^4RSP-yU4HewD9$v;R52
z>G~t>`&p048N0G-+N{r0{CZ3B;?wmqxffrbHB)7rwO%GS*6(Scqr%^dZ3@Sy^yt?l
zR(V*a#9TUQ_=?|t?Y`-c-p`e*Y&h_x{rS;p|DQ)h+cw-#xfEUhfZ6);tn<>MDc7r>
zudlV_OS9I>R(bXP+sq!9yL)QSnS5LF@%fUu{__8#D)wXsJ>kw2WUR{1Q9Z*haND8J
zKHlciCMNmgYcr<${GEK1GhJ4&&gS6#YlT($Kb2nEF8>~|-c)<`L*)nS%1;&9$IbqI
zt*PYdiOr%v+<Go4FNrgX{=D3`>!ki31FzW@?`AK3^*nlI{}IPz(@6c`KS3UQE;Rr6
z6v7?7Td9>*`{@nIl`WUg^%(xH@kqWGtPph4_VlEKVdjU;gLS8WDY|hdSo&x+q<n+4
zC=UGC*ub%j|I?b#8~o>AyT7-($ezG6$#;)#)Pa1lhA_Qwv6nZwBR%R?KM)f6@}+rl
z1lN+v8qWXRKb0nF3LRG6;xJun25U*{roHuRQ?%=t4Fxh%Z|f{PaABkG6)8RkYldhK
zW(V%(FI>5rvVsM&4=kU!D%|b2;=M5WP16}}iF1>juGF>E7Veu9JKIU>{F1*mItI1L
zf$Z!d$5uYcy=RczEOKey{k1;#KAn~=HJ`-8ZY$YUoe*~A;Qjj(s%EXz+>+tj_4#ZS
zpGfRP=~t6J?$hwSTKD+Khh%Fn>vW+Nd<_#H^Q$(u{@XoO)4Ju7{ObcwVJDWz|1T^5
z>m~PRc~#J1W~tBND(}7c_zf;>jcHoDpTleC6zfZt@<j@XpMQsH&VTlPP5e^Bh(jD6
z;U22XUKbzYTJ+$_m${eUdOz5-Dt_O5Jzou*sKsY3uCF=&<g{l$#}fg5%aY6Gb=KvF
zYu|_6DOeRZo!MX>^E2iYp~t5<6PM55ulrcxDd(G9?tRC8P4)dgg)8lCtjh!q(Qjte
z$=0V{=l-2??e>GW#*%q=ZyY^vv*{Nf`-a*K87V1Ao!6_b8C31R%NBKkbCS2ty<1&e
zPIBw!J<s?3{bH%|+RgpfU#fi#OE}%ZlT%UddoAss?IZ27uXaCw*{rtiIThZ&J<d5*
z?>v|8kE5TCES_*F@JiXNrtQ-+jJAtruT%QDWdGazIe#6izUA$E_pxS5#1ivG|CMIg
zdBrn*TI=$!`^ka^?(#@IF%PFnZ{}=otl4L><z=Kz4bRNVQ>FQ7_XMqDKh@>7JQeyq
z_s_RKPwvUID+lfJ7n=B5@#c@sztsy)wpKlUck0RGxldJAGVF4{RHw3C{SND?=L<~^
zecRc+yhP-Lxxm+%^Wy{WJT5PuedNUy+x;7RWUa2BV`5IxkrbMC{ajyb{*s7y_cqPk
z$Y)Ty!t6Ph(0^fF13le@%ZWQW4VC-CGL5!eWf#yqwlwmHUpKh1oWLO_aPW&`kJ~39
zbDNWiOwCI~^1H<(B)KoRbO!G#zO~cbXe+PIRj-;zmPy=FSsN_#t-mZ%kh~yo7?Z@U
z!IWmaI!7=1%M<RL<LkJl8LypnGRSPxi*p)c3egNZOt>HXvaw*B+c#~Z$q|P1w!_(m
zjR&_?T`6Q-?_;WQ**$W5CBN~B=m?fCO8YI+le&$XXR1C3yQA#z>FL8O<vVtDGhcqb
zbmgf%y3Y?<8eT2;>#a#mj||V5;gl6OP4<%4IjvU410`$+vMl2gPv5#}D0-;+kLA<K
z*ZbD&Wf0?+S@Ejt-IZw3%a58(DmfkX#5wkvnr{=DKb`wk=|b@+p2W#X%+~2DD}Q<J
zmicw>N%`$+9j+vu&tEzEuY5B#Ui`0kkDEB-PqC+`I3M~RnWcXBjrzJyeIKs(^ZU1~
z{l0GYo{2I^^YUNv&6bhAV`&-J**U{@g5PiP`?EH*mK^&&rTpk2u5CwWF3ffM`^8=S
z*z~)0s-G%D8+zu2?|0y4dtPtz)$+~+2fMc9>($BD`vme|E#I<7IH2KhU;akF#aAEd
z&SL*mTd8*M+S%=)6???a{44JFujPxFb2iex=KJBvX55#ntm@0^gWn~*mz@9qfpX>h
z66v)@lZ1B_XI)9J=>K-9x}>zqWA^UvN3Z^V^7?bl*9dPt!-o$&LVw?{-dx6AmbvBN
z!~C<4#Lpj3c&NE~*I|LP+e2-<)TaG4(YE>B-yBft>KrS^9b!^eyl*Opk9zt{k){W?
z)4sXvHgXnyH`!hx|L}eDm?w2Q9arD&k<Pz<QmilJ!MW)IR|{8#9G3fYmvb7&1xr1p
z38Cg+EoxrNSl&OETPgHQu_*l6_Ww0q6Z_?KYxFJKjc;iFxbXh%JFAx5lJIG}EB4))
zzqf|Z{$as`i`Hf;+y0-C+5NY#(PiUH2j%%wt~z|1)FEm!W1d8=d>PO2XMKl`3SVS;
ze3mI7>S(}Psd%2FH~N1VSQsiQtmzP5_wk$DkCpr1&&%@u`FY<l)ya>he0{(6<h{qs
zlh&U|%B$I%Qo8t>$j9Z<LFE_t?rU$F{CGpK(hNI3`CT`jZcJ~Cv-!Qr<j?Q(VjB#m
zHkp<eJh`s8wQ<gqw#7X0PLX?=H-A>0J)^8btfS%n;*&FO*3Z`6`L0-O!HRV2`dcU0
z{hK&*=3)PZFSKgB9{zsez2?J&;Kc$4#|yVOT#j6H=z+}U7|v}99tBV4O!qptecS(Y
zH!RY0Wb6wXEG3>N9P?nDaNA;94_ot3y|m3uUv)BF=ZUFYHdwi4R_1N#8#{O8ZQJF0
zaGLMQBb&}?HT*OVDA%=f=4)SV>0sH$wWGFZiTAf1yAL@trB1v3S?O2WZ$Hlc8dnPx
zY-MJJeOA~QEOcDeQX$sX@X+gAf0hN-%`a*auju{yaD820g469ekw1=X@#_xR#qwY!
z57UKKg<Zuvue3+o-Z(t-qt2Q4JZis_toO}1_$vJS49n-MA37Sb9|~P+R%Fe+g2&*G
z?8lO~r;R1o<Y}D`RS38px<|io-;J(Yp7Q%@{ND@CKX&)`lhdEqyf-Q|@?tnA!!23!
zv*gM2g^NOqrgwWB+Vc3NO^A?-NzD0O;*U=%EMBv|?7+>B&yPIw?Mh&rpt$fsZ@zS5
zc}9)<n!L%<ml`j$_ZiIpFo)x9sczLf<(_iZp3Al2sX@{vX@6XHb2<sc&NvbpS#m$t
z=I&<Wr*B<&ck%JQo84D-_~DA;1s@l$JI<4~=g!0F85U;cdhwujB6&Ir8*4iXzX|qi
z&fa+1gnxBZ_~Q2xhr1sZMJ`wrEC1#4+EVVP0?rkW=bc%fF{f_ltw-5=KioE!wH6RQ
z%+#*;_C@yGe;)gHZ!vz>UoHJyz2$`U|LXW_cJ)(UuH^Z2vSf+dbob02G41V>uLoCs
zymw=x!~GYjZ?;6{tju7%$Ta-|(^*+JhL?vA*X}7jQ^mPkHSFmrCbMV_hfg{@o5YP~
zI-Xwhi)q`ZPOq@*ch8=itDT&~Y`M_pOzK`$`x&Lu&JWj^%#}>>^hr}qTiF%K5wo2!
zf;~I7t6qrdnxKL&;{%xrM)ik_oNYF)vM$VFeY|(xOjQ=Y?N%=rc;EP<-gMJQ?C6(=
z7k$!ZawkqG)tLRM?6L8a4~~w~+n3vzIIP;OtG_4l1gmB0k!iJ_j?U#`uKH(r8qXei
zu$J{trghTo`j8tFYxXTw-?ZoM1ee@~mcC~Zy?T4Lgv$T=b8s1}h`W*PxdywV{ofw=
zNB5@08!Ize%9dQZ8^3$sn_cVA_t=&Dre$QO?l$}}pItirlaN#QC8x}Vemol(ZtM!%
z92BcBmHZ|mVWGpLrw49s+j#jy;H67qJ%*lkZDK0Z4*gv5v`$)izh&>+s%a8ypKQ6r
z>9IVZ;NrexDz|E;*w}28Og_yN7R|hQhIYjAm3>E9RvYT6Pda_$)nS3C)y7dvt@X-|
zMXp_<m%%UCSadmbrPkI783*p~o^t$<TkVf^pU%w;N#A|0$UxEOl<%(U3(wwW80%h+
zn=O0mQJ>D{uTJhY>}5;u+1hS;7sT|Dp;fh&_paoXcSj{xaar|EWmtJwVv_ZdJkO+e
zy%sa}eT-p0kq}|wR%`CVku9ZNCUo0!2dj%gZGm*BCaC)?pldKsr{O*;!z;0u_YYdF
zPjNdM$+N>g*?ZA9<IK8l<~RRL<{!wZd6$`5tmJR#yI{)9#i}}i%872WY^!DJI8V*J
zW>*l}(va!6{6Gg&h}pFTZ312^dRwK}ylIhIb5(Q4eEo#=_mq>1FDXSvoDf>zZQf*P
zFQ_wX=7a6*%}49fb~M*7zN&V3f!ppZryr8(Gp9YYd2=nVn1L<bR<Uu?yY{r>mmW1s
z?Ag*_pIg2qO8n<u-_YzIt2~RITYpaO-oJlk)z6DkcT?>C%{ntd>0gbG?BC4Q6W@F{
ze||LP&-<0vzJ<-T%*t;qv#Oo*dbZSho&QI-b|!zawD^8h{`!r=dLzyK^Y+CbHE`2e
zz_VfVCC}A+wuIdYef~Omb3@esn+x5_j+=ffWX@c2`M9OYYTh?n*M`Py$tX^L-En;p
z>z)k_A+x${6JP3hEMnZr`rK@(_|9WBla8B($M0R~adx6c%iM(<ADT)krvI<Z{$=OC
z`&~<=+m}24f)9m$T(RKn&Wnv#+vU&Reh_0W=q~qTL6Yv#W|>d#WP_Fey73jQJN@(5
z<ju04wrSGO1>WAv44l9()NIWyv*uEf%IcS$2SxwQDc-f(zh{35sPW{Tp1o1+A`_^)
z&?Cy7V8MP<tW#I=!B(a}MXDjuX^TJR=G?rz^^8&`gSX5V>($dFnQuoH&u)vY>6;;9
z7WOh(eEAdh4cUtpADF_K{Au~&10ns5X;m_>+8b8su;!j{Tsb*F%X3r90^N|)8xDWo
zzEpl`zd0)_tH;j0?+<w2dEfY6w}b7ms=rVD+1W)$P5$aW@W1W4TEwkrgV(v(hY4FB
zpIq9xd(*;;a*`?cyS%=qsjyA|`+D1!zkj0*&DQ+^MVH}8nc80o?|#L<UpK#ciQDYO
zFE_SK`h44A=MjPRK`Zi4cUyeEQrg+7e6!6keA}H}xo5Uk&N2THzCJ!-Nu}R{4H9i>
zjpF9v21zUK?^Kz1wB@$>mXE3vkDTs%eD=*+S<{~D_2RkbqTOx!s<>~g5i9U~kZ|yv
z?#?}Lr=}#wtW9PN>iMX8QNsVWs(Ox4O=aTu6{-#nSv&XWeocQ{aO0fkIm_vuk0;uS
z-jX!`zuhG7`(a@_g?fu~nR?f^ZaBL%w{6RH|1C@Ell|7znqG8WTp=%_mscrT`aF86
z_Nm*wh5xH_A5H)6(=Am}erKDI^=+vj&9(2IsPA1>A);FKJ0$v5{5Lbb#y`oom;ayq
zd#C*@wz*##E}7MGO)bqWo3Z1?1OCN*i;P?hYDH{$Hfn+*xI;`LRjh-df;S-Ah_N}J
z#~^pEM8*8{B2UJKcijo=wi_&6IY*V{`lc20UpJ`Cn0blqhwTpjz>Mp+&I?~KPVT5x
zOMKP#!n?nvbajf(F6moWyF`{vP;083*R%5&3&Z!x1uU*-+lo6M6`i-2ixi08m*#fG
zeu~jeQOlDOsqg2+&oVX)Te-TT=-20{#ja{o3M<9=CK+t7l@L`ijy?5uL*2_O61pt=
zn0MXZr~g{p<iB=O;MTHeCbrM225!5p-lW}Lv+dA~`*T*xmN_OdPQSyXG{Meh^5fNw
zxi^B3*Criejwp#>*q(QfWu<uUIn#Zrw`wZP#EavVxILD9IGyxV<-A^C@4Qt%t!BK8
zKWuTY$#DCg{2E?^*rZV7w1U6KpRGJ%SfX#Owc*=(^JVWYMCWW0p75*UbIR*KYHKZL
zmF%B?^RC5Tee<RBWY*rT{2K9E#y{a#!R^{r`>Z!#PT<weHB7kYe!Ih#{mrK*PmG=~
zUA0ca^K|#xUma4thK$AReTDCDGTpm-H{im&w$d`mw@o|h(?XV(?>rKce3VIPhF#da
zDL4Eec`(M0`-Vyp!|jO-*Vi+hu;pd2=G~z?A>Poi_0Aa)As?|ETaU$QK}Bj@ZQPd9
zy__4X*O^9ttWH)=uxVgrY2V;b)0CU2p)4V9$@Vqj_=&2^0u6oHNk{wHl=!7AX5G-r
z=#$v9fuT$!;^pHxAuL66b&FoEJ9#tLsUw#E%cZMUWm%K7*F+!p%RV}-3)GF;lGW2=
zmn5KD;2Lot$KtTcr(Pem$i70ux4-u6QI1}9Uv1(-*}cz>M#LVv^-!|Vb~D?g|BJNE
zSWc)e%iiR4du6kB(c!4Pxgvp^KYg9VUhQ*j!Fg}1WqoM|e`_||75hh~f7$eA!|8W3
z7B7eszqP{ja$n-MDkm3K*`p6`F<p4|TuioT(bTO@Mg6a@B)y5)+`zG{f2+FMw(u<t
z;Ks*cCX+AQ{TVX;Z>{Cu=3*hgfyrdL#0-%)($dT;&h>ckD_lNyzN3?o`M@2q27iHf
z?F(c!Tv9FD>$Xm(E}>s}E?4*Qr4o=ztYB%>fvl~f?*Er<_~Z3<{tbOdT=aNzd(4p9
zbH$4%$J$_>>ZA^)<d{aQZ>bx)m@jOP4A{;1HqdzInx@Q||BSXg+Y~x;y1<gTtS9$3
zU%j<b*UH#%(@us?(WBEI^)2bVc<NlM<ie~;5teb6=T4Abob&(Sp&LO9-<dL=zU^fd
z{6JbQBe2TjFsQfoVXjHT2}jn+m(QK=wtoCQq(se`J+?P?{nc+$cmLZfE51ld4E9&s
zIQxIyf=s>xU-a)Rm#bJFoIf#F`<jZ(%4=#%TC{hC`S+}uEwjGF;`@Sc;iWS|Y}Izx
z)upy8))WV_zg18B92D}FGm)!K?zzUF=0o3fc5zJduF9J)^)O-MdyTg}Yu}o%te+_s
zdwGQ^+rQTyr&Zo4Oj<WrGS<pE^6E<s2K9Y6azPUc3#D>xWKOdAcynacCX|&-EhvAv
z?NOh_8!O&jl}RppjT35jM2Q!!TWS4hR?mgBTbi0V%=LcfQ>zQpV+v>6&3bepO)|f^
zVCk71Tgn;R-X)4R7Zp#mTDl}uUQ$x!`h82)MVIED;gz|!=`!!#nN#u;t2!+9DLcm`
zI<j0|zkyw=%pIBvjxz~apU9gmaC;9w$McVc?u%Vo?j&-3bb(0k6`K&@8n)|)Q3rUu
z;6c}BheJop%GAEFbo=j_@43W2HRxr@7sJ#kUp_OHO`6B{^XdIJk6JgcDLcG_>GDg-
zB`z9P5(>GSr=H}j;0r#bvtB;`?7zqNj?D(`!eY30t-io*Z%2LH=b)x|<%YQoH%#A8
zi~T?Kz{X#%)uT2}4qaCHd*2&=@g*yNdH%kV+0$sgC#)#Hb~<O>!Z#25tEEcs&yn8T
zzrU=sr(ym4)j=Vu6GCU!tnjklxh8ghe+|P<uh%l$zdE`9+10=L-IbYZws6njKL4h1
z^Yc9?4ukh$y4rm5{JQE<SnaOAddc#0L-m*N$J!fc|5}}%?`eJR{^S1N+Hd(xm#eac
zb2+|!pTBO`Oqp}pRcmJduUwEh^H{>eXDta=kLUkB#*+7Zu725T!>Fa@`-(rZy|Rus
zmz{5W;Y!J)jeE*}?Aq^LYZ4Xe8~bdnee|@7xEbG0=3YI0-t^_>FkcR9$Dn@Dz=dUS
zZA$P}see*M=8E@!Y<N_+OHJOp*Z=Gv^|_Om{nEEvzc(`De7$VmwtqF5htloiCPg@9
zwj{o8^^0O)DD2s|dwG{%RO_~#LFe?}uX-^x$#tVVq`nf+eNfMR<G_rG4eHKChDp%j
zOwlYx3%=DIKWiVhYHz7d+78LZ26eg@j9d>gJ$mA^iorOvN=B#P*PNp_ww=`L%#e1l
zmb6SxjLB90wsQ$%P*=Lb_adI-!9GF}CzB>QcI`~Rljy8wyhKfKN<ojG!>$Y8Qgho<
z7+%V3ICES(rR>_zx%sg%N10zHEPDOO=}GPa?weO^r+qMe9~a~Nrp%6CHox}sBUvYf
zxsP@+Z29-t;WKZ^J3WD;f2%(~_PbDP;WPDrqW@!^^32wcyS=TG_k9g~x|ik8qS$Lk
z0xG$GozR_D6SE}l!(;!|&jYGjdnUgXNh_VhuE}=q^^1)s^tZeS*1Y)XOu+P<*Y<C7
zBaSk^oYMEk$(?K8jkopXSI#|Z{8Dj4yw6d3<(gl9!8T&*Z>(p(J-X|^{5^&{_qW{V
z<eakWt0=>7hiy{~tLFLVuHQTHjikb4iOmc4`=6J!W?Rypy(?C)FL&RCgN8pAhZ=ot
z`tq+^yLqeU9(yw#1*6Nyj};3=Tb|MpkkU!FJgFtNE@;CuK4$Ip*A4ehtb2C^v=*dc
zzK+9fuG<C|+RYa5Hj7Am9{M6Z<JU^LjS`SjG$xGOLsu+N`|l<@#*IY~v7QO6YnDZb
zE$&E;e<rN(OS`PLPHJ=1a)!7V=4HRQ9?S{s*nV2QFF^QLS8Kg8r$bHJUEYN=zplAl
zFlX+M#cT?@dUWTPh|acuYGb`=;d~kQ_<P!)-&CwMn_MNfoXPpFri8AF@5vjJ&P*_V
zEp}wyz1of1s_JSE<r_9>e+t;D*Z=OWh2ftY_3uyK@V=zKUM~A2Xcy{sVUK@r8jtNs
z4_hi0FsnEHwy^fqwiK)RGetsvy}xZ#HZS`X=l{3Uc~1+>clj7oz^r;_lK4fJbHS%?
z@oLT&&rWM_tUn_8<9YXk{`!0OE&6ZCi)p;AdcJQB?~!Hd3V)vcpLWOV*K39=>wo>_
z)l|2yG^qAYUang3_4U#0i|+&so|e}am+2hebgqj_LZN%pKc8x?h(BBkSFV{FFV_C~
ztHr(frr3%gkKOg9wU^8^=6WPZ@%enenin7UW|qrSz4Khwe;?j@6I-%x{U+^K{?$^?
zkG8Vhs4WV7VsN;3veGO*Tce-%6J|c&Te>PYvnA--&Zi4G-v@0>^s>J2@8r%tkxl%D
zSIzBb{C;*PWNO=t&p&s+N}F(}d$n4EVgvK#^<H5+V@e99y7x#&gE|ZwzwSA5EiR4w
zh&$V+_WY9x-OJsI^Je~h<`CVzcl+jPd%wDx{?Yif@loKGhK$C<hy^mY{pt^I*>%@v
z=^taZ|3`QpvZ)@{2tK0~<p5F;n8xTCd*%JPd6KEULbH{@f{YiL{%vRoF=P&SWK?$d
zGUJ+6k5!B}r(MgvwDG|KuC&d6*cX&7nrosolYz<8s$xrF!kXi^`X>a)ObLH|#Y1|=
zJE0Bdb62<@$_kt&Rgh*nIm&BBXw}3>m8^{LEt}G6xwUn!Z~FN&QRj?Nw`6L$<JrHJ
zQ_S)Xe0iV0YF0S&G~>t*t3pCsC#&xCX4+_eXa3%1nHB3#^=2$Qk<K3?Y#k%Cc+&RI
z4pv9&v-5JE9xHy8^rGZ>ZQw1_mor0qm@nVAylU3^j-5H7F5UOrb52Dr<(K8>J^6n9
z*cI;mF8}>Di!Zk{=dS*6;mwQ3`|m`~zr>sNCH}kVv%mMdo;ZcS)L6H}b-h*G>e(L8
z`_J^>_I9wE>uhxF&vfa_W%s|hoiz76)za{h{rQm)cQ(y41+hmJB_=i&H3tOG{lv*!
za&hO*h_sXh=O~v3Z-JSY{+FeGN)Op)@P0G<tlCY{+1qDM?u?ux_(+5O*7LX5|M<Un
zeyHegq2GCC1uOOolXd6h2CTc*9&8(U(1@4uWr-Y1?jQSgezva9&n;ZyDj&1hu;~3Y
z-klGXX1vQ;p#3DOs`ZQWo3@Nb!R-xBdWq-l?V5Gx3SQ>-Sl*Qy#2LNYXlm`_V_yXG
z+anq?8b2m7S{R$oejTxWmF(n(vM)QcYraIkSqf^yJwC}~q%9oz^6|^Y?pje@g{Xv@
zNQYxeGp<bITi=$cqh;8#)gj_(#)4UIPwqdk#kYOtkt!A^Wp##C^Vxs+E{I+`*Ou=>
z^dT#TwsOXl<L#4eTONCKes({7cd<#Cf=r*^#Oa>Wg}ef>L2Xt6+y1az_U2zWZ*fj|
z9jA6k)(s<%3oNq^%@oX?9BDk`i1*7L!(S}YR=yYGtM}*JT>D=)*7WK7R{nkA7wULR
zTc23fZ8!Qkqj>v<zQpRZ&{KI#ms%HY3GHVUI5sV{El$8nd8tLi-l-)ymk+MnyKv84
zzE$TucvdMW%<cE@4tZ6-(WX$n{_V>jcUd>4UOZ@I6MpOgLkXYd_7d@L)5?!d{KN5p
z^K>%zW6Nol^QUdO!zr-rb<x&F=G8x|@5%UIy?yiA=gW6CuT58Food1SGGyEOGG^^L
z#Z$P_{+l$PeD!+MB=fr;t$c$r7*gCNv#-rsd`)~y@opZQ)Vlo(bh*E)s&E!x-?Adg
zKl5mH@^vO9<JvcN)*GKrzrmDvkV{N3DSYOO3$t$QGhQnpZn!PUPbZl5=Km@0VkRB2
z`?YG`#ahD-F^OWa4PFLcq*T|crTI48IdSG{!qX)e<)<7xH&f!~w~3qQrSHp$;5p{{
z`gZoFeNs1X9V>siXzlUupT)oZdp+Y$*vj6E{;{)n<bOTVk@j$}_D`9pmuDZIpYv=>
z!1^b3UA4lxpjza>j;#%@O3wEg_nrE8Fr>{_=f<|#x*KIusv~)=KW^&OJ-t5p=~Q1G
zyT>Bjh6#%_CmPo_@Hp(6bFw$_j2d^Vj*rCT!xI;}coi9#yxih%qiQB4Kbvt|b3@Wa
z#x42BpKO^ulbJKY+<4`sBWK*WA5<G>XB%(mJ})?do2_Wyl-O*qRI{V8K3cPTXFNT_
zbZwV!StO(PmRuL-_IpC#Ts_s7i=GR2;dB096ntyi4S}N#C#wq@wp6?3NgQ{3e);mo
zwZ~<=j$h`zyOMv6{(E)D$h(Oh%c>LVR-a+m!`X5D#aXSc*+TLz+%kWiN+vWVI~Pxs
zdzKJ#F7~pL|DKv9U#l2)T?DNs`g82gi>|uXuW=z|5<9+Kd|Ga|F6PagWw$e)emN#>
z?W=h1yG`t5KI@wkF5Q_a{&HHbt-a}zQ=yjoZsmqeT9p;OQSF;dO6D`0`n<ROV(M>{
zzwKYW$=xdNYs-hKGyJ}4t9?J-c($r~Tg(fKpRYwdj~m^!X#TruQ}=<AY2xeG97^i_
zJ+*&g-j0ZW9Z?P}m))=INqQJQiM6E4`0}?IH$j6TTi1s2^?o_eXSzwe-ez+0YW5RL
z=S=(c@e$*u?C?pNNA|_66q}H|BX{#m!wLKU*iP{<+9>;VyN==736VE?6K*bX_;&Q3
zUguO9mW#*VKYW_3d--96`r(|d7oVhGVdrUKuG_J+%JA6zZT0sjZtUMy(D%#jyIH}S
zn^kLWl&;zOIVaiw#J77V*Y3Jql(J@Py~W$9Mkk;D*!K2slC+=aboKXZyM-d0rc9c8
zE$im=<x%?n%15Hwa~VHc^4)H^u~q79(4(e^(qR4*X1Ry;19Tp8ZEEP<_bk&#vb|8p
zGe2X7O!cH`_Oh)n%+jvCzVYPd-JIL}Y^PT1oaSd-ur7%qZKkgjPn)I59_A-LNgMkE
z&oZ1`E5<$9u;EF<M>Z`J9R?x&`Oo+->|)DUpHld9qfaJR(iw@fYZ!$M?#$<VR;6d+
z_4vZEBDL*3O`Hc#*j6QR9%ngi{UGGhBq6yAW`=GzKTTfBnckZ7@nF_<M|J7>30~zJ
z&ra}?53fAHwY^c+@7tEl>dVJhJ+WPPsY?3$lSwVb`ExU(uOx3{-CMY+bepxo-i4=6
zG`?IBzin%T@%7lBlj3xX%fHWkY%aFavSEhF(e-C}KfiRZ*C_k_m{+s?e8Ry+d_gw6
zvCnp>Z@+r)>!*v;W=((Zao<eYv(&R(=eYyhv)|mw<+i)`SE|Y!EGoNOxv}_I^XA3h
z9bY}&av|)>^+js6R!4Qu7Q9^&IwQ_TemA&5@cWIP=jo5m2h^6fZ|eDxAhu@8#%+Ge
z+6R}cTz7P>i`Z!&p0t~uT4Ev^$!|UGUvS%$`^_Pt#JfGYTv*)Gdif@y(43$s4YA9)
zZ=6cSS7*JETVN~K^XbIPbbhfL-r1!&Nm(4TCkdYVqNZ(Kvu^Iza^8r5ZyAewS`x&T
zO)9P}x;o#Xb&9WptjM|W(!vKnW&L^QJP%*A_PE}Pn@yU5F}IFn<ZYa8{r!vl%}$w_
z)2_ap;g{dY+Y}kc#=)WRL~L65rdQAY?cbg$vBbSa>&~`tNe2hJt=HSWWm!g7t%~2H
z9+l9*niD4L@H)&OhV9O-FM?)C_Y+QQHvZNTIvlub8+&I<O8kq**By8!O)kBAGel!s
zX516*cDsa})ZGXFr8aofq~8ou64|BtGBJ0_^yABS%ed#v<T;qoXTlM=_E?3El19hv
zwiWLTPnyqr>#^|1Y&Ewl>MdV$wU=7hT?jjxHdiy_?V1%{mwcW)cK`L_+%@Ic_ftdc
z41d2iQOk33|2rw!Fek6b<MtJn-_{0UzkZ#2c}*rOdqT{m;_FvV?Amzg-NS9d`@g<g
z@u=zWVcq9N&;H*k)zne5nJ{<l3mFj~|3f!kEL$zypy+wy*WIml{WopC7GK?CVJ4aE
zk#+o>FtbO?yiI*w)zf$X-Dh(=FmXfRjSlwkr5jDoOwI69XnMK$17mNSnx{qRtEDqu
z=pGE=70x`F_+mzWtALB*Ws_J3;ni#=#r}4^FLxB)ZK$?$ZRpH1Ro|p=vZm*wy3y)4
zRzLb<8q73ycv?MikeOUM|CDTW=jO%CufMIIy{FJ(EnCh*jwQ?O48v|d=n$CW)5Dv?
zJ;BqGgQaMLSocx2=Er`CWsX_z{W9n5D{_iBxJ>CRlhVI8>n{}P{Ak?y_G*GzdaSMC
zpCbo;OBg@ujWE~WcH*jSd7=BNw?g^}J-btv?mn&kX2<q2Z$5s5ny+^|-`nq8n|E*8
zXW8Nt3$=9P?tBcpIZL_UukP^GJf`xmzm@;3%$|Jy=~33L_McN92Nh{}zx{cB<F(fY
zf>ZuPE|~i-FtOTe{pYCLvm`It-99CsD|&i|2`FoLCqFlgJiT>u(v9u;=Mpvw)=G$#
zwj8+U%Pw_uzxmYU>yh!-J(!+$np)=kOR-*Yr+oSQHDU}O&Hq(g5}Y`DQPTc@&r<S?
zo^>4YjLqi%eLuXb;0^QJ_04lm@xOaip!e;)|EZhG+tV#K<@o0()knTRuutaWJiof4
zcs}+A<;{h4-}U>Agr`ZiKD_tzWDnbMk$oTAC$HJI)P9GqkM@*OL+h!^KI%6AFYj78
zC+E!U+p8mkdM3r%voB$Av}pHycs|bVdBOqv&joVF8D1Mk+=$(8bI9k}zqfnSBjc;<
zKRGE{GdslbEMl$M)ZiM#_~DmATh;d8Q}R=UF0T^#J2Aa*4_~6J*lo=k!JQ>Plzg7O
z{<*AJ)-yr4AlmT#Y`IXkpVC{GF?a;r@qfej{L_hG+e@-PKl`q%jn6vQsdxG7<JsBk
z?gg!x>3;QpZhi9Ulkfd2xw%{Z?l&oT?;gGCy+K4yqu0!m$A>>BYwLv;$KL!EbhTFX
zO7L<+HoG5=N3VZ$X#Fg_F58hS)4$5a;j;dE|2wCxPfMMxy%)XkY2A%<HP-37tK8j-
zZ?|7In>{6WdXB_QS+m2L*FB!9e)}BjJ$dJ&Yy+`7O|iBAIJ5+8>x%OEt~ILWJiM~P
zkbPPI*KfZcOLnDph)INtEzs9KCg_sZm^<&-6EoS#+PiLS-G9hy!CJ1u0-3B_#vQ$j
zLwFC!_UIX3Vkz6quvTzQiScHZ8|CS3UQgEuZ!%c+fMMAMR>Nq8FE=9%j3bp5S%lfs
z7jIcD+#(XKwfR)?$8O=q&k83Fx+Hy9FId2M_)y_jjkZgXN*5irEuQ_1`})%svH$<t
z{;f#~Pt{r?)baKPQ&ef%8!m~sA1^O0xg32i`<amSulV+;v|U=dh9T!RZZnj-vcEL`
zT3mQ$P4DWk*HcBan3s$4zm%TO_-Nsc-di5r9P97QEEM=UKjypV*H6l(yKH(73Vqwo
z+*`Rnerl9M<_5R(dr!o&8JFet<m@QU3SqW;zrSLKCijA0T16bc|GoUi>SXZua}Y0k
z(9$+`lZnZfC+1F!`1N$lfd<bTPjh-AzQ5S$U?Fkaes}#L28Fp%9?1>2#nm^7hJR(R
zuP^c6=B_aJ(=3maJ?1`M+==zoCASz9=H{s%XW3Z)FTv`#%=62|&i(he6({%w#@^j0
ztYuyL&E}w&<?MNHKL{QQKmP2wtJOKq1K&g@CI7zh!YScyu$+`<pu&NBk{-){XwFQ0
zvyxrylBRgaTK`6g|C@`IcN9OY=l=AjTVl(Nhnzpep9!9K<^Hm{Atdb7lE&GLA^RKl
zm%ItyR$2G)cdFb=ZokK@Z)9$3&(O$W>Am{C=bV3+l|@^Q(~3Q<->=+TWy;RcSn}tt
zRhpa?&w@DXi@W~ZmJijFd3o=5lpWiwKE~?g>6YKl|CUm#)>{02_RV0e@89FBE=`*G
zXZPj)bG7R-dVh!1mdD)O^?qH2*$mbzwqmW?3Ufb!_TwFis4A#EdVT)CMR~sxQhCen
zUr)D=s66z0i$=huRsKtNgzkK}Z*2!Rw^a4>xrV);{wA-#>72h&Na#S#g4m8j9}e%{
zd~^Oa-^BR+cCHd{Exwn`FB81AQ8T07a_Vc11MR~8r9T6Du6zEyTen_YOc=BUp}<nt
zL4rrE*euHD*!wMSvwHT5C$?0Io;Y%OqpkJ%y7RO6o<(tSbND0#CCeDI9yQu*#FkuQ
zFzMsPjS&I13U?RvGDpliq0(k@Vqc1Z<SEm|JespR^*v71+{l=-i^0LFm(}=yAOCSn
zdr{XZ5;YqgrUV^(D`s(jQU6!9_>2WpW*?cK`Np%zuYOtJ^Lec==15#mIP`jfr=PXt
zjeV>1_NX&EXYFXpmT$PHFzJ%?jHCpf|3(=+&;IG0y0_%L_zL5UJDhI2Ib9abYE<Qs
zd%XH?i|VA!*L3-x<{Wl%G(LLxd>``)&SQ%hk8XXX(|31P$=#LPPs}QL)~LXLw?wqc
zv2ukR57(*n(E{;&H|NTR@G!no*?m&Ha?iPQ59UvQW!*d<v@e{OyW`B4jUVUeRcik{
zXxn&VHor+hv{>(r*(J9Zx#Z2(Gkvj)l{4pe;Je>FlglD4uI{_skrG??JIqkoq3&&0
zh~KW3OYf~No-up-{x+wpf|32sMb+=$t~zQ~s(IY*eDfBc8D0nTVrQJVckkh^-?MJH
zJymsK-K17J|2^NT)3XaWE*X4QyS=}i)51RR(!&#OI=?wJzulkrXqKK|)X~DrFI$#G
zeY$jbM+DO*PoXr8hb$%6?5*#dNPE_@IXW?&>oeno<xclnBVvAY-QJPiwO{JZYUSsa
z?e|JfE>igKq*Kpz)HG^;-u!1=qW`7k#coVqpIlsb>x2`tf1~JTf##~H!l1-}^-ar^
z?9c7mvTcFARwv7Cr<^LDKeG*PTAclIE$Z7-*53WnMO(t$Crg!;M$JFcy`fM1y+{4~
zd6|J<b)zOf`uce7&HolRO6K0n3C`0BI$vEA_-E!to%J8jz5G_WKVn*hifZgyK@MH3
zWP_WF*4|F}ZzQv%?_iFtN7=`12blg;N8VW!UK_cr^M3qN@4tH@jHT^weyW$vixZU8
zx&FVRu%=U?+kg2!yXbhX`sX3hYPw+$m*3x0Ru-~Uzw&rdRQK6k-k&~Y-rw@_SW)}M
zZ|k|DEO#Cdu*k6wzghqMj?;-JuC1&FIj!mM->s3o5bXcm^HKi!mtjBdZQfLMFz*;+
zYlY;)m3sQ!A2es~zx_V=X;KnsFw?D@)xbiz;ZB+8jF$`t)=JcSdsMNyOV)%MYd=Xi
zv3?@&nvyLu%axA2+~|EE!9C33d6?C~i)?XLyg$|i1Wcb0XMN+fQpLON>2rk_=d5Y$
zGf(kXYc$;3-w@~D+B`Sr{)(%|=H$LQr;&M){d3OK=Bcq0|J&W!bZ6pq@3hnWVn^p&
zM68{ANn7>NiF5OHPI6>V^Ln*bBHr`=uJwt}UVl&F-BqzZ#DiC=*>uaIvoD11{NHrw
zT|jZrnoVgd{>`k}IQwGK)H!QjbH-0tR>WZP#mk-LWa$O}<KE}~^BVX>r{3+2GCq7q
zN!ZNd$nC_l)v4MWr4OXm|1dd}+!Az-(QAok{Ka+rLRR9sE3UULbJwbFsrBz|S$zMV
z!t1N^m#5!awz+`e^4iHtHEAC{KkYoeF_BTwM_&5;%SqcexhAL7*KBW@yKhC}T8rn0
z?d`OZH><lCTw+xhn9aYW@?5*PdH4n)hFv1N7P9_(lU$Z0lR10-`fbN2$NjkDS*Gw?
z=ZVL^4|$sN=h`wW{F}Rc-J^A$w~mU=@JsHD|Nosyx^J;R)8+aCaS@dU!E@{@{;l`d
zlF3z=FgM*w*5>xRj`C8cPg<Meo<C<<QhnM-eA?FEDctG#N}&_4DfybMpPL*1CXR3C
zT=VVM4YKq%uFKeznZ9tkW!DVtjQ5NU?~(*S`!97)hbP{UJ+bdZy=Y(8wq${QEXhAZ
z87_q<Pw6|@!NUB>TcTr+M(SLK&*h8$J^tZbJ^k&O^{LnY+DwX_E_|@eaR1&nzfbL6
z^n>wwY_PIk)hpG^FSa{%J~GMhMb~ZK{V(j-=9<?b*?KY;e#BNC3%1^~)BW7Ej170!
zn7aReUEqE0yZEu{H?z%C`OJIux5?$jKGw9~IQ4^O{Pu>a$-gsw|D9`_?He}x(bUCf
zx$8F=Y;&=5;?_L8@#DEo>ts&ZTbiwS6_oXjfBD(E6Svn+E`RsY*)n-mNAOZ9xA_w}
zO!8z-C$8CEuX8&tXLm+pU<xDC73N04q$b@b>mSTH&MSM-#NipU>!Vq_ug`9<{!mzB
z_3?#1cf*2$N2fondug&S?rYV{S@ZRM@6^WnMFdy(Yd(0qS@xy$xmD@sd^y)uynVNZ
zIpq1;N<q~LTW04O+-ADIPJjQ?-{Hq)v)*rHvD>}>W0caOAGh>Qtl4s)GAp!jwcXv&
zdp1AL?bLD&t=ZycUirE5?4EY>u&Dip*1u}PrUaF5*romJ?&Vi|oUg`CKKC!<l=q(-
z+zy-8d1m)n#w?!uq1~4E>xW#Gxw95z?Y7llG_m3Bp|*ECtFjB-PE24-ihm=oRB{b=
z7QnXsG19f?t5f;)&U0!1`1okY<jca%*8fXaSblffy!X6t5_9y=edm36)@kM%F9|J|
z5LP_9ljqyYiuYTpZgpDSb+PL;KfWZO)p=6t-Q!oU+nHGJVqmB}w0Zjy+5aD}&3bAo
zdu+yg-H!X^|Ie&A^~-PZC;s?r@y<^tIfupVuYJ6;&!>x<cfzZ@IU?6wo+s@n(p6}B
z8UAjwY}omR(A`(&E~zMT@Jg`Sxc#Sq_)YdT<{UE|U74npvCMc@qwsuT&5x9N+xk<o
zvkw~OXI<a?LqoM&e2TV@OV5vo3@<~(-usj$Woek4ZkFLG*yaAfN}jKh;qu~xOO#GZ
z+wJ+<nk{BqfB))4mB@cwd&`+F{dC|dmGe<|Un$}-ZP|tx=4Vwbi>pqy9dTNIz<bYy
zpN^1B;KF@|g*{EJ;mh`!j+c+AJhWc;)Z^Ea_086^+Ba4SZJYfp@qiuU;%_N3&4SNw
zTr)dj^)n~o{sd>WXx&q54s7f^sp?w5S0I%$hb`^h%apcHVq196q`yAv;r;UdLu3B$
zJN9ju%l!2abL!mH>TWl--{A>*VZOdgv;LX4_2<QI$t5|;HIFvWx+!tN^-o#qr-!vu
zV++^I=Fd&ub}Oaax6{U1!6yFA?d7+EK+E41izf<*>RfNSdGEuILqC*?b{t+VC|5qM
z?AV^@3w9e<thCnDbGcY~dP#?nfs4(e;LVep^~FrTE-7ou{q*YNwaR@>FPpbcG!gk+
ze)YU&wsFzw`h=}5HObCDld`sJIi%J6+xBJgA-fAZt)mZz1)hJKKd&^@t3dYV%bS0F
zcg{Y)THSeikfo~Kj(4w(7j>;a&s)qhZ>3IBgU#;!mEW$oq&`;u{Mc*X{+(5ZbA(lo
zw4N|se}C%E$>n>zc5}vA+_BlXV4?NKu;eZ;j=&Nh-vG0}6@`Z^>ll2Vavm;bpH;MP
z)rDo(Gk7lESaf@{j`o%}Czp0UuWS#hiL%LlTHHQMZT`ylo|j&nJ-E~G)UvC>hfeRe
z)7~}B|FFE~+x6vF&+FyB?R;-`lWBwK_P-A+H?p34FXOp7NB7-{weHc)%qne_nTk{W
z_5H1cmrbcu-RpAOpIfTBucp9l?+me^f7`ZaPWo_n@!J*k?e|pFq*lI?%l7bSaG5VH
zd*#pR*OUF0xUv-RXy?ygYo7Y?uejV+w(d0>HqM@1qI0@zue8gIe_oc`Z>@<>dlp`m
zva$E;zDY8&Zw;3G^)X3X@{!we;`!#w720f*T2trFnJ+cz1UGYK!<<P$3hv)0D>8Z<
zWMFDzysd3>S3iHl`RbD|ZF{?4Z8R=gVN~9{wRL^B$vO4K`CFnt^Zat`dzJpYtvH4K
zmSM!*l<(`E-!7cnwo&lge2-g+9(+Njdxe@~)T{p&U#)vy)!|`iYu)qttFyJk$w1rQ
zWOt<tH)eC_Hx}ght4vH1NVM^M5fH8ZXlJT=tHI)vhrC{wtl1Oy{KHS>9Obu*dU!Y!
zp7L7xEt+_2S70L7wF_wp@rmt?8|3m&J$>s^I%oBpq;)BhFK<p_j+xZ5=YsdEg<rnZ
zU3z!kvmnm<eD}PIN3C?TbIN}T`$X=raZ#8x%k%yA7G>FqmjmOSJtp+54$t56VDF2h
zh&j(q{%6)*y_EPsY+lRx>HmK(tm?XOK2QA9<QjX^gp5n~=Ym(XNJw()oZ$!+KJh!X
z^0|*x^4}MmXI0P1X7YF^!F<o&c-OC`hjp*A{Ft^?I;iF4@?-2%=jOh8Saz(*;Yn#l
zMctW=>UOJr%cnm~_ZH@7-r=v#IjJ}OYWU^jmb&}gWK8u=@n|X+%u~_%zi0bLH?Q9p
z*1WJ~Syo@PO}X>NEs^r~TUIDGzub4jci{v9|9B^}W&Bsqwx3X#`D|lr;n6=|%by(=
ztL=$B^dmAXNB`v<XZc<1YpaBhZShShwp+XWvF>DLsd+b|F8i7_gq&9SC7{`Ppk${@
zR_5z%E<&AL^XrbTnJHM>(r*#5^z5SIhX10<Tcu~L_qK_v<L*fK^|YRC@jh0|lUnDT
zYh~0}R8AYQPFCbH=~&XJ!T3fh_kP|}$r&~wU-)z~9rD66CloY0@7v7DaO$T6-^z8z
z#Nrbb6y`QDHG4C?oqO{hi`j2wf!A{mlyI>uxa=_NM0N3jeajt$#rEmPv^=nX_51w0
z^G{9AXwFKBYGe8^d2{HjDV)r_Q!016wtY*S`ukv%y1`r~%UenYbJZ4_OkC_R*`rSB
zxNFD%PO%S1nS``6O%G0<%`4UWQ1r~46>04IZ=VUX4?FioF!o2Y|KV@JKNc<DU)3q7
zulcV+#bn1MW`^mEG3~eJh(>#Uba%P!yYoP{(YH%kDkriZpWB!!dZ(OE<oTAHoH2)b
zLYf?Bu}l(~6tzHadBlQ`EWaO>cbuQTp>w9zw5	)h!b}|5ybddBC1(#J2X!K3#c{
zhl&ClBsHc>c`S~a@n-X*zLzFD_<nn(-6)B(<>yYFqq}W-))$)vKj-#N4X-Z=+k0Um
zzXOw@hn@9(!^e9~BjwL}1bw~StrQ@w@K^ST$ub*>37eTiZ>_xW*!sQ3;iH{CN*nJ@
z6R-TVHaOS2bED(aHFHJZUR^ccJoirg%I=#XOCl9kL@|k-srmQoW^_=$mGPpxH-ANJ
zHA@rs^(k3)p?2!6lAOf3r=~ME*px;W9)C0a-H96vZT*kgAApW={}htRV>&<fg3rQ|
zDa(U$Jq|s2qH*9$*sWQW|9^e5-k^JB-(B|arwunO`(!R!dgjl2+erD(Qa`7NKiK;v
z`~B3l3coLz%KKgY?sKDRp>h8IU7hlun4cUD-POTvud{pd{Y##inVYt3jVKnqp;Mj7
zaPGPK-BYvgYee3@`ufSVM=NDtyuNbTe#R&LI>D#6<-_z2Epf;QF5Big|LTQIuHXOP
z%;iiFx%&6`{WE#*CbtyaFS<X0WzW&oU0hxhvZDK=w7-h|-JZ4jK-1c;sI~67udh0<
zy~fe8LPN<n(4nQ^e!9h>i4%VNm8JjNe?LcUW{%m(cWIk%KK@-Y^QQUxAA8T|l->Vx
z@BQDp_jgTR)K^vK%ATD5xU8yT)){{B*L~BCr|*qPe|c`vJpP)hlD2;dOST`bU1Jqt
zRCP1A`TyDZ_Tlf|e!8oww{_|FM4Nx<)rs-jUOmb@vQ~I<?7!W{+XOS7f7u<kSnO69
z`>Z?fPDr~I3w^6U$K~ahUq4UTE@rN_p4@bufA5zmM%`One|M!O_x}&4Cr0(i=uZA8
zdB-|z&!-C(Z*|W%R`4A=##dFer|xa?YtyBFg3J3ZB>ma2>DR;SWiO82wc4iIeeb$!
zvh?wf$3DpvUN8N-^`}Aku{K`+zuEJ*z5aN}@FefgT?$V_!#*xQ)BNPqGKce*`p!Ku
z|905%>HGx}R&Ti6{mr`UzW-@wKVrUR^Tp`SFJB+cd(Ga}(_L+vS!BNR>Cd~fce}E_
zH7oVm9)H#U&l|6<!V{B|oRf}zz5M=dwwZohLHpmh+P#|nQ{EdU3GDq6;$WS)U3SxY
z{@u6q?Cov7U0xU^EI(=YiS7F>R+(Qfs44rVam!a)vu^@Z{WtFDtu<kb^#2^?xTwDL
z<L9N}5zcy@GjdCFFYmO}J}`gbwlC+mZM3mpzhm-`&kO(C{&}Wvz11Rg%?0yi-%p->
zYQN8BS?b3*@$%2z|Jy3w`?Kb0;V0vpy`N{VO_rX0ta61=LV2f?$Yq(i&jN+C!)MpM
zXqv*kQh&{t%PHLko;e+t(-s|=8)X@E%<TWA)$9L$)O}uKoc?p;r2cXdKcjQYSIEDf
zF)z?vZr=j;!+D`*hkiV{rg+-up0&l?$44$WPdU?d@8sX4Z~uNjD-PqW%Desd$;vZ6
z+2?rUt#pjbMH45@JyqbIbt=$k#*M@0`R-innDQd%@c}(ak#LXCJHNk=D*7^Ov53W#
z-&fYFyUy-1b}K!pc6d^o=*|1SFE=J1zWAwe_or3t*QN)vtGQ;xpP2t%U+zeyOz^3X
zKV9-C_kS^6aZ)NV&whJ1+YwPe`H346kM6tF`D-)(cf$qVM^1{bfBSoCUf}sk;p*d4
zf*3CN*KIfYZuNI&&+#9>wSU*_{1uRCU#0fKvv@LFpYGjwzDd7sdso+NT~L><zJ0FG
z+YcWncJKcCVfK0dl9Q_6?f%RY?%m#Zr@rq0m8pBTeg1v<Y3W0Du2rT-|4waw%wH%M
zR3O}7WV_>i{%*J9lkddew~46Uk-2(db4ueP^|I=urJi=t<sX*B%ukS>ddbbZY}<#2
z;ohC<-}_Ex9DQ4HcJB9_(z@eU{zZw_as2vGuYU9e<42`7ouw9A)1LmFeEPSE#<s2O
zOZxe<O80!^|9tt%RFkr@ie<~+mx*-!PP-9%`2+Kj`{6Re$!td_n}+Y*vo_-5>)Zz%
zOu9lFPI8)DmGYmzi{E9*mmY)pw;WQxcZ;sS))6V6E85<@t^32?_^u9?$r^fQ-8)`r
zXkCxK;#s>Qj?4F4a;9mQ$S1EgwX;{9bgBG$J1k_D$f_-`mR*}5w0DlktEFEQ^%oXZ
zez8(q6SPl5S6`^JGc2$yNU-pW$ZDPqT~qZlCVLvLRP%IkjZ@3H^!Q4eYVF3HoI;tr
zxEsaGte#~Q=JKyAduwpLSz*&w<DKg-h5hpoymiJYCcLAp*D(9GtH<>Pe+}*T{<`83
zJJE02Q%iCFF58MlJM<o&{;MjS*`4~ScD~&1PrG-!>s)LP3<*uI`_ldR`sb8M>Mv6c
z)-Bm_O3d5n`xDPCap!JM+~>aJxc=Lnw#DubQ|@_rMXRXqUgT3A>UPU5%3@#mmBv3s
z**5+Ll}VPXszdlrJ9YDh_H~tnN4#IL{ldeIK@*nNwuWAl_{4s1zUcbmY4UUZZmZod
z$O#W$R6O@ztdgI+_PUpjn^x>Gow}{_+^X=h)hUc6GuDJ~t=u~=XXT+i?|=OG@YPyg
z?z8^f`+wFKAAj9da_iGfcYfF7nTkr8{I&@juJRRL|D#cKdwc36rMWtxN@s&iBd6yt
z^en%6ig~GW#lt3XgJpAs>zmeA?Ra~?uJVyn*plD{OT4)5ZBaCHTDR2m--OsDN1in-
z+AHjG_WN-yA$!HwQhN3a!>|0)W>c3@vD{s=b$yCVWRZCu<71{=m2Y1yEG&(tH&2}O
zbw{n}v{w_h{<r+|K-{?~=7N1${`r;-mtvQ%uHL}PcT(JY+ULpQ_pMT<YJb0Wb=#cI
zUq|Dr58QA3l2F2T^3@uj4NpCt;x51a8XFhO@x|PY|BTIY$D0v>xmO?U-O%$Z=ECex
zuQn^Q+paU;osqG6<N9NsFW+P`ORf9tsi`?3f8UzzJ2G;vtNr{XUH0(%>)GuS4(-^|
zIkCg|XYZbQ`_oPSZZZ3I`t8Q*gU9QvT;43*VYl<=)hh9~R((r;>hG5kes1gW{p-x{
znR9PXko$X5pfEMc+}ZE3=5gUitX!w<zh&idebxRIG54zW+AY%0c3fWf@#dqLS>o1O
zmw%V{UlMtF$Nv0P%Z(E+a32Yo|2lhSTfLs#_0Zdr)&Eb=s&m<Ws7U|U)r?&AriClY
zPCdQrd+hh`H|yOp^<-pk|ByU7_mr~sSN(nViz9!%zHaF@b+OUh@{O<Byxh-E_{%87
zVWB?3{qy^nS@rTu^Pg@s`Tval1e@yW&svT1e;@Qe|FiARsx>c#60^4a%TA5pkGb&o
z!B=JBe=_otY5y-}EYy+q`H)?l=V~MAqw=kXfBmdG=F2YJ)N{J0x%1sULHXhbU*8v<
z+8t+V^04P!ef9G>7wZ*$*QbOCY(5iGmV5um2eIew+V3QOv9jL$(>3L#mUffu(G=C<
zZ_LW}Po8(Iaek^<Y8EPY(RNXZPyaGD>zMxEan9%Ea(*~kM+qPM(xE4J(KaafLylFP
ze86V^*Nmpu?;C&m`)Jj!4clr?>^{|7yZ7?vbKmdFX?p9e-q<?PdU{e>;$sWHy87s>
z>&e@FJgREHF8s9aPvgEdb?UFr7<TUDuDfDwt-0#sz3pZ7?-$OUYcGBO>-qanr<5*z
zUt4<0;dkGS)om)z;!V{*+U~o)`|~Q{pd{a`&%gF&7p=Fy6Ln+$6|+B|bCk17e@Q!Y
zl`#b@WB)Dtt)M1{cU|<U0^vPxCvBJAwDmB<r*b*LYgbA?86G~;EBciGZpQt!OUmnK
z-PyM0)|Cv~!(B<Xg$?I7ALZp<T(zz2iKhyC`l}U3CKqh0d6YYKG5aK;x!YD183fs`
zy?ygamB;z{4$~gUIE4D2uPs<`HFsf^WqIGPX5swl*M#$brtJT`r?yHV*En_mSD7Ut
z{;$g|r-$$FQ|8#><*;qyuZtpnc9*IrxBl!+5jKqSnOWT+cl&$S<lX$;7N6qheGU4n
z(x!7R`2GrkWx=IR+r&DpRNZE%Y$`CkywS`@qex<9-^-7ao8$@~8F*+#hI^Q;4mh*5
zq;=lE*Q;8y)(K2BRaH`3<g});Q>|xL@`=bTyvtM^W6~sbLT(oXUHGo({WT<Oh5goP
z&bigOpZ?#ww)s!1>PDw2Iy`<GcC1^X9{z0E#c;{qz>Y%OMdz(nAMSdlpC`X-PLtK@
z(3FpBG`bf2GfB<NdOxE!{+{6v<GZJBek#qoxcc?yx+7lE-?ZxlBc~S3)?a^CIxaa`
ze%2Q0XP4r%za&h0a=Cyd_vjq2o3nUC=6hKeStci)p1itEEpMsA%IBMmer&#b^zM4;
znJ2D!`c2+#RQLJx!}<T}O5d%RzVGzzSt47mu6{4|Z{_7S&C8+JQcQ!shPnOkU%#$=
z=jYQ$EuZV(4K0ylcmMT$+Mzm|%(6GlqAx{yJ}>UM=(&7`seQzr9nY==+|8O(6H>SL
z!Kr(^X{)#Yd;Z~QisQU<Vz2Y_&&U^~S9$U6mW$l8*<kMNwE8LW|BG%+n(>Q$iuuW}
zr#|1GyURrNSl-Fcr*0WLM^5y!(tG-LCbLm#<oq31w%LB&8f&k=W{c{-)Y>>L&CB-B
zzkQp#_R<;K_dc(G&0QMDcUp9B`KI6VCf<GH*t3amde$}PH}xM+$MeLl)0diHRC9Ue
zr>N~yUfxOfTC=l6|L>Q1Yqyx~P+Y{mcXG<hE#GFY+-oi=e`nd>#|^#AQEyMIzEpR|
z!XYdgRLDk%B|kbJQI@yVEO1AH!yCuPTN7-T=WnwL4F6QRp|AgV>D!%mk6t-hwpf>s
zV^@R7%C(wR^Ox@ozJ7$mu>D!jnv7>Vm#y2dR>jrk_}$F6R?>16+FMn<y=$*dvYyrd
zf4cUglT#BF&tBXW;85Y0bM`Eo6#I$0Y|&FbOH|%FW^H3BcOhxlv#duy6?W@K?t3Sq
z8LE9$f89&{igjnXZf19EUz~LHYFV__@lES*6u22L+qv)F)Q7Dp6XkawapB6`U3mY3
z-9}%Lmop=GCLPX;;$Qyp|C=tmTM?1(Zm)lL-DT#x8|v@Xel*^BoUyLBW9ja<a!S*U
z-pEa=ydio@(D&J6-#ZG|r04vZlX6M)<?+v2N9|Ak+C3v?@e)4oD`L(wUK`wsEM$yq
zUcYwzVe7D)OZnCBiGEl=^|;A**Kq$=%chpfOt*f2v(=w>=kY|#a~C3v^j@jDT{Pxg
zF0KDEHGJ7*@5Ph1OrBbvwZXL3e$yIzx80dvOV&NS={h}ff15Igi9o>B?-_o3{bn5Q
zyC|L~Kg%x8;kuOAZRu02H#wu*zIRNWt1TTRksZyIm~0vr_4)LS>z6FwJMGDg30n7Y
zmvMQs&Spb~WgGW=yfgFBy^6W#E7zGPmrM`t`+4r(<b!;>=jH@Ne17N0^={=_2|w9c
z6-OdvZTnLTUFG)f(Y{l0FfmH|yiQeg&6DSv?&T9~`pe4RNAKQm)0W+4UGJCWn`SgY
z@<`DH#nYVW->0lsWEE%BxZM1@t}54L&t8*OxdMxi(@sCvkjRYNd#L){n&KCYp2m-I
zUcA5cdDg!C9S-X4C$ee<BR5OL&fJg`=f7Up^3IX=h4bF8R*s&3eu?hdJv}FcUwSIP
zIhDC>!ovqE`Z`5#_}?>IbFiyLUctg*g?Vs|&*S;a3>kF3A6yp~G4tHZnoWl;{@pD;
z^=j5XrLEjQ*ja2;6*BW8A9;q_&Yfj3U*z?Y9btCjfkk3}1)q0u%h~F#>dsx1(5U3a
z6(Tr2Mkj3Z)2&Nqy0&C1MWh(2I;mxsc=GH}c2YS%%{4~<mPUlI>Y*k3u6$QDvEJ3D
zlgTzoXNkx4DeA{xe?7jg^rG9!|FRp|9c8W?|Jq=Dns4u~nVWXKxSO)k@tw1f-OXT=
zmYPq#PqOzs{jmJ;t_eQA!kUTN_x^vhjNha*cYVR_`S0}J9lUno&%L6`^ILZ*&AtCW
zrRGy&(6Y}f6|Sr9ElK^gs(Goi<@$K1ADOWe&%KwQf2o+C^>F9^g6fB@fA(!P`MybR
zw*CERmwmTgZ_K|ocb?0sFHd!+{?Ex@dE02)m1)WQwAW<1I@JDh-)nz)>g7AfxUYZt
zoA`6yqLNAX|2%D+<aln%%M{j4eWEX8;`aB;zt@pF{l5BW`&>h<OP24~8?)7zKWTj%
z<*?#ubyxc)qq*<={Xgli&-8f~bD?(cI*;d{%6<JT59!-IcS&{sSeUea#$UI{N+G*t
z)mIEZZJ+0{?_zbaq~DTglaCuW+*<WE?Lo2Mu}=HQ-PzqPTPlRY;ug;mUi{{E&CNxg
z-~Z>We|l>_=R~!$DO>lrbzaNfU|(wWJbbdt!OeUdjrO*b*_!OUa7x<$_OB^+$vLd&
zeec`oS*lEVx#h`Ly-Yz?JI+f-_MNMIK4)IsDVr;42Litp>blI_F}=*NOXE|eZ&&EI
z*>_F0mhR2W^uNDT&HM8;@!AdXYYcbfCVp3Z<8<N0xBl9K{hNMlsCfC$FpAmM<Kt_;
zq^EBW)xXQln9S=Nx99vlvEBPVrrj!95b@_*@7-I6+5{>W)%R=UimUzp^6tLvuEcqj
z!jTVNzl;si`Tu(LvH1sb*}ad%%So#5Qgg{z7jxsf_3!EBKEF?fD(?QXa&wmb#9L2y
zn(V(edG*^(4^Q(hX?S*oP3`o)%XMt8?wd=U*f43P=(}BlT<2&1fB$vvq80Zt+UJ+w
zXy3Unxs=UT`bF2vY@YMB_3?!vf1d9zi@gypeSSjp;)}B!`>riy|Nr6h#mDux*Z#cn
z{8e=4{pkHO9k`7jKDlfgwRwrXM>pI3OMm_^JMHn`^JPuY7kTCHo>fMAQfuxq9+sNR
z-N(lgGF7MdYSP5;IE7uuu5Fy@-OStV9Jq&-QSHOUKBu?swx@JL?;2QL{2KZ|;P7!b
zrD*?~j~V9PzyDcbtzw(*gZA4iHgf-1n80vP#h^vU@5`6;pt9FpCFhFMey#Y;pQ4>V
zPh#`ewwXG*eL)J#l$UFt2<^N#>+gv$l`5%nKd0wfmotUVohtFWtp8%A@0Oo)baQy*
z!YsOVx>C*_Ppa8+_fy8_e@P#&rCZ)I_5Yt(Zk2O#iT|^mb3H|Vc={h>aGCR6)_2m^
zwMRGWy<Ka&roirW_Ld*dZBzs@)EiYc8T2pazn1Vzd7iJ=+oHrSJFP>D%DO6~HARKb
zbFI{KpObB^>C-RP5qHs4>GGplPmhaq&n(e+sI;nUjqI7dE=k+vlpMdPto~Z;@oeD}
zOM}*b3@=v(b}dw18Tm^0&TUJD<FDS_&#SL1wLiB%$-`;&zqilUTUGw8xmY51QqOwc
zzFpU~76-+=Z)R%AD-3CPXCJ*Zl4G;=loJLwkKJbd`l;s0H;V{?oj#c-^jB=jdY)OE
zEG4qDYVPdgYO)umzI@`eC{TH!_zL??rKhJjezf*5v@d*pa+RXYj>*i7mLiw$RVgjr
zTqg55Ce=4)mF(1~IWwKQ#g^Zgd-`VAvr}<=0$<Ktt=%m9+a@?}dF{+N8-;fbk4?j-
z^uLpv>+ERJ{xm$|@sD2*7iIOD{;XcJbK<r(t80e)?1K}xe|smIuUELI-}dn#)3bt_
z&yqDJEIPuiRV#GpbNA}ik20lS{7g9*pT2X?ceC1suKDI`%X^l2rp)gT-@94=@zU(y
zb$2h-UV6QyY>DpPSxocGVom<Nuv&S5pYd+enOQTH^>;+P+?yx*eo4jZ$aPz`=1jMJ
zkpDfR_~6dJNhbAE)|$OKc{J|ksttX~xw+jd!wkbiHMdN3Tk%vawEOObq{!pkTi!@G
zCn{|@%6~k&@BPc|H^tIc|2t)9A8}9pqD<Po^X=J%c@;kYf4!O!vGd|S0|}kv60^Sl
z6Z3NyE6qJ!@J6=IQY}p8wdO?uZuj%qZ-20zHSRrC$!8Z86}UX|@at(G)?O|DDs}#B
zrFo+vhsJL4E6HC!tP8dMnCIu0GxL2$>iNRHeJ9W52mIkvVcoXaXzuTdhxY8VG%oG5
z^}m(cx_zBbI(KgD%yV|<O1GQ7(Vsg%u;P2mynVkE*PpyI`}59sx=-R-YZq+3Q<Zex
zU{1yHqv@|r&&S@Xj^mi|{hQp!@5j}<SeWJCF6#cj$M^em-7jB{FjiOkT)!%<^-;;K
zIJIh%Sa-wg4UB#7^d|GpxUox?lVy(31l6uN*<Hec8=uULUifnbPwa%?fD>0ZWpq~f
zJ}cgI!tThmPgQn(zML|8D|*+h30>s*cj*M}<6-mFXJ0$A+d-{C|ANE5YFQ0y_N`Zf
z`HLDC{Z@JMVDWT6kBj=IGKbH8`}9Vy_3r!)uIBEu-p>+PH#O1R;7h>$fPV}>gRcu~
zZ?F`YxY$~pFYeUsEZM0it8DXSE16D*y;!4=estfw8=FoT7JiVwx9oIhbFJOi^?T2X
zyz48i;C>gng?Gx-r}Y)(@xpvD({#)0>YsaT>btp+)jFwlZ(il0>*ZJ2lOGlP{V^z*
zp>|qkOXurXnYrA0nO|O8G&QiZcm%q<b5;3Pt#{EZAne4%XD8Qf>0NI7Y)Q}5-4nfR
z=5Naq>E7(Mbw`O_^r1=G;=lTZcb7k0GKpJg>M{wVzv>h9OFcUUHs&nMepa@7;?v6T
zP5p|`S7+I8nk4B^tCsn*%)4Z5XWx(Kd(Wn`U%Jkd@3ZlqmEFpjn|3X=dwl5WC&9am
z`}SAG?q7583E$yJ%TJwm-*lQSUlI0uQex}Vsrm2spN_OWr!3#Wd+oI9)v~9B?|*DK
zw*2iht!<P154M_J6gyJ5<gkHmRIyd<&4wLQ&vjKpGO<nS+U=7?a`nVTDz+}Ub!)}L
z75hDIKgo5|E;^Ss|CyEbC(D;pVz)<ZUY4V~W{dXaljVDtDSCAnOtp?yxR{#QT39%>
zc49O0I;rM$uTp<_?5{G~o&NCclKneZ9RFoKUFPNaYhUyK6nrb<l9~E%^T(6Y&tfkY
z?_Rg#S?<y|+H;%demZpUOTFx#=DUm&L;ptkPdWGNkaLmG<@4tz9(uJReY0Qx{em@r
z<}_(V%jiFlcv|-S&g5+;%>N|3Yd&@*XX4?x`=YhABX6v#zr8+M+xT0X{kbVAmA|sH
z+ibRFPw(b#{K_k4V_9cc>wh!+QS|GUbr+}KKlUU28{68QD-1m}-e0LMPhC8<NIHY*
z*C(z|sds~twl^kc`rXtuj4A$nNYCGbjZgk%{dOVkCRyS0GjDG7@u@u{bNuzZJ?Y)$
z#}yX8+2!W{rmH<z;_S~-)qnGnQ_sFCn0Yqk;kxswy@iv0NG-1odC##^&rr?(_FX?|
zJBtUJzcZg#+MHi(S&?BSxoiWssNb(Sao@GiRZ4%qC~{2k<B!UR&3D%r#7BKl5!dPR
zSXV#qX8oc?H!Dt`5U=Lc&8t}DbAA5Co<$$@cw=X-W1n}|z38Q0Zrf!)fw&`&&NE6s
z`0%_-c+I6=k-tx3gub{hxzK)h>DKFgqES=dEA%&9@6>4RNqkaxW#-@Q@v81ALXm54
z`;^)H$!fDk@hdR>pJn=Czva=b(rjw2=6?_IEinAy6jCR4=i7x9_On$yOHG!1a{d2e
zaWRL(%zKu1{_YE39C|kP`d4w=lUq4|i>KVLb(6pM%i%^+8}p-rM+=<#8e1(mM1M2I
z{90T4`YGeAo=u_OT|POUnEdAYvmW93>ugp1CN0{*9a%QVTx;e!)sE1Vn}`0Lur}Iw
z)#i8h9;L0(&gOfb7EL~^V%-~aAjaaDotxeXe%Jl`GCiDS|1Rl!W%vK#LALE?dql(P
zFUWXs6)Y~D6xXi%^L}N{q8koHJj;vkwVmp$z4CdR-|-{YUas-EE}ZXucaMEWdH?If
zM>byO+;BgyQ{?WFH_zIca%+3!9{BL`b1d1TwExwr7`I>Fq&8)250pH6mPPaIVU7Mj
zE4;6?3At;1%jin=7kmEIGd}zBJocCJfA`kyZSK1-qm}GkVYnyy$;NGeZrc{rZC|e^
zFH*m$KFCFtb>kZO>D5i4S8l!0?g{B_3N@>rw7lli^;Le7BC`+cZoahb*s82jrL#|p
z_sM3)te>%|`1i4Qso`?3`=|9UQJ9lG*Yx?io^!VfrakA&PXGGVzqPI;?AQ!<$H2pp
zGtX9Td{{L1<E*-OH^Kt9`u_F_S`z+GXZws(*`dPPM|U!?)~)L9UlJ?r7qtDxhNQdt
z%D-3E2lM<nFRbk!Q#h$<t62E$NB{rS-kX|Tbc$`=ucsGMcRAJm%yqWkUmfZ9ar$pB
zwIX>LiIQWx+1dVnoVCxp-b(Rj%jCSMGpE*X(C53b$nh}$k-p2Kh1dW6eV{AK#%lF`
z$^ti)S|44r7X=3=Z>tD*iEF)k^<vLNPj=Z^Jo~n5)e6p9tKF2JaL(Q3ou{*!rfv@V
zp2;<{?iU35pS7=YKbX)g8n5#0>bn`cCQ5tQHl$5-yqTvH9-18Q{@i9}%>-+;Sx)BO
z^R}J-{DZfeYip%V^pwvvb9G<M-eI@W+`>@nTd%cK@1cm-t;yP73WAnx?v}1A)Ka@{
zu+QU<&O-IaevSE=|IY8uocX`<$*YgWQ5P=Wy;e}Eak(_Bt|VlN*6xp-C$IeRKe)YV
zlC$fqgwIL^mS;EbzEE>dZ({7sIE^RA6ngiG^6dSo&3MnzQs<~#$CdchFXB6I$nV>K
zg2}SRnc-B){Q|xnabfxA#O;?dILT;q{{4El`0#?T;|J>4qP=r0i}kl2+;^$6>GaFF
z`h8uAODxh<ZGWfl|9p!vsxJM{VKHsDRIA@TDXYbjA9?OG{JJ=I`kx{no1b;1FCVg{
zR&6<$bs}qqjP<o-wq2rKb-jO$q;!9MtqI%knUBFX^>R|%MfJV&`jt6;OlwG4bkg!?
zbok<zD=r*nE|34;enhzJRKo3&P_@$~yMJ@uKXrYJIbYDuw;@ZvJ$1Wyi{Etbtv`2H
zybEmICI9(Zc<gD#r>oMEPW-#_{eRUP>)fkJuQNUcN5rOvF78^j#&>_qtyOQH{8}vj
z{l~+$?|gkJ*?(OBTU*|{**tf*k=4s(k<)gsnQiJIDY|vb?8YFcB{x-H&zM*EOzK{t
zL$c}owHwUz?np@Kuikofd*-Eg-q$lGUo&G>%Ki5=Z^iMti#O|4?oYSAygTjh!&P>-
zd8bF)%(XN6_@(sl<NLpF^Z#s)i=AF>`1}0*!d+P(xU<joB*_1WnfQ0_|Ke*4@=yK^
z*!Fbx9nW}qjky8Y!H1vMPp%f#ynH`zV_`e@{by5DWwqD6J$UoTqKEFk{-qw*Pj|K2
zy=i7~-S^P#OG|CfIA;CdymrI)y{G%`y}97CI^Q<)^<mjEas7{iMK`w|ejIJ~Hh1gu
z13Nw98=`kbmYM!)<&e1;J8g>Sx}4CQWvy3Fij+S)wMFDs;f(i1e-pHRBp2;3|NOL8
zsx8DX?DTT()vml}HNP&H#4FDH^~k!q&EEO{=jNBHOcb28eP!6`4Ds6+-$hkzyYT$u
znwV#mwYLhYV_dgAOH-CRp7Q2g#o4I$hv(X#-xEB&H~q_#oJ5EH&#o?R%ob^#dhv4k
z?ek_kmZc~!(?1#NYo#k!aH(vaRmAVtUkl#2Fg@E9>ttH%H%<L=M&6x-fQy<!7k>$S
zmG%pJ8OYmJVzqFlr6fy)y`bsC%DGD7R;-Ky42=vb7k<2!X*lHkq)++4zi;z?J+~{|
zwxqDJw`cW{cXNV1{!(w>+83F*M*8WQoq@*hu6&CB@u6YI)9<s7%srob?al%T-97q}
z*S9)m+*?<1-ut%6bOX25+<FJP67OlviIjiP{<(*->s7|-rRJB9?RC8+7P$B98r?p|
z4H`Vcx-16F<-*2~kMM{5{OTCfo4cX)pZNXxc}9`X7kKE`l>}_IFr0Sve6e467Ps)P
z>KzpqJs;fI$>Ndd_e-QIB0*erP0tf?_JjrEKkc^f?_W|`Eirfd-`|I)8ND%TQ~xD9
z!R}7rh3RYdMb2Dx&*-4Jqrh*qmh3#&8|&lpJlA@kDh#@l#dYs)`YP$Ud#`=Ahzxrk
z;(13%FOEmzzOmP4wz!@9OJr)ln_VudP+ELXzSh({yMO)8!?RMgR;|k3bmP3;o_>*Q
zU!H6WN;+}3o4Zq7NobKz)XlZawN009O5WO>zWVj;ExR3GO3E+0EB~$Z+&gcX)6(}R
z%a^ZxGjY)@mE^kU=Q`JIMOePnZ*Jee{=MIgiamR_C+}UjCH2dF^ZAoatAnmzty#O_
z(ligQZ(&pa95eg3<5`K+-t?(GpH6R?a`)F(XV$AZQ`PFs^eXQz4z7Ba>bk3L?bgMc
zt6zIYXBTU2;dIV=f5Un0&P$oT(;RlCaWSqg`F8ow`_SImYBwsBEcX?cy3PG6Inz<>
z<(!IrKbD_qPI`8S_t8A@?VXnE-(L8WTRtu3-ka}h67^2!<;+nnb6m9Cc*~3jURnOT
z|ExLq@3y6{+(p;VQ`a3gf5fn~%E<2e2Ep#lOLlK~`0Ym3-YGFBUYyt5_1i1`>z<kE
zU;OL0E%~TDi}~rRclyWqJq0FbtjU>j<fMgP{;8&Ix4a#+6rYyzo$cKo<ozeW<MS;)
zuRTpm)+)U$wlBKxAs70RJ9{>Vv8-s}(_a36b5)dfffKER;)N%PrU5N&bM`5IVAh;b
z@8<Q<E_Sg=egB)y8(W$l6(oJv{v~$AV%Cvf$t61()uyYY9}RC?5PV#N;pH@D)ufZb
zlebpAslPkpM%SZ<_U|v~%(}6aGkU}3?8(CEt7mDP-+8W!H+{>4**S`jUR=I-W10H%
zoL0N+fBYtjrkvY63b)lxi@o{h#r4;B^5!*LmM-P2X7p|QdUD6ciB?flDl|V&c$U1y
z-)l<3|0&1ku&g(__ORG&`;#NNb^Zt4Cpsvc;LvbVlYg_yMLU{HbK*_gp7P6Gwo99L
z9{m@4$baLbOS{)v=j*QZ*I9q%<EA+-r_No}_j^74?Vgn6y#IQFl{p9Zye)Zp;LoRB
z26tu8zPz-nU{lP#Abro-j8R5gE_Y`fw-o<!<HP6vexAhDuhhg}-LFte&e;)JeJlO`
z?c-gmluNIxRsDXqAx-<u%=dD;%9bAddh)h3H`}W`&tu=aif*3TUY}GEmzWy*XxU>?
zXODIHy+ySy({3z3wMJ9tu`6>@|EI#3ehJ^#PT$?@N^|ddJpP-=`u~4zUCk%?^>24C
z^bVP^H9qM`<f{8`Z!14lK5_i-_MElna<@!b(Q`7Q?fv5hy@{2qQ$=Hcd_M41=5)cm
ztIk6G|4W(cb6zwD$NN=ZK4ib=gF*GljokarYRaE=J@;CCcQk)eR7|Sz>+g4S#C)H`
z7B5-TzgaDM*Y<f^N+*ah_laH2IQ3`c;am53ANkC4ICXlZr_Sz!n_fpKoP4u#%^IKm
z4KWW314_TG{d=OHy<7cWR>a}Lm63PNOT1fe@46Rc>;K92W%tw#RXWmTdynf`G#PH>
zFSKQNwV|%CNp!zV$(0#;Qnz1-UA)>b<89{j=O1(C6zT1>d+Ed&m%o2kezS$lnPC5%
zrR&toCQtG$?O9oV{REqNS5@S{gU>#GWD?vRk#hgdo5ivIfBsyWdnd7K){RN?=1=YE
z6leKz<UD^>Qsj-P+{b4pT{q@W;kl%5`sPpWi36SW0zaO;G*0@!d~{L%BW07V6@7Kb
z+AprJd@S#=+-v2Ni<5qfB~F?rsQo=pNNMxQJqv<eou4Z(FeyBE&T?xT&zH}P4tx4k
z57co)ZruGf!LwOuZuJw-CGXf;)u&diT5`a)uXC+MN%y;HYFBbf&qW*Tl;=B=yv50U
z+ws3u(G%BgKkxN2C`zYbnaB0h6-7_(F0@fBU*=)H?f5ixo4Rn3mv7|SoA(wyN?3o)
zbMHpqIn53agaSUd`u+-7u`S%Ac}MF$mp!kH8)_%!pV7UplcN7dzWPG=9E*-8JLCe}
zckKVZO)SQwJoR0~U8maTFIU}gx3iyS_0Qir|KjHzcfKkMefu==&YUB6zPxw7{^xx`
z<fZi8-Miy|p4%t?eBz|J>vv9GHu3$t#8_q1ryDm{%{y^JTW?LG;UjbHqVvE1Ww<`t
z9J$=B;P<=@_uSU`ud7P`|3%fByL#VF#rg;GVe|LQy4}S$S^MOw-wgA2PfA-}e*4VR
zh(?>g@$G4wnC4FXyMC>;@sEl6eevh73n}|9<&V8kD?i_O_P&pav)Fpxp9igB$T_#R
zqrN!#Ujz65UEi0VE}K55deM@w=$48PfAVfQap)_={JXYKdNbctwarSi-%b8=Z(7s7
z@V$Hg->SAdc`<C#P5&Pk#LMoK|KGHDa)n^a@2FjyS3Kv7Qh4~8(Nl2Nw)AVi+t>Nk
z=YIF?E*JThl(*(U{FY}|RP=-sCuPm~x%SUDkI47eug|qSzDcY6i@%+XiR#;_cem@@
zjxEs(ow(y%lXmsm=fCEhn00TK!@JNQHAm-7GL-HQS1FL^ud6cod-!duc+tM(te5wy
zE?wIl&3*16E911+&k9dEu21>5ZOQuZ{o7^U_uB7$Zh7vr$m6zh-J+!KNWOjVydLea
znp1UgkK1?O+%?y)l_uZ(HgEfK{W8u)%jcc+ox1w_U4ww)=Ed>+Umnj}eCv9L*n{G4
zTNr)YjAu2K$A#<{o*K)r(XeavO53+m<(Yl*Ipwv7J}2`9tXgjCE^TfW>-etY+{?4G
zpS|g9o+@?x)7*oLVyAaA*Zw>5;Fx-=&S&vEhI5#vTHT$Zy3XspT3lU1&yU0U?voBq
z_-yos`R_08x7Jb-LVSx(87{TBdgbPkIIe41KF1=wc9q&{|1Q7!zQuIf(Vg=JEPiCI
zU19Q8fq7zk%c?)}GoALjaDB~n3X~4_s!`OkpRii%zu<&NF_nEOJHsPhKkG7b)Se%(
z<o6@71^faGjSmzrtYGDxXPBsBr*`0-|Br{AwcK&0D|Z`QJbZ*N=))#e*ZNJ@W|%HD
z*R8rOZ?J9RrX9+c7R<Y-G;_+gcPi5Y;{^(q&$FL0*JxjMv&6CX_BNk64Gs&016=2A
zxM?R4TKQjiqb>Wy<<l*sCZ)N)>0Ma6ar+kQou3|<9X*>|E6wt8qj&Y1-O9%9#fG1)
zSQk7Imz+`P(b{(Z$s(=ji1#cDXI{8#XQ=x8Y%i;N@rgy#r=F2~FMCDjRb<(wRrZxz
zR)lT-qFVlC$?_82yY+F~-^)kTt*Q53dUxu_Sr0jvFWx^vJ96WmR|zj&>L#9Zto-um
zUA0Z(*?s4>&s!r^bbZod=N_YbMv9rXKWA^xe%&J;W%2*xjwN@@&7QrzU$iOf#m7mv
z<i1<{V0r#vVSoIkVo~;pOS5<UzB8q?c)wkx?eD4exwZHDUo4FmDw+Rk(xNwcf$QX^
z%_xuCKO^>ae`t?YOvL5%XTRI)&J|7eyLf~vd7X%;+S>=py2jq6T~gl;ZadWdqEs<9
zh1vP+0>w5l4mpeU(zYjV32!&8etB{Jg>I(rnXfm@VoST#;o@7le){G9_xZYeBJ9oO
z*h@Q4{Ec?|udnv4Z{u&N)%z<>F4%E*{yWW?HFKI4#QpvLxBOFW(#*rYLAvj`47MG7
zzV>c(`R!9yIrn*umbrfsEPuCnu}|X%yX|W0J|*}T6}z8pRM^vQv}yNr{(Y8;d4B`#
z*S1!#lV1_ZZCzZL{$|mJEq#wZUyx_1m585h9QGk*qHaU|=jn5oWGjig=f;?CyL#{Q
zw=YK@r#vkb`N{G8z~}ydDIIAa4sdhlY?b$Usa$k;1^d1|Mr)-bpR{MQDSD=ep6k~5
zwdjQSmep5eoBO|QIlbHD=jk4M_o`oY<%&;VzHhlvzbNK+O|SHv1#jDG-@OY=GJU&E
zcioD`tP4~OH|;mmy%2LuC$f0c!bj#eD<2*DRbBB|b?=ss^K7DZ*I91goyK`T;KIWv
zZdGhar%U~mSG=7jCRVZWsZy5x^s;AhCfinT^Eg;~<84Z1Xt49rw<62C%0H@I=wq<m
znC<f~(D73=)2bT@nxZQ=Z}0r^F7(Ueyv1i*p4{tA=JDL9V*3BvB>iGX^HA0DPLXC0
ze!f>PC)@~KwB}l1^dH7Wiw`{yEs68Voy^#%89gJ*ylV<`x~rPV%Xvv|wgL=|7gR5N
zcp$fn!{F05p@eef_zPiS_liz*GBdn%Z0N~-Zxf_t7ytg#yPD*yX6K*ZGmH81a+xIe
zw#YF34@YM2X5GDs^GHdj<IHo8(<V45{NOkdCb)K5=ld>^!pQSpQt_(4cYFwo+O#_&
z*rUALQ2d3S!JMtKnUijmbJ;KU3RK(qQh4oVt|{r$wy=u}W|qbMTcz0f^K_o<r+0Vf
z%gpuC)2|A;y=k3?e}}=;Nx!?_D(`kRnQ^|*;go;A^q!VA)0V%rkG?zo^yZ-Qn)5T9
z-`_PUNPU&P`@Kr`i;SQ)vlTptQmgNaIPWUd-{ly`l~Va=@!ZziCtXiQ1ToxpTjpXG
z`LZPZ*bR=!Z$Elht@-n}-amG+cFJy-nO-;4x2;}N`f<^rC##m%U-`YcGX8m7u-mmC
zbIw<8_w$JQe&%ZJ_E%4TNCcj@@X33%=}5#g&{_<;YR_}5@o)8I<8wo|>e`n0PmTMS
z_|yOYiu2d6Pc+@N@z32S=WWiY=>PruYO?($@t7Z0KPTDe?%ed<e6R4Kt*(6M7XSYA
zbk6yHah;oiQ_fv`|D|}v+0_x8c1o(BZv1^-dR^zS{tNNoinAM9dPRInY_zZb{CwnV
z^<+_-(*_2bljIAII*XUw{&Hiw`mQ-DuIBsGRn4_ute3u+dRhDOuiEpKbIy5otw`bi
zJn0v!;<@D1it{@*H0|qrn11|ghuz;NPi@OvW=n~4?%-KIefih;_=r-AU%TggGW-(u
z$Zps1S?BketbH%O{lEHib+gc=yRXfN+P(eVmZN%I-=2hKM_toyKk_rj+~cuFP4&0f
z%>NTMdYMnE{hwX*-KkM9Qu5ro-*RUx+|Sge_iyyr{Q3QfTvly!OWrq<i}(Jpne%VM
z%@+&TzK*#Ty>f4DA<vwd({DH{Y?w4N{@{#llO0dJI{Wj{rp|As?>L=>Soim2-v4Oy
z@!+b*$JUsgzkOLRI{m#`cjdO8>ywS`ALZB1Q=T`|)~ne4NZr-^9UjsB*H&M!+Nbqz
z&zB9)^kdJbxZ1vFk+s!`GZxrY;`I2|zQs)O>kl8#aQ<+0TG`zqOEarS`x*bfZ}@Cc
zcZq$q|A)-UN4R}ndsNz$Zu(}tFt$0-SJ>s>Z?)gOR)(jXqARreJLg?o-Qu+|A?33}
z74NdSGm=}hKHu5WJ|S1L<lc_lLw5x)rnkhXhE4c1r>tRRdZy=pj<pxw2zcEMOEiAU
zX=&s?=iFL>-McRSvD%#Y?nR!t20xPm1Jgnuhw2yHsSFak%Q^P^J1KZjFUI1-q<8v3
z`7>)9^rNf2YUY#%Z8o^{d2i(s+w45^<5FGce7yE@c3w<6CR%yJeeb@t#rCX)xhG5e
zv~w5oaF{4An4wX#<dyv7tj?zC*JieLcDG-jq_x|6rQFS;<MU2Tn&x*)_4)VXr+zfL
z`n(b^K5=iw=X&<Z`d^v%OaA4Ols>hNEi0z}eMi0cq}0m&M?Q!xUbpmnuja|RgeUE9
z%75)vPCdD`OKEb`mYSILGnW0{`TA%Q>$_dm?louhoUX~~zOK6YV291O+h?b5w>9~s
z>;CiJ?~@0vF89v9{>>{RvduqF_ETwPZ|+vlNg-i3tc41~kNrOPzV5}hE9=%zURpNE
zH*kqJ+p){ca|G6HGW!Nz`6sz}_nw`v0wTR?X5`E*@_Q4Reo$-Yd1<%ON?o-*sr66O
zf_V2AKVGo)%D<?x?4mV0)Q+7ve`9@J9cX>l-g`I7HYw%h*ZKbWFr_0X{d-$#Rch4#
z-aoercPZr^U-_yfZ)4)sm9>kyzWzR(TeP!0UusKR&7$>7e@~BzTHDh5QZCeJlY7YF
zMV!-jr&T<XtlGQ$`)rZib-wR~gOBKi{@69|N1be4&K>Dr)#ZI}w%6Xwbq<r$UKi`!
z=s&q_nah;kmx=%Hyq#hEO8?F@<=4~Rt+mdb$oiKdF6Y6~+0klcwvYPGG=@z}Q*kd3
zFMWG`=jNWi<;&P>t|Y%G=zsI>sU?@o|5MD#TOB%MUo(fiO5NGBIA~(>qaM}kVs7@g
zWrE%B+upcVx;J;R%$<JqlUMic$UR<IaPNM&%w?U+kr!9KKb4aG_33Lr?r+D>ioL2j
zBdvG-<jg-+a`ENY{AGXYrn{~Ak(+<%chRx_^&f5KC39W9v}(Pf?dN}W6P?#d8lUj(
zG)zd#OqKolRZj%8u4j#3cJ0gD>Pze5ri=c`i9NSYc=5@#UB=!Y_8vcVv{pa&av2|w
z`uz95J~L<^?Vq<&q&~82$xQy@ed#jH56ZXmi3(?4ROkuYVmVhKe8&GDpENXY3U4{E
ze9zG{Gj4qK+ODU$Xi|*i+&o2r*R$t8?O3^GbGSKIlD|y5$N8VXzklnzQ}g_Chw1He
zW*V1&e%^YMH{8nYXTd#Ama@njTfNuba6cEiEJDL9AXheb{sPykm3{k6EF?5EFR9);
z<#by`<=a%RUs9Vw)T4Q4abG^)rFFUV_xWx!`DOe((F=<cf;)9D+}z=?aOY;1&z_po
zKXtXF*y-h5`@1O2%Hz}*rH#)z)vT|r&=cZlVBlzIn9Z`lr(9T~%Ub;azyBW7$?-A^
z)HUsVZik%|uFe0gP_B0H{fSdHipvUjFWzj_SGh-ar}5>#^^4}#sTl7*S*SiWH1eqI
z`MhNdUQe>{teNO=Ae$xS=swXY&tJ`*&vs#22)FIicgxw^OR6k9j5>~V9aTH!dU&gu
z|9<<hi5`r$rtzNUJuiwjtmSfBnh~A)(<=JYr_)n6tx+l6(v!C~uK&k|TQid<NqwIe
zoS6P;qw6Na8)@FTyOfq&8x>WD^L6HLI8i!%YoK4=>$ZuVxq0bEQC-rnt*xK4GSyk`
zex}ne`r_P_7w1lGa*7J~vfh2Ja!t)VuXl-2yVgD5eR+M{@w>|U-@bU~{EEEPeA_Df
z^v8Qrr*~Ysc8Xc`7Lyq3Z`)_D4o%tE?_jlj|37_?d)-W)Y5pmHpMH)uIUVh8F0zJW
z&c`QgYd!DYcqkoI^eN9FEW+y8zJ~`UbN}BN9(-ccp`EktJ)iT4*Yn$zYhR~5x$t^-
zan9=X*Nz?hTTsiVb)M_ezPl%ynP2X_v6VyG?}L}n@3YS;=NMW}-oLg^u0C13vTMSN
z<s1K=DoOwQ`qRdxKi;0dVe9j<Tl$x&-KJ>yx{b4^@7v-$C7I>vq~DXL1!dPtzTR<r
zUEFkW-F=GNb=Uk!TP%9Km3yzu`T5WHY-^tKXwo)er8#}<FN<I2^i;{#&8z&h)>g?V
zM$Gxwk*s$*Un?zs8`c+FzO;~i$F9$_=j+eFKo@SIQeB-InFZ(F)57wcKg8A9Wu7+e
zyJ&C!KX&VtZF<&ybzTamgM21^%vRrPmp`Rwn&+JR_K`CdXZ&uw)LQbA^GoU>yAGw7
z{P*8+wQ*}NnR5Nv;Vjd)#@fQALHj1GS>hu8^U>{et}Q}e_@OEBJEz0UceTqD?<X<(
zO8k6!x@6aeHXXY=!Ct?E?9ywWPR#eMtDNNbB)l?+x71-lDU;dH1_RCis<FRR*NT5R
z7&^7q{gkce!g(Ikto|@0`n-9!=-|`Vk4@Uo<#V+y+D{*CJuP~@-^b#qm+X{}Jq!`m
z$8>LGO>7FhRko@~p#Sxx#*OVwxi!AW7I#gKop*0n>6@P0vr^a}OK+Q@x6v$==k?=~
zElJP*Y!hjDv#dE@cj>fs+tu8?bGJ^t9~2+6er;iaWL(09=CDuIers;%E_C?bw^H?I
z^y}m6RPy#K%}vc*Ss#1un#h?d$!#pp!=qQO-6!*U_pNms=UKbi=<vV)RHAOk-?ux(
zXN%jFGhFl6PqvGXPB55OT6XSf#LV|uM!RN4y5~)r(A3ND(_sC~Se2yR;T%ubM1};k
zM0e_3))eSov35eMP~P?`#oMNzs@HX`Du_C+_`O3jFlDA+yq&uLC-Kb7xk7nLLB6Nh
zpKJ-}zWFFERD_*TfTM`1E9r8XJo`_B+cI)fUgjz}F$yp+3UCNAsd4>UIl<pG@~*eC
zoP%&VUq?#G{5RGD-wt?BT;kE6^CD!agV&^zmNRF|E`2O;-Zxu)Ygd+%Yt_$CXW?zu
zyEYeW))dKaW{B9%sNteFf9;mG-~$&|-QZX`iRqOb$Fs!MlRl*y^y>>8Gn>F<w@Z4;
zytR$pr<R7#Ol4qjVDNNt42cSR^85D==fX4p>in&|?6_v=o|8_!zvlS5O0zedXMXt0
z9#zk{mHPkHmTOldAHU4HdGI`w2WT{lfk{`O!Duh5!3j3dEX4%n38^Oqlk4Z|Pg=C-
zq*e`M?|$*UQd4iv<#O3NvFN|f+o}H-uh>?rWoC0%C1Z*8qP=Cc|K)!?^^=|QwWq;)
z%}M?pX`4K)U(BAJS9Q%c>nA5;+0o7auP5LCC|qde!}Dp^#Q9qatoKD*^=!!IJ-_?z
z-vwXZe4o|4oGpv%s3H5Sy;<w;Om;83_~%oxbu|Aplj7M|?tKfCXK`QvqXoH40n4_Z
z<xmCj6-qdA`qv6g&fYJU_$u?(9ZN@!V5!Je^ZoN2EN9KCZsfe>-OH?X(#x(jNa^|R
zPjQK><8-f_czdX7*@nHoi}(E(-0}R*yY_s!>0gSLbTU0uYII-_+OX<W_?GYYtoNBS
z75r(`+2pyc+Lz;r;STL;jlP?)ZnB~C{_cEzvS3yE)+(7R#>*aB{myb{0w(|lMu88u
zj2f3~&&$|0fochP;SCq&B^J8He2mz-qI*MTV_(QrCWVuGu1;CBQu9fbE1z;{phRTX
zWNwM8-v2~A3)fA$HD~*lP1fRkD^$9*7J3|iS;PKytMd%y{o;{gd~!E`h<N2sXlifR
zoUrPku~m=C49*MFgFfnA{`U2t%`peb+rs&8#X1Y$t^F%!FMMNThEw60#!r<cYR|K{
zCRNRd;oi9Zt8z(+dhM6{F&p#36(g5KyXMuMJ{Wg5Q#jQT61og5ca#}YcUA8RkOGCa
zQ}cmUu1k|nP04SyO4zIM`d_+Wx}1_qaJ*E_t_fEPmz<aLc^=y`A!rY?%R`O?)+HwQ
z-QT`fR{dw#HTO%i#^!fo%U*e1R6nR@q`M}iUFoj%fs0%m4Grd;9;Z%u>lOq)@m(@+
zTj{0+Q6IFY1ZnhHPI#X$%evUM;E=8AF1t9s4X^SpgzVMpgCtP}2Bv$845_=yEf$G_
zvzr;ysuQP@?qB`0N-(&-Iz(ezQlN08i`e7aCzJMQKRrI_(QMIIoGHF?&Rri0s$4H`
z<p`>r{UT=B>#zHbC;4i8neXz}(@Xd2O2z-ruR2Zh?9`jMb!U{@+o_XSYwWH*58ucU
zUUZb-zc592$6o~wP&WDJ=^&U@8GiHI+Ilbcm7wGY3o{1>mOY*fr%LiKxkQ5;D=>#8
zWaFcmHBwUh`OJ6<yN>F-QhMy-A{n`6_GC{>Irrl#iZ>s%=5VRY?&zHIM)2DwwW8aL
zCZz6+apwAJq!sMO^*u>zvVO{I&cyB&3gUVTQ@w652&pSDFg5xdP|;kia#}A+#RZ<`
z7{G~-qYScn<=799%N8u=NQpd=HMeF)Ki7HzP0LLh+|ylS^gos4O={ihCA3Y$Q$@;3
ze|2E1rI%>nCZ$I&LfyokSAMzW+Er#a|IYRZp4LJxCItp2Mh6GcP|>HZk&FtOE-YY+
zAl8A>IuoPD<-H7ozMvM7!waDe6O3Xi4k}$1+Qng5B*4(vpz5%?A+k2!O7O--ZRTlq
zEbJ)y?+Z5=-Ti*%qB6Ke#Q-K98iWH{qgKyYyPBy+LI9rUA>n+0mnmS`ZboibNLu|f
z*&)~Uqo<|OPu~Ai_<|Z4I4Wj4%x{Sm*|4K{vdWR|cXf7abQnQuNr>qT91RXD1R9KN
zL*AJj;{<un(dWRsw1-FjDaa|cu`ww;aA!HOxBF_!q_AHTJp@=;Kp_dSn}NmQ06WV9
zpYMz*C5#}QplD`bQV?)q5qbJsZ`#|VjwVVx=sH-LHgTs0u%Mc=V4lPE<=g9-C7?zy
zK#g1I&T#6>h276348cJGcQP6S-t=rnlR)LR$T4bMc78uE`3RadXxu!Z2BW*a2^xG*
zSD;BlxdK%z3w-$ZR`&_ww{-_77nnCLw7}~HhZlkkMz&wxS-SDywe^7-qsHaCENzdV
zu7HIU8kdNGGf^}+Y_u(UGS41HMu7}BhEreCJSHH=BdXC0ogW-)%``t9bmYC!ycrkl
zOWb-VHZ}YPwfW~uW`UABYG6Y}90CMpBzm3?Xf0Ve!DWg_Ue$|T>c_tFbu}gmC|qD<
zS>W@YVMT@zYHENdF9w!Ba}V&GKJq#xGUlqna+RYlEv@GpPfur4vS4gfV_D#n-^84c
zk|2?^Iw<_$JhNp<yyEF)nP=U!x+{u)emc=}-_I+rj8{@Y;1MYHKJqiPfd?DVf)B(M
zh+;9)c`42vc_ypy%A^-c?pplkt8`2k9yQv*z@+HGaO#V|o|h6vIAYZyL+Hc?ADeKG
zSN^wU_Npw^2z)d#a$4ZV+-a`hR#CvR_7lCw3?SJS(+f_s9oke&7pa`y?mZ<&N%6F8
zPSKY+nbWRhOh2_)fP=*nv=`+MTXF@+p{Vf$6)E_8=ZfJZmH$B;Cv|HTGVlIwJN>CS
zSYASgu}eYV70a20d<>6K%bt3b0M*dk-k_xQAKh;%2!s5!!^<ITY3#abu9gA}jeW`n
ze^+ivOmqD;bwbO@!*OhGrLT+XU%j~!zUlsnb)PrzhA=las57Ki9k}nrgAxxCoNMY|
zzH*Yf-jyIDeT)T`z8_dK8SB1m4sqSvFY+?+?aR=hqZb(WKCNayE3`f}jVrqPmQ$<%
zM+lSSN}uqqoZCw+47CFCxt_2w3cQG63i9>Md+yb64@&?p_{)+O8T4CoQpJRrCDq4W
zyH(t@7GF;94wU|{rsI^kza@i_g{j*iK<o16e}$Q*ciZkZmTD{Lf;`;7z|kVtD9Nop
zM_qH-J#U@ac6@s;q7@ex3R#v#t9=Pdd8znH@zm<Q9!WKA9=nq}Z+zj~y1Hs>t_d4c
zo!^1A4a{|RVjb)-&k8uOq(!OqO%46Lc9Kr4<>QrKRx^8)bbu=qXp}K9)%hm0rRHom
zIBC+})O)O~WkJeIyUb&UwnSSm%H6`r^v{RkR7w3l>j}Kj{L(b#K*{MN2c4$9zr?I?
za@7pO!;`8KC-8|MMy&_u3(iQDoT_qKke}Pr>%|GbRK?f{8x{U8>wCkT+uX2Uu)*l=
z{dXKs7$H?rqk{T`M9JIjYf|2+YtJlr9FQ9M>*kr(OyQnN3)B=NP%!tvn&UyjBATxR
zS516UrF-7nYtjy#<by{f=WTD8r63Ri%42JOvt~kLNnrt}&)>H#BA#6J6Lc=k{gFE*
zWuaHOU%ggtgag0OA(V2|L0v$h`%6UhmLp;>HTS)Vo8%LouH&)y+Vt8*`T{R>m;#oS
zi&cK%hK8@ATf(Z;qMDK=FBgVwpY5fj<1_W$+*HrE;z<<-F`$+aqK2MvU`_k^?oB5r
z7N3}2>EAD%>F(+|Y5Dyf3V%P9nKZ6a5D);ly5lyBBFxnrI6pkG^9qWX_^xZJ&l2BB
z@=q3cHri%+pq3o6{t2&`o0X^NU29iS{QmaGy{LDNQ;+oL$5?z^G|l?~7YB<SqsC?C
z|8hwz&<KBUoh2xJd7EZu>G#R@NA1>ed8Iy15&ZapciP9!b=Mp%QQXa8%4BoB@71l=
zM}nP={|w4PPuS1hX;8Y!{sfbx!vbZF8P<yew}2~fi0M*{-4obPTw1Sj<3+;u&J#+W
z)6VewZeC)y|IVb?mu0mfkO0Z(Y%ocF*s@VK_Dku$(22z_tN*Q8|J$`5WQM>Cmb9}%
zYSH<H=8}tezRuqmAfuvKXIGpU_4ZNa>}@~#V<xb$>;Wx9d|%+A4blNFiTpUuOt_T4
z{EeZX-p(niD}&Y-L{?2`(%d^E^JeJlAXO1a+RT{Bu#<Vg`moE)YsEf2blt4-&-rfc
z+rxSnMM3$H#g5S^D)7mb-6yu5&iF6gqq5rRPf?rdTYYcc*=4`x*G%AGu~BC**Ppl9
z*%uZ{jbR*jb|{oi)9EaZ)_nA8X=9g2ZNI{4>2U1@G9WXd6%qpniy@<?v)68Cj>@L}
zeSt!Ubi6O!y*%GX`q_FTCq01=Eeum$&I@psMVQNZXNN-WDy@^@TcS?P?Vcnt(=biQ
z)_`Nybmd0mI?v&S(20l%W!);ClmE<XP&sty(9-Jks7vfW_pT8TV2M#-NZr-FAb<y&
zcO4FxGMVemshFtcX}%-u*4Lm(0hiPDJa*swDy4xGjS37*2VE7C9e?>Q@z`(B;3n$%
z%z!^gW=S2Rim?l8qZp^d%zH{wF_73~U=%2r4z4(~p3L7Beg5gENhYb!n!aS}O?P*M
z`3IcKIfR&ubyu1PY`A!uDbHe32bcIIXUTexw~_^$oD&2J6c|!>F$?SgEp-Ab|IowW
z`>vfka^GZ|jU{jXI0;?9`y=F*$oVBMx~Rt5Fuie``0Q3Nv;U{6&Q#5?O<w%xtE!y5
z4)ifLH8TjlxX~V~zy>xFG&sP?B<b(-!hiDbYKsdK^x4*EJbKpC`ts$z7u>b**a9&e
z(uHOu&a8-Z-x2ub=L{81$Cak`MN?BlCY}H7)~@hCl__9ZHK)ofSOFHGYVfyw@ow+L
zW$c-MChkjes8dq)6fHj!GsEswc<my26eoW$ZIqmLg*)|T$>jCF-){JqGEqc%k^H_#
z8?UtRy=LZZc3>B3Fxt!fa8V{y8%G(FdBidGwpRDEf?IMgKTn?8;;HfKE~qN2xo2*L
z(jjQ%Q!8jz&7Gv<ZU6c~Ud=rl%hoqy^4e!-CP_a1oz5}8Urd%w8dR6R?Mt$O=AZ}4
zO!mLN-jIkinyeICa?i{+a#4iatG^s+PZrPa%R_D|)Hpq0^Zr#aS>^AZ!;b}jrkI?y
zo*1$+e3Ry@g^~R8#jpO+Dsnhb4;l;G|DA0TG_5w=7n+eODJmx}|B=Bm`sK}+PF_Jh
z8$65C{*(n3EkDh~h-`zyg1Ia=rY_X!-p%s<2+y-6UVpC~xy8QR&cH%xMtkQ*R&mDp
zES%hWDH6OKpm<{7XfbPy+}z?hanXB4F6O_jEh<Y~1XtRb{JofXzW&#peMa>Yxr-nw
zp=85umK0Z!E7v;H@2)-m)c4^dJx`&#wkvP1vs(4zOqBMvUH98`as)4~nZKrRzg|{D
z!+Vf-s-L~MIGaVsX`ZT^YxU-n-?mN7ycz14yW(vw`&Uq*eZZg7<J5~U@e`(=@u;8N
zFE4c{>5<%=B*_oQb-ewy?mx+1iQ>rvqD(WV^)8Rqe7?r#Sko7)^q;bG)mJV`xG8TY
z_x?rQ1v`!sGe(We*1!1{{4W!np%(O8W740B-BKTmo<{on&zS#nx9-olJh8K&5N?PU
zII&^LzZQ{~nRSXi#)>K{<2OyIc`xF9PCHU8u{4aO5*p3mvJ%REz{&W-=X;CPD`=oN
za0s2)u*7$I%1@#H^1EzwBE9x@Kbi6;KFosyTHir6f_hsTpz3OhHLE4a=L{?k3xqk(
zZ1CCRGihnIbs)?4UDKX?G!|QS_f?$g6nM^q7Z(hT(?BJNk#r0+^Dqi1FqubOQuCbj
z=dnYxBKONC$!TAD941SD0*&4v7k3OCKhzmDF8}?!&p`{CH9l}N8H+FLH}br6Ue&8k
zal&k;6PtEq9c$X6eRmDk9?OLuhAA)4PEcTm6^ad@COf}SSCx6CoH?tJU=-^N%jAbk
zJ~LJFOpt{sf;TP~d<HrG=XVAnXuvhx7o3qd^Mbd>tG`au4*m7BFlS9#T@|UZ{p=PW
znb(k^N`&7b<xnFVQ@}Fw1qV2xNw|SQc*e)uo+s08ewi>~$&3ezPQOY{np8iV{J5Rv
z{0svjs9%wci3jx;r*F9^4mIJ2W#h|!zV69NS1cz@5c(Sv^8M&!ml+(~yY&wn?d<<O
zH4yGyxK9+Wa5~JCdo`b-2iC!V#p$z=r-P@vyLv*-j@_)fBAzauPL&zgE(-@WmGwwS
zK<$Eh9Mp|(7icir>%3u6Ak>8%I!wVj!Dl>!PSrY1{lz-%iJXUacSZEt%yXjsOVug@
zq(f110oW~0QyMH?j`t?Qn_?0iDUq9UmITyCu4wZS_g8kQt63xSx2OB)-l#KLQ?7eW
z*<m}+WY$)V)6YC@ufWp~Ja9OCnF5xzKdU{y0-8DxOl<JtDqWqtO#A(EeW@_@tQgIk
zjHx0=pRK$jtH3)8rSt^H4@VMH!Dpc(=HkxGU_RVvG{y%$(8$YS!L^7GK~1d?aTY^R
zar@(0QU)RmA!kJx8)aAx%_;=9Q9-SNVTof+v7rex(4lgT50oZ+{LT5dHR;{8eSs?<
zzFqzP?yX5WzfyI4{PV+Cs~z38M6Y`F29Bk%Zx<kSA-6a&oca>>-~rkQ4Fksy;YO>~
zbL7=ln|n@@c;#QT)m_Z9<8#ZBdCv=_yH-qlJ>~jBkMA=SwBVU^hX+GmTntNV61q1p
z++>*+y-AyUs;@p*=+A|lQ$Ke93*G7wnmzyQmYkb!#XmW4z`WA9K#d`_D&T$`4<BmQ
zQ@FvA64_GvanITO3C@#651N^kA6v{-xM$a<Z!cA+y;Z-nvV9Z6-UphD8kcQV+p7>l
zkWjwR$s%_C{+5Wq-1l5Y!S7uRtyaH|ZQZuH+pfOMJuscKp$?j87+9S88K%5kSbp9x
z3FbIRVrW!Q+)#3F+W*MVk6OITtvuSUJTnLh{h7Bh_wwn;X>%q>7S9d{0H<tFT0fu;
z8m{6@D`7$}MrEB7Zti&_8|E3M!nt)va@B{$VWO!wCsoe=-<YttuEbw~3!2|r1VCd*
zFVd1%p!w#YXTr_N(*54sUM+83^^EN?hpOV5RZI44t@!us=|Ky-9!^*gDj0A$%#@S(
z&KU%cEJ)I9a9GV^q`RBz=_2FvtDNp$&N|4j+oS7?*VSeJGEYvPy`SIUqM{zu84V5-
zK;uOx@-*6*QT-yYi^VA3Dlqldyua#Pp1bTMG(1*mtk`;UUTn_Ed(u3Q4@!VS0vsev
zvl*tmY^?pSU<)&q0m?8@oG?=}$NO1+RW!%f&1&LIZ+lnCP1d@y+<t40rDVn;=Qvn2
zG(J#aNUbWk7v94K^B~x42ZbY?I`d}#?OfHhpF8vh6JP79y!B`FBd6@t*sby)rTqXr
z_#C|t$ZSzQeBjUg2G=dshR0QMuDI=<{ZVUm*R80@9}eAE!FzOd(?Ua-g^<aZ4Jr4Q
zZM`0?CNeYivaOR)AS0v0)J<aNZ-22nxr(i2j{tZ?=}%WfRZytv>i4djrvBdN|Am8T
zpE^V8uJ<z+yDJ%dZ9FJ=QRyL{;>qM@UXLjvw)x9E^6zg@DGffC^Y!i<^G&aBT^GoL
zmxG{@0jtK;X}Y_Ob&|zBmwjOGbe**SQ-G#t*OrovsgK3~X(>MecWvs`6Usk7eIMjx
zC%_TI^e^?2^bQ%towHfO_Ec?XYvTCpyW*6W`D(V(!qCtV&lRUn3O$$AxYYbBEV%$4
zBnl6LnY?vw``^g9w*2Ik-#h%R%eESaP7v(;b1dn!n#QHr)V3nHVmpwZ&QDzI%*;4d
z<kkEM-)8)^XxqbTCi&>ficO!69sg{;*1+J9CNyRtNv)#wfcmtxM?*P}o)1diz30+`
z?NWhPxfCPgV~p}2yXBj9mcbJVsBh^o^PXVH#bwbf6_;{^OZ6vCt>R!8590QHwo824
zx!ar0>Q-o=j#}mkM0`|uE3h@UDpKRs_PmhSI~J{)rLX<TxchmY?S;a*3sMo$2`{*}
zOlF8Y`Q*gJN0C7*<g?R*8N*{mPbo&Otup(3GIuAVx(y3*`-Q`XN#sS{$3JiCy_|k7
zncQk06dw1lWLfItmwwZj?;_%@!5cJ0$tleBv&rGv<D06M>PO{67pDDQWMBWH`Pi;|
zTa17E&zu(9QT3MJoW%$dJRs76L-0q%yJzP^W`7d<8Mysd&5czzv~I7O_y5bz+Elyu
z(z1x$+tAK(M(&!kslW>XCYh4SCNJ~PZ&@-|$@TNrNh;p5mjaiYuhN>dK{wW>{<QgA
zJuVh#9RV)EI3k(6b!P9M|1~6PqHo~)#VgCV_&az!ntim`yw;<3qMSUu++bi~YG=^7
z%=|pAb;^=%-P!vBTHgBdosukJ4!^1uxpnV9Ly?z`<=KV`h@j<gVVYTN{#H2cW6!Ft
zroJijpM6`Ua;0C)e&<W(SL|J1`dWIo!HZZcP{&ucNwl}Bnc-*PeJz)_@|p8uU-W+a
zvyiu{;>2p9(DUh0+unR*uT)@0NlXTfs{SeSR|WahFZlWDBA?=-E~`YL`kxhf-%jym
z%J(9L6F-N;%y-NuUK}b0C+{nvQ*<9G%$>LWk-q8FTGLg(?ubu|<=F5<e>)3GC&5G@
zBJpN`*3DJooS}E`O?cTV>?xY+G;yAK>)$kelXf+vw0aq|S2**<#f__3EHnZa&icD)
z(W?8qqmJJ{HStZh*y@|>^|q&k|C*gMfgPo^Dq)&AEp}`7^VI!XA*$Zzv(HbAIX-8j
z=GOB%+j4B~%qt5JM=1Z($iOt=*WuU(o?M|5n>bbr%>HM3c~V`F$17eZkH(DxOIDeA
zhS^0~)qmce%O}JNa-4z#r%TSQ@JM5`w@1$j#EX9Vc<zOGEhu0dCI|<JpTGU*Ue9v1
zc~e6&m${sr{Bf0&ZA#FpcvaTgo$6l$kzD$xmto4we`W3p{;OF|EK0q4ruF6WmPP-Z
zJc2@2w_5*Rv?@g7(648gDsO))-=^^b+>*4K<dC*(L#Gc{TsvgoRUm|A+PS-bQjZ<_
z$m(Qz`A&4pPE!llvw9(q@15C{GC7b3sqMkS-gsccJ6^tP60OrYbmna@oUXIvdFJcP
zebZ-ssWN`@_;1Ksf$nKh3p~DC#h4&>O&%CCc|Xg&tD&O()-EV^yI%Ozsol@h|C^qk
z#Ok{1{5ugVa7hVDs2^BClf#aZwH0$6mZd(<3=0cBT$j%2S*KVowu<%q<oYk0<K!)*
zP}_y_!Vxy*uN$p)K9l6~xz%Ray-LIE>HQ~4&y<A%ziyoIf)5c}R&yApyp-*ekS!5m
z<a@R=;_i=CDl4<CR`HkL4h`9u)tUM@ev^*n)pyriD?!eIXUz}cO!MtFe)fs2{CW9_
z?cGd;K+n+Z9JAZ6y>(}Y|9&l@0Bh_wFmP;{&M@WWTh)$VbG;8JP3eE0`zs)Ht^A~=
zJ!>zw1iH?hFr{8^Yfg-l(E+aIm-ASxkV~-_8cg%;wy*cu6t>slZGPd4#aC9{nD_t4
zrqb2UcK1LD1X3eA1b}M$6?y3|_|iI4PfNL;SU*YrYnR`G0I%A8dn)3-wfVh}jzy>k
zO|+lqNQo-DsyA=<%hJxWl&ra0k4o11*3J@nX=x|SBn#@RLd{keY%tn8`M~1n_reib
zKNIUydq2%lIimS~`@21^MvnGzv7vi2Hl{AV{C3^BY(=Du;K0DbG^=6NTfVKD`BSG}
zwQG)WWUSR!w1E{44bq^Jmszaqmwlt?`J9vXtDo7g=O~@dzvOOGkoNx{KQ~)<)m#wY
z1rKvDW53ddBb$Eyo~obozGv>jx}%JTK&3I*sV#hr#V>e7m3c*(zCD<4x&D02glUQ`
zC6(?y(Sf<~0R>CXTY<;P;Z-9#<3T-3!pwK|-&uqz+Z%RO{9AUk>*}v{S|O{;XFUm<
z5)~SHuSw}{waZ1MEC9D0!e9Z-?=nBsb9C$H)TzxnUFA9T-;z~XGmrOP30G0ADqO1-
z7n5}I-kp2f0>If1l=~s=A!Jfv1IK~0-U)a0Jzg_1^6^!@eVjb6x1(o=`=ljyawd+U
zSCm5!{n;)$jd$t1tM0l295zhPZT|#L37IM&UNOlT+yu_>b6B-@UDWS$U*E>X<li$Y
zg4(nonKS0!hwa)bioUOI#9gogMaGu?2BjnR;rmM-D!<J>`aj!w=DYvjIfSYj8Q3}J
zrb>QoTIBXW#v^HuwePA`Z`++V{@kn6{bA8=iE<T(NP!nJOo2h`<4<0EwDjCd*}b4N
z)=<v5A~5j(60!Pscf&49fvjj?;P}zdpz>(tbQSq8EqWlw97txGwCY$zjAetC(@Z%B
zbM8gk`W=>~hU;$qaO@20%gZk;w!Xa<(z=hsE;i@po=tBzY2coGR1i4B^1#@3#(TYE
z&!!xB^iz8JZI#=%Wqht=GU^7e-RAY`zTLJI=ZW`>R1o7B;M9baryD0IGNe|WxRb?m
z#pA)`%GsMOXRJ_}l9X&WfAuPjfU~bYpMIA5zF<$=cl1;P>8yasg`S}K=soWxOa&wI
zPTHx3cDb%v^2~%Okblbb>o&WebiN7ht+c??QROHF^@o3{D;zmKf2YRoB>^|TD7nhq
zTfuSr*3nnXribs_uvh8G#)Kbu+`z{o&~Sy3<6fIwqfYmRNqT?ZFElUekw6rApuk5?
zZ3jY`Hu#iJ6*+~-^Wf|RNzrh!MF-Tg`F1V&0Upa07?{?9Mms`-G~iQP;E+MrBd`fH
zEOj%wzY)<C!)EIaFNRYk;hqaohv?ysf-)RV2pP1$Ibm3U6rYR&AH<o0btWE)l2u)#
zwAiKB)a`^<Pk?XYEpLv^6JFYT8fHFPwF$JSMIcP^=Bq8!mBKqXid$E#S|?PvSL<jO
zd-k8yj0vbiF-@%uQ(hh{Qa9v7vZ#?k!Qk{d0r@+Jb%ZAQF0%3$OMIBN@XFCn4a>-g
zsVRw*Cbb%rE=`^iFeO8=lVjnkL=%hOcVah1wnnZJUbRASQ;4QW@Kl4&%^}+zj<n1b
z6;0=IeU-d)(XKOUO1m`|EjggZH*edmhbgz17=$3pN#+axSnoPRf9nOM)Q+V;BsQ;D
z7wEEf>h+9?pwV2Hw$n;BK4MEn6~Cvy+HbDvqS|a^>ymHN#*5Ssd{E3ZX-Clwg`ggF
z?&upTidO=a_GOC)b@2;0c?p&ZX`Tt4JV7OL#g`zD-gT2)cdhbj+Vbqnxp^U0E42P<
zubkDo%WJ7h=)!u@K<BX6n_|=-`TSI}Q)=fD>-_N~Hp$@+|CPMPA5k(-Zhgrze0pAX
zvKIH{rMhAVSV4jIf`MsO*55s^-<Q98*|01&tEk5)Z<fIAX}dx!Jh=X5-C`(wJmub^
z9ee!F%=+&<p@-`*hZ_^;mKUGj?|eDS=<n2{?MFV9E^jWJAragFT9)FFCvpV2^5<}1
zTDkeKOH$FE6~;%oZ{CT$F-38Sh`&<pw}_wyenr(vo1OVDeP89C%kD{csVdTa$yKuW
z-)%$x`t`C~tJbvjFWq*yHNb1-s+lEUgjU|-J}Tn9DPT>y%Z#-lBH~lJ-k86VSve`n
zu5HbuYKv!&!TsVA6~@U1u4O5|bGPbhUb?Z_zvuVD*Yew*ZVNSwSQl?H&CDrKYjw`I
zU;!QHBNk092cAsyoTW6kJvfZ>^cRJlYZ<M1G%hoLyXJ_JcK54aP}U3!X<Itc>bK3p
z71t*_?beT+;y&Tdk_XdHa>Xj1E}r>k;pVS9uM0(PoqSrtQ*m|q%x`a3zdWS&cShe6
z_gGuKGoOrh?Th>3kz{;CY5mu(Hc+ltIKX*gQ;9C`!(ImYHt$E@-^;1)Qong!zwSoz
ze@%mnYAa?={~~lzu&h7j&V;MWx)!Hv9bI^bt8mkTTT8dq#%5&Pk0|`75jELt$t)|^
zIVaW&*_hdVZx_)h`o*m6E~(c4Vo~Z%k<gU)+GTs*eE$A0EppGl6BZZERlWXvUY-2t
z{%XG;F>EC_DnW}nY`(paL@JRKJUAmnXM4t2mCsobux!%QA09!A_?67H|9Vc8^XGb7
zrOO&UaWY@*cBk)Z+`g9j6F0@Z?!I%pVAA3r@<FfS9q;l9i7w+*dCVmkRvNITW}%aI
zQ$5(fEu0hfui>oRFL`zG?a#ZL?zOW-|9X|_yw|AK<-bMY^-|_9Z_<u;Wv%{kd%0!O
z(Kh3=>vqm|tg0527ha~hYjxHw@pF&+R{nLlWw3kG)j+<o#Lti1*^mB;ZS9-B=ik1y
zhCjRK#K~8%o1f^mS$SA!YuvXdQ#v=TI=9_N{(^GUm-ljOPnxU<_hA<nQ0fxfp>!g?
z#*?-6%*y+FZakd3s9B7&WSY_3eP-z=lf+I=wf7SGX|i&?X2zCdFDK{N_j?5$ZDj5}
z<gT(>sa4K((<Tl5Ry(_yd-inn2P!MhK`yCFj2K&M=0uB5)j5AsD%R|-?{T>$9_ra&
z&duAF)%8_R<ut2`x75k9St75GF1qyP|C?C$8kwVC=ViazEABllW#_5Ai8%#_6Qq+3
zu72TL*wUi%$CJZ%ivWkGNAm@z_ht!J63NLu)$d*RM(+Jy^!@)n+q0Ra`zF5&$+`LF
z^ZPx!Hyh`t-8(xkZGYwGy<8Xen{=A4RR8z(3rl@@)Yn<8CftAa=r%kMscK$csJQ0C
ztCe}3nTyvvFlPL8_vKZua$6tS4Lyth{jQQcny>FIcK9LZzxVydBI`@tf5v$3{x!#<
z@T}#>MXZhf;(rsJ*E2Zf_w)Hz{od95$LG<(?Dp>aw%fyHPORIw|L*eZcfV#X*s|H*
zx>jaNuP2v)O!X9B_FwgZN2cn9u3j;5S6x=j#cQ?m>hGAkskr^_3g3U~+MWly&BOOU
zxA>Rv>&;EKi&kgjf4oR`ZEcsYP-bSF(9Y>Cuv3z^;gszLRnb3vH(uR}$Pp}D=zHkW
zHJz4)qH-o*59R0y7N64*XY<}GuwP%`zU|TYzrughF1?SR64sc;c$Nv|{S2`SItwBl
zHXk^d*m!H(wFAGVg;gwgEiarVdy(tX4xTl~q7N+36Kpo<t@DwvHRR8a?q|!Bi0Hi%
zwK&pYb3<C=<9tS!l6gTBtln&%apmyR<NLoo4qiNeyJL^nf%%#zz8#+0klOcG`s*q2
zNCrJ^7oQc=B?RB=`!WCUGw9#+;)}+__*)8gF~JLBzCW=S&M?xj5$dto#(FAu`#<e<
z8A9Lp?(lzjEp7ICtLh0H++RXo?fCV^qVWIe-*+!P&r1vcU9#x*%e?)5C3T)350&NI
zu{XH!lI!&BDYE>Gc5WN<{3G|zI+1ea<~-X=r!$Hk>0a2M@T;f#Z|l+H(Gt-gyH|e}
zW_#EipXvW6h^;%Uc85T2-JvC)S@YzYO!8$miiV~O?J^TSw^5_#esK3v$;^(n#LUB$
z%ocK|7cTbN66J6$RqN2TD_j?|@&XThb?c8x%vmGkwqC5_+Ld_~pR;r!q1u25<{vvF
z7qEW5o3{C;bn(p%`x@DQ9c&8gE8F#HVQ-Rw!~ZpPD_nzDLc(<NIVJ}s9y#56U;OGn
zFUj81&8w!$wbEE)_MC>vN2i`u`kb-%)6dNOofV&N#}xiIJ$&TL#cK<ePCwTtI)9Bd
z*ZEKT8V<(hRG8oV@ICyAh`GkjO9wvIX>51<^LX7xt{HZzE)xtC#mc!Y9{6vWqx070
zV%emx#ft4gA@?*kaXk#Tc37hSqL0lZd`sGn%*HyF%MTmhJvNs4UDw9hT{uPHS!csB
zcaG+R{0FLjY&_jpl+03hL+7ow?ludbx8Fa6^F{{-&beMB-M=(If;aR+c|=828OOto
zffBb*<Zlz6FZ^Ho<uma`sn@}24k>gjcIpP?o3WeCIvZAWe)1yiU79oEKff*c>3K0`
z?L<gAeUZ=oqV%#>(D8b)IlDd|akqNYzR#;mocqLd)%dFNt0v8TPbXgI>-zsMZ_guD
z={N5c&)Mn*t@*LCInM6Fb&Y4feFDis%q%ZMSG|sQb#t7{v}^M2IYPYt-zz5{I+|3m
z-CNO?t8&3h#sBxdS1PCo91@M)v}g0<i8f5HwkDa}TFl%hBKPE`X2+`=ii}B4MH3_p
zB@9HL7<|fWUcPV1-i_C-)177<Ix8pqOx2@*^J9;xKWb_ha&UeqSn2=m*ORTA%lf=N
zh;KgL-np3L&#%Mhf(j4sY<8__+5W)kczt+E*QTrq77k1a-;Fn%IQVBHM@MA9Ee*bB
zS30g*d%5H`C)@qHQJ`;_USm-vd*Ha}y>mQ9r}v#W%5igBz?WRHTMny4>+=^UonPCo
zHD%vvn@=b9c%SPLb3UN>?&(7Doe?(_9@H2t-@Jacv9%SK$mzL)g@R#uMUoof*UJAk
zRXlv0Dlk8ZrzrQu98Ezb*88S!Hx+qr%+0y1Cw{g^Feg4LZtah{>Q0M<>uRr*^nV_=
z7ju~I7t`dhhx4M{g{w;2<kmFkdb7@F+#1td_}c&cm%FJ+<!w{7JZtY?{k-nYwuG#f
zPjkIu|Hwt?r-oNe&0VwW_a2@@7Bl(p{#nPdQSRuBl^OMJE30{rN$<D!m5E-Uf3Rfk
zjPJi0E@s6v%w1EsK*~CHOT;6_s>7e}JnFu9B3|8w(dfaX$80<2u4)e7eN-sp>z|9g
zM+-YME}M4U<bVEPc}1^5)HcO`Ul=bO46(eF>nZx{(=|?Wi(Si?W`aw)0~1-#RCy=<
zSZuw&pv2$%G9n?$S;abc`&Aqi3JYJ=m?t#XlH2!*yD+SjY+!BBm5jc9=78Ce>gBT>
zIjsC76aRj8`4G``XWf^{`+Dd9sR;SQw&&%P!jp|=26N}#=Z`U#d61Sq<K>qr0)_9k
zcznG5|BYy+DsR%G3l4>EYaCp)6w7x_dB07p`yj`=7hQUh%T_Pu$vf4`T(f-J8W#Wc
zOkcaBUTmy?y2qk@&y^=#@f}B*x6Uy$a$!{do*GtuTb=1aZ0F)KKN%?%JsV@UDa<l)
zCc0i5xl0d3ht}{;n6;o#r{S-9zHgwI=8`%+If00M0tddy$KTTT@LV+QB*(L#3>8mQ
z`7Pi3%803m`S%O0j4+tETU3Zq?f=VZfn1?|I_Ka2*<`t~$;9D(?R@>o8xK5Ep0_i3
z3&)4d{>nb<<Vj3Nn^sPI)MlT*T{o#D_4xApwn>-O{g;GWt$(W-w{DKPN2H44kJ@if
zKKLc)U)y2u@yqUZ(S$jVL*E@)kbACG*OPq-+kvHeL95Qp7Fo#fdJ}KVg$EXOPrf#v
zTGa2OzT&LwEVDOd0k0>0kF_jI-!3i27bcdyWZgz*CRY0?>&?YA6LzgWKktg;>`y;G
z1s#`?(0d=%@4CR}(t$7M%mU*}|4j|PU-;Zc^{Vp>ty-7Wii`@sBL2MeWXZ6TT~MX+
zL(%j4qDg+@tG>w1i#o6|Z!JS~rtY=BZVx&exvOQGv_iRxqxW8@zS8+*_a@gUjcJF2
ztm|wepW2DYb=J><)RzZbxf{NTIX*GDh)6$ux(Qca&hE8R*5I6-?HJ23UH*2N_i<Lg
zkX5Yp7fYsD{n#brD%|BCV!WlwlpAD>fq*W9LbBV*y@&U6)c*?j@3z+F++2CX8}jqc
zgdMmXy1t}TIQ-!1@6*LqHSe97creC%8yBM&=aS3K+qvS_eYv65;i+HNG?m5swUlhN
zS8%svMy>PYNJi_}i&x`Hr?bCWBJd?AXX7Eq>F4?^**wCv6&pPB3#ycZ-PC{GZYW{%
zQT@xi_wwHdD`lJZ?YyMEwa)!{-!0A07yJIKXnnij`}h0xmTnu@|C(pB+}-@(2kH4t
zozA;2bpL($+P#D===OU_-YUa-uO&>&g1^V-N$WCgJlx>q|HrCgr#nk@d~1Hxl!k_N
zTf53l-rp~)Tg2<3F6*7VIg;`J*B}3y|M+svxHngJ<?*;3367arnZnkWWqzIL;y%~0
zO@CR;{pzCzsXiNdvpyQUJgVC*d?fb&jkSJZ=fwTX*JK}?5m0!rCnI(GQNz}%i5KSH
z%_#r<J}xZ${y&|azus(|qyDSzyQ*`}eS7hgrjN3>Z+qz3o!GS{Ii33p?+m$HW-$^?
z=U5u*<m(n1T+(&>px5;KkIP<*3%_H}+d5xAVEj)h;P4d1tYh<CTYki>xgIBC#n1TG
zYco<c1*+8-=<5ok7)_~orL%Ra?lj&R()x{g*+;hRTeC#J<R#0BSRHMb?7z{AOkX?1
zueeeIis%KA4?;FF94O(Rd-Ai)m*-O!9hhZyip%d)^(46y?`qPQJknU36=t(=v6EFz
zPZZ<3OS^yd<`*4#F3lxsFjYi|rP_18;_JVO;VW-2d+{{Ae$G5+NzBh1Vy(AY=RG#%
zU3(}r_U=||n-k6-cckr;3OE;XeXG_3J5Bo~f8L+Y`FQab)6(AUbLT6ZYYI1dn)GC1
z!mckYX}|XVzpMQ%?V5&zne7%%soj^<r<JPp-Fm5^pXqlmcB;Bt(=;{tEi&oNLG9e4
zB2|Cw|GZ+H$ezVl)xVs{;CT}Z<JXlpIyXp$sERNhx}-eQM6rK=AxCXS@Sd}hOQtH=
zw$6Ji<dOb=Ud`(bPKOW2vhCk_WtVZIf?27+YB}Ctf~Ul;d_JK4d+PgQ6|uY>1qT~e
zPkHxdMM}`$<Au&za;N+PUfsC7X2m9^7x|*~{|$VXh+pYmp6?x~QCl%zp)fOa@{ix_
zFBk1u7HQL3{bgQy!!3>;es^m&i@y0)&o?mLyU&p&Yw73df5d{ZR(0Kt-Uj=XDfjl&
zZwh)_)N_?TJi44E<a*@F$#*K_)DqG;nm+~onDZ|=a@T|6Gpy5|L}=f=7}}q|wKm<$
z?d|?#oeSOuwxRlI+xWz*480@HJU@8cI%dYhjVE2@Eq2WbP@J&7qMzx;ae2pu`MU!u
zBD#)Bu53SQRfAkC%wjbG=k^QyVi#V15#wLnnBOO#)Ymh2;;%rbbU%j}g*WbYf#=lT
z%7*T_c(yXbTdaoRd-A=1Q$Ji%WD5*=es}iY1aaS`kcOn6759r$W8IaL_gz|=Cv6eP
zU!8g6vRjm9oQG_Y#Gi|{f=v}q?9?^9`NN9l+H(i^b$s!QyY=#cxB7RZmw_A{RSmOe
zH)!lle9ai)rO9k@|DD4+m7R;dUv02I=3dqPd!5v>iVL6R9OF)Xxu|GnF!9b!mgSl;
zDnCwFXC}Q+uosknsoch#qq^S2F{^66w<5cBtHLko`%g|i$@`shI-~s2nz>~+4>aFX
zkgI;YAn1MEg;N14C%n0&nQr`4y&=M`IQ{p2yHf)D_ZQ75vgH@i5b{xH@K?<6<Ezm#
z3b-NO^mUR@+ZH2{D!2b%|C&6te)e^`boYU+vYKv;8tSn;N7w(|{nT=QanZeCzLe`7
z?(z01dk_3ycjMwTN%bWxO7pFc&U@tQne)H@Z@tii?J+V{@(G{y^?icgZJM9MeB0?K
z&!1P5?x}Mn-;ncJzV6C(zK*x;YvVqgubs5=i?Q07Xi4EG=Ffk;<u1IM_sX@!&BpxO
z*WI7@9BC|$=Mvwi@#oz21o1YO%gP&nJLol>*mG;+X4W-}Hx<u%9X^@s1*2co^qn7e
z&Fo(+>QNqhxbVcv<RwA&7tYUV-{vRGR(z^(MaXI=NGb9+ud#OevXV2ee(g0|%LHp%
z6kKI`aL(k_8LQ0^e_wC~@a=1~u)BZg<YR?X;Z5IIFV4|nxfMOJEjWDZH--52x;>hU
z{g&o!Ys__=eZy?q#%s$1L)!R!-v5*<iPl~O=|Jq;+;DF7maVCA_g`9Xob`fnE=zu4
zSFM42c~qJ~M5gf<Etk#^^}ZK=pBUY|6&Jkg`Vu1Ucp&`0L14|78<~%$R;%AS;Mg>A
zecjZxA#)k*$~0Npt{!q_tV;BrIPbga)Ufke%kI~D8ZpFW|4+3m?_am&a@_Oo`Kg)?
zyL^_vRW<PPujgEH`nSE3hT6(NjX=dPE4eBscLfgZ)Y}Waq$jLXd0@}RHC5(UjK_kU
zJK^izSOn<Z`hUBPVTnjp-)|)b4cmV&@A_HtH>SRv_|D_b*2kHvVnt^2Pq7YKxukac
z+9PaBj2-pNKFpbSN66#(*MDpHuBow|+$#QRl7rRzLZ2MF8-ndK?!Bp1U3Y8k?q7E*
z<F3s4RwT~!Rebd^i<Jo`FK;oe`#Z_<s$jBH!+F*jGm>|%w!hCZ$Lzpf=_*6nIExso
z-#7L)Mdd-_hR256rRPGwW_5y7(u>E|(|Gq>%hbIDuJaF^zs|HZcdg@x93$>yqMtHu
zsL%3sZ@u_IRcU^&0NdRPi-j%8uWZ6z=$_m#N&VTSYv#B2nazx2e9O6L`9eQ|=A`M%
zZf+9JYiZxmzR#&*%QWp=ldba@j|=-fvw!_sZO^W?DFvRstG{oZeAsBwAD`OCr>@>R
z@$bsr>tzxzeqHjCs-5<BwS<k%na!UTT~u^uuc~H``tA30<BDG19Zz0<DK-ig-skIc
zcGIrEd|Mb_xK*qCH|frK#Wd%QpQ`N{%_li(H&%#UD(Z1=aA7+3ByE0MLc6VKdFjN@
z(M<<7`Bh!CXFi|xvQYP6%02rjw;9)*`g7^?oa+yTD(qr3^O}wuRbKb3H{z?U`Mulb
zBQKM~t<v~aFF(ETvVJR?)4qLY)C)!3%C($Lx4CWKZswLgvcbas%V91CZJBRxxfN%&
zo->bTX$qga=aI;zdpw!T|4%4gA+EAw^SX_U3?-_2{yB@v{Y-a!ySd7}6V$zJ?Q@%U
z``I$t<=fsbI$;x4-ow80%1Py)Kd&y@H!VctQ(bATdt4p=w?%r|@0UI;zhUmN)!Xsj
z`?D8Cckf(vf9mt4Ax}f?IT_3A51bS1ikb0nXH8hey3S<oFG4=;LEIg&R!29N#$0Uh
zc)Bg@-OURwkW9L~i?w9h=F{)wXN&(j@~(He*YquQ&}uwoyTj(AZ&Dx1JwGrlV`5l{
z%Caj@JNSb)9ep#cXyNw^<{h%Vc1#ZGbL1JtEoJXM@tG0+`4+<~(Or8!&e_shrESV{
z?}0Vvvn^by+|M7KV12InC*L;dxb3=++cF-`HnOvNxAC6pmuHKeu5SyhS$|%1y4<Bs
zT?fmgq{V^GLhnj<+*K@&Ke4XK>VNgV6T2mLmF$!I<^OJu+{}C8RVx?UmwpeG_E)pr
z|6A*OPif@V<fK2oM;R{Px0!jXZqe;5d$IgEGV^}EY&?~b&^xXBc#FuK7vH{atIN&N
z5c|Su^85V3o8f!6KJV1i*Nl5Hao5_p#}=vQa`A3y+MK1u!M3L6(=`#6*ISBWzPqf~
zxf~q(CcG?gFVlz1!Pd-El)spST~0c&&73($Ewp>@ivKre-F~=u%O9(c-%i;Sezh$0
z`sPtl_5a(Zl#LZtQz|}8WuE)x_Uq!3w_et=Q-hbcu4I~Yi}{$*{Ee~})3<0RzP&H~
z-dg)`HOs53i`dRzeQ7iO-r4i(Pl!A@u{`%gu%AHPwI2o_4=s;0n^#!=#-e1x`n9r?
z_8r}qmwtV1?zu$!-$&)st**RO6Mdogd-cBivx~#uT$@{0K5u&S+Y<9DmCq-+2=d<&
zFFK+>*V<!c+ZMZ44u#d8*Tr7Q&UiOJRcu;|@`DDMLp%R9h<BfGwEmqVWdGpzYSRGF
zP}lrc)&pm&OCK^PcQ2B=KlkV67aOj5Ld%H)Zj-H%HM4G=`KQoc8+qioosO4J$Mwlk
zTGy65>ktW4c(v(EQp64KMLVvl{Au0(;aiuWuG+cfhaW1+ap`?FC^6*GE4Vi!>GuIi
zmfxEt**#9|wC41SW^DSLyIf9mnYJwKEP(Gj+xLm2dLO(P*S1IX(3MGgQc~P5y%_?6
z$HWSEZ1Me}Hsh$B?Wtw6Ts^a9bmU8_eL3bnrE-t!f#SXSPuI#mbzT_gZgp2x^I&D|
z%DtcjV_kn@1*c^|jC{$@9UJB-<TLNz%&2)bH|_4XKUF>UipP%sNZEb+L|2>GWA|!p
zc2%~@Yv*L85+1u3OnbNP-{Xn$a&va|d`g?YM4-PaC%?S!kVQt%-w2fnb;rBjoStcF
zdFAubLc@K>ckQ#fP%N>VIc#aT>sdb@9Vy9|q0bdg7y7??d&A-#@4Mq%w{?D1E{KSk
z`sv|5ySs^(`+e=)O!uBV-rPLPDN2AT!&Z=K^*)XqPt7cTos*w;>q{mvUvey0<9l+Y
zBPhVV$BRQi@Iz3!d)2D$5Yg@bbHk^HZp;;Vc{O+bOu4&3OW5t!$FuJ@vAvleH${PM
zd+^Z-@qhcn`Gxma=dIUybm`*Mva;nXLZ1rdmpEA~SH4u$?Pa;TGP-D@^{VBrtZtLF
z(po1kn7yQBV%!;<dmrEE72f?*(W*TE*UO)^>N`$ve!#ll`0T65_p4;D?$$7SSMK-m
z(TcydbJa9+R&3Ya{dGxP_}gn!^UCLmdp>rq<%z4--KW2GPqVCetghRN&;RC}f7-IA
z-R#7nB?gJIY9AVByhy&ZUV1^z6RST{4c<GeUtid3lLf8SgO4$Bq^281e%U_#zun8u
z3CoSF-v75wty)<MEx6RDvX&I2EpE5Wi0v_dSmAKS^DM)^z5X+#J|Fqg!Zwl1VYbG%
zCDE^4CYKv8+MzfrVzc~%qWp_0HLWM#XrHMPzO&nB(b*&?zls%)ogbbHsN>ls=JGSE
zG1<F2<Kp#$*Z1;OGOnq<(q$E}W6Ra>?z#FxM_f(ThU|@qf32&uM~HFRa)<Cq7bM*u
z-Q&1(_V{s!tF@Zho3kdZ%#WB6o|m)0eSX!x@5|rq+i>vTA-4V9_q=vD?)o;}WZmEA
zj}Auva>}oN5wUHz(zbm^*G%BhmlU1yEKk;4@ARzAW}C~_v^!p(^S6ED|8E8n2Q$)N
zzI*lF^WUpuW``s7<##>FC@t2oQ8^pl!+rC|l|wt`X2qs0@H)RvTk6}rcU_#4Pkjv=
zu74?bl+#*XcXR#uo<ASkJRj=m>fGpet+0KSVtZh5&U2%)Hw!;XyHA^TZEIT(hyM9r
zFTYJZJ@={gyqD*W9jja~8&|MoO{aBE&ZFEb=Eqs9ro`DQmb|}IsQ7c6^zsv&bq#+P
zwarRfXszvkO|RFVFNCRW)8E-ujDC8SiyyB(pZnZd-JtfF#X4s0z`)Bob)VncnpN>X
zQ(y3E)t9Au?xkL{xIgW)=2Ba^tTFJjN3eKvLY~uvb3YeU?R>pRkxTScM8K_rjSg>@
zzF%=~pP#ntg)q%YeQt557r*iLTxy!H&R=W&zBsbYee1jY?IPXN?aE(14|wR8{veo*
zVQ+forWpoTq!yIiTT&oB!|(698y~lnRVh?mSZ#Se@?S-Fm0`!@-RoQLyj*wVY~bE^
zcKXi~&09sPUViQuN#?rn``_F5MMpQ8F28er`<uM^a@Fk1)<<5}O6%@DWTY%H`J==h
zew75Ja|%m-*iFe@U3)9y`&2=f_UnC*HZH#1qxfmEfb^Yt3cKRhhd~>^$8R!O31)oz
zu>bBPt$7!hYVEH+zLPPk6xKmnu#MqD9@G711ND7;pC3Cj?D@_p6R($5y7tlRJ6wjE
zTLY)0&UBr++?D6k8lh~tCAX(W%WZu1>&>~+q_S0xPZW(J4t<@cWa8LwwYNo2C8FZR
zaqimNM>FK+B<LJAtY`1YYc$_@ddB?N&^~?R)6#n{pPrReSzb0Ne*U@rzk|0wzPDrk
zthjqAy&u{ah4$uNJCXlxb>VK8y3ozf?_U>PI-_TiZfV||yzo71!ev&-oa*{x`=wNT
z8Ou&z@1U^zua18@8}Vgj{x%7lJ!?M;q`y3K`~IX|TJ<&zMK<*!-6wSN`8Lig|5JZZ
z`sL^A&#&~o^I#00<SJ3TI$Fuoe@gVtKWClOkCwgcxH>&~<En+eoj1a-=h)>vcD-Bf
z{=Dy)o3@yu*^?`lC5uk4&FWj^YkxC*<KMTZg~MN8y{IW0TCi#9%cD0}2OoW9ss1u1
zx@oeBHRHp19zk39UsxZSx@c`~+Q*x}oSx39Ien@7_w2Jr%@pREd;YvQcS4mb)4sh|
zug1nM4VBH_y`{5O;F4tW2FEqAT@Q0}^;7D8O4Tk2f8YC@FD!3z+Tpk>wVU=|7r*|+
zVA*tAaaNVu{;2$-HS2TT8}2wdD9m8cVvK0*zoR!xf%EAN`*%?aciGDoJRD-y?%rLj
zr(^Ts6XQq6S0WEz=ZeJ5=)2(^UY3$@;o*a-<vFekLZ%Au+dIo{|2!L|h?}homNCp?
z$}7}-s&%e#=CyG5@Hc7x2Oq!t?4%`k>f9Euk3Xc_rxf~FEk3kKjWy)`>V0!wRrS?A
zy8E{9aE`pq!4H1<JK|?V=N%VG;f)jH+r4^O{ioUAIlOOsXI!6q=)$K412#EPUGBAt
zoxhhm`L2d`s%Nln=$jdF&+6SQ;cE<QlOR=khFC!YW7E}L3qKfJif_^VFEVdmJjZ&+
z`yC7)*GR~eukPUA$MgB><)eqHlaE?{zrhf}-WB7$i0{=RzgKCpYr0D}fx2jkvqco{
znr?mftg5#9rof*5&#{M#-X8g2$9nm9@b=w*OQzMI6Zn7WVPf1(XD6G_dJ-@9ZQpLZ
z_wLg5UaR|RN~2%S5jUTD&N}8=f~(Rl4VU+Cba$4X;%hqDTywdKscxRR%?pPu`!`m^
z6&>I+o!h(li{7lu>f61tqJ8Awt($Rc?xwe_*Xr7TNNAmxm)!C4(?-SD7y2iCE7~>P
zZ2CE!HMuW+WO9~o(LerZ&qBuC4&0Ng|D9N1TXuEM1n>0Z#djwLW?t?nSih#*oqfH1
ze#L`l69waD9(h>t`1Y1po&4)`B&waH?#-Wis4}(W^$+7;o4fX1sQuei@$px!@X^|E
z-q%a~nATMu<(+%ro!`tkF*6u$oMZd{b#r;|*2?Ym+qWKBXui#ALZZ)#={suHZCY2=
z^K4gJSoEz29;wAWZ=>ezSmph!{OJz+)y<WeFXu!YOqd;{#>XuZ%(yJ*lecsW&!v+K
z7VlXuW?H^2#-m`Vf0pcCOLYlfEm^OK1<Q9-i2aGa-5T?9$>NEd>hvWZD6lLwKB(;9
zd92*Izfe<<Bk0$k{NvrX^*ur+@oi-{O?9}wS-t$p<exMB_T_TfZr5+tt=((oP~4_H
z<<FB#cI$3Ubg*yU8`NihlsmKM$D0!#rjy&*JEyp$t_>DR@DbTw{ZqU5_0#%wx7IGb
zn#8<I_2`7ft70qTQuZHHVePV>ro#EHzmU&QUUpHzz4=+$3;Nz%5zmy}FJ*hF?d_tv
z4?9*~o!)t9myqD=N&)3OZnL@19>34fl@NN#_@H@#=e{4CPo0c;f3Q0J?Un=FvH9y<
z6_#mTxytigs^WXZhN=GBgC%_0kE}9HeKcqDqZammiW!@YO`~=}({&d2i;~Ny<fP`C
zmpklfn|o#26{+=*0Rp=%3>i@ln>+ar6cwy|UBX+jIJaf@JjTz@vI2Iybg&7RSB4#2
zEK~eqo0ve@p{G3My=`VsSC}X0>)l(XAvR(0tlK3DX&05}J6cNdawK&58}Ds#JeYBX
zS19+#i&Dk{V`g)!j98ttJztyJ?{<}IuKM+Q|J9dQ8D6(~aD20|ve}dWbN=t(^7xyF
z&tJZ<?O)%-)4#mp>tFu-lu()y8XV2=RbNW7Vu||KJJ!~}6zspH&Ru%@`RmuOue|3{
zo3!oN?IrmiSVV(T7jLcq_SNNQ^_24-AAEnGZrZW-QuF>z3v+MxJpUda8gX07P4>Rr
zs;*sSK@okmk(GC+Cx2YK)YIN$FZ06}r*2r}PRPG@U+j4P<hOs{A6vA9eV=Yblf2g5
z-@8vQw==8{n14KJLh0NUi~Z7P{y(_%(ecxEAxyUOSEQ9hzMo{gShPQ>X=(j$1>XtG
zH3dty3rS33HJ5$hw&LB>OE(RZ-W@S;Ts7ebhx}3IsA)=Hk6Ew&^kgZM&Fsc2m$>IT
zXk^B|V6DCPGO+9UYTd=JmMmI!wCwBg)$QH$zfIw+3cmkc$gF7R%G1kEv`&_FU*%$E
zv4_*^SW9Jh{lBa=5&O5xJoM&yw5Idnwz|4K+maXaH6}6rdwt;;@9$HL2WB6+njUSw
zU%=p;X3`6xpgR`tF58}e-@3f~yv3{CuP&+Ei`}SrRlm`~u}9_&|AFm+J}l<{K2&Hk
z_+(8twf4@<vb!Ik@ory8q%*7N=>Q!e7R}_8b1ep`cP^Msd1^1B@x9_)jV413=Z44M
zgM6e#7wPEb9OG+e7icI=*)xB~(>*MO7tct1eChB0@aY|mbH|um7~Z%jrdu5iwLaA<
zf2-{EL9eB6KK{x7^&;To{`>wS?Zw%m>+3n+)H9!&y-y_k?J+BD!)+Y<K3mtl64&Hf
z|5MIjkKS>M(&v>onX3KTmfM!ApE+9b++8;&r~1>}!(R8VzPweL&i#eYXFhMjuCp}@
zKN#h;-P-!7{rI*AJ6z}dlg?lG-9t$3-sg{p449&$+z&?A+H9LF6tP`TzckN1fBhM*
z3HO%WZWV9sb;@T<-uE+`CFGRL0`=EL>7QcfE!z20=TGeOqD5@WzF&J3RSBvS*w(N<
zU@Zu2N?&>+>&=ZmvjpkYraq7%1AA+3hBMn5#ADT8+^uLkv@Yq`mx%VvovdFZtcr!7
zol$+@e4PEDYw0elCz&o*_by1<Oi41+ig{J_kafeHdv7>`PCZ!qtbpOwk{A=FwsyZ~
ztshn#_+aGd`F8vN7m_s#b_8~2l%_Sst2a!$w=U~_`m?*0TNE>VmXwEvG_2gY+pToz
z?9{8jD*5g*mg=TG?y0z6RJ!+7)s<9<f{)AB+!AQm@a!e$KfX!<kEKoje){)*Ud!>}
zP?Ed&qHAw8q<8K$e|oL8r~lhJp6|w`3k>><wlBXJU1k4ks?57kY56Ay6}H$;d-{8d
z!p4fZmrdjB)J(2VRQi@!9M|krV{LeCw^z}k)t{|iFW+1;EjPpLwfXa_npMmBKCS)o
z?LzStE0OS92b%um+1-pcocVlu58u5#Mr*UC2rZk=S+?ozy1@6hR5`1z&9~RexZXY6
z`{O;1_2)0K`7M)adU?0kb?duBuf+OeT}vAuFRER==+B(~;`DEu{L?oCU0|QN?8wis
zm9Gm+&OKOl`_t>D1MS&z;*VqBzW7}na&GFz#Q1mHGABx&{uU;4Gw_1JnneqBCzx1t
zD#nFPxoG#<v#NLQ0>#g6J@eTnwajNr5`R(ZwPEriJ|)i&!Ck9G^iOWAW7wX$vhVS}
zGg%*h%x<^(n02|h?CLk63IlfyMyacNUdgIj&(F1sNL;M>@w;Ic>vC7Ed)sX<UX3jN
z9<E<IQB`hVUdyF0rdQL?yt{MpZM(LpmiWq*J8d;yN-k)4=&s9h_oSt7;WhQrfBS6y
z3qJVw?u(Y=y)veKn{QRVPL<91Za-~lIhV%C_niyMZN%6;1P{OW(KRyA{k@4-M0@gs
zoo8Gpxn5rMA@KQ;D~EO+S>Nz+k2BXKX)(TcQ#wlwjT%;Jm#LgMZKcV+p?iVb*5-wi
zAJy$!@Z8ous4uDUz&s(3d#_&xvh99*F7L{Ixo(S3J0CG|s6Nh^TkyToazg0ZDW%`;
zE4(p#u(|N?{Oz}oZQZ=b{`r@CjW3^WWfOd5&c!C9`Kakxr(=DE+-!qSM`j&RGhC?<
zJY$c&z*>e2Uvt0s@CzK<=3$fj<=DY%!LclA-|8%$?-gix^JPk7dEZgy%$nI7{C?{)
zIZl0%$#=fkcu&QbXGJoGnFhs=B3i0v+?ny(W%Wvz7qeHldd90p>7HG*bYD!)kLuYo
zYRd(ZS{KCqeH$RzE8dr<E@LYxz}_WQdp#x0)vxcXc5!OZV?WuyD{s#&oAaj9)cV^S
zX`88Uzn$yb<mj8W`QfUM3oX~}i}-S;>-csNwV%IS72UtwpQ_IG#%f<^_EyNk0Ef*E
z>vbAtIE&db6nTpr_z-ZW_-EO@G+U(=%7<qK=~N_dvsz)Gp?T`STo?ZH<!m}f>t!`|
zP5Qm#|My?tE~ymxZ+>+8c<Zv0Y&F~V8+~d^of+=>OWND~j@_%b(RQb1v7WzYc&58^
zzvO{}i^klK_sz97S#Y@3Pb|;g>hjs!o8v5Yq~r%4p1}V6SdG-%{8jZSq5p2WZe7du
z>S6Y!>*qbc-T(b-si*y_&{JlNE^GfinYHk}Zms&;q$*Y8^M@WbMlaoNaPW+^-N~8X
zYSX^1*tE-R-_che_T~3|F`oAKSI>&A+kH6CCrN6uJQbdIt2VuO_0h){D-E^Z&5)a^
z_x|G#%bbrF%73OeCR;J}l{Xl-w3hwUExDVr+<RJes<-DM<81D9ab*wHOH;~KFEl-|
ztvt8jz2m`zeSh|T(RN78neg-Z{-P&Vs^UWL=F9%QI=3u0VPCIh*s95e7wsg!C72yd
zViC0EcE~9To_%!k&zEm&51u%X``^O8Zt})j{U7V?PS5-Ft?$#!LdIo_=9c%@b^p{j
zcZpNy{x{ESZ@%8#y4d6U={-DgyDCyHToc##x^W;|bKN$_XZ01S2cH}Z>lXicUw)qW
z-eAVdPkqJv4gYaGc->TDDwg_jmfGfzo3>wmzV74-xzqcK&KzN{l>2*o(T@Cd)t}`>
z{PTW4I>=onYNj&ZSSR83?&<uI%B5$H?YP%m_;<Fk(mub=51REQp$G5GS@q20&h^5W
zcyrA+%a5PyysP{9v~uH{_?kJz*UmNVQv6hDlBCZql^D&iD?jYZ_42bFTWl9^(&>-w
z<t?@Q{dr+xoGQ!Zne!sAy{TT+`0|Z-Qj*Sr{*P65+xN(M9NI6${?OLv7F$!j(3;6L
zFImjWms_)6E_ygMs^7-$!KrDNZroW`QIsyTHhfR${yQNZt=^T}AOHB%<@1og?oFoV
zpUzEOw?AAv74z|RcYDXHs=QsVG(7a*{jZs{uwvuqhYL&&RPPpLz7`w%M*a1qgtc4C
zp8wBxQn+}3{|5g*e2zxRjLW=NvKIVi-m~NX-$FNahF??lHyvHC=NDviGc~TtS6HR)
z>*~I*+t&R$QuJ<HaL~yOdR6kbx%abwdmO$&KeJ-b?VTq*H#+5isMk9$P`&ZTLb0O8
z|Ih5Jk2GKB6Hj*hs%ZH(bGN>^Xy<gh)0<NNKR*5Y$l7k<qdOyBOxRU9`R&cAVXF6z
zlzZP?rei0fxy8_pWz&>N(J}pZ1z$R`*_9scOP<6e!e9C7Mc<=VhfVFC*E1h>JUyZ6
zUp-M<uD*R{itx0DdncVzW4I{$I4~_@=T?a=mU44<JbJnC--&E?mdaOGAFZh?owx2>
z>{jX0J<SgmXmb7oO=8aWV$)l?<MW<7ts9*lGF@QSe=lCip>*bbnT%M9N41FLqooxG
z)cHS&W_hVBo$zh`Sti;26Sk?C`#HTTJ@km(Ki^nw<*y9Io|8`Yi(TjFNwuW=*zP&>
z*=ozuS4VRMmCvkYxh!n3x9{NTCi{lMX~$ba`PWOa>0Yg5C<s@}K5;ydpI=ngrsToO
z<c(Jr`bd77&M30pJ&JLip`q!OL;7;Rb^oc%XJL!|B(k|)oa?wO!;2(?FFr?8Hu0uf
z*?Y?JzPWl+-|pO;{p)ua76z?(zc4>v>RaXdDK|ym`QI%I<=x@EuVc-*6?~r!C2yVc
zd13RgWcx<@bn9K&+NE{XyJfX9zt{#{dchudmMh%m#Fbramwqn)t-bi&FP(JF+f!uT
ze0?>uOy7ig<yza!?&bS@?+9&8z87C_{Gz_F{=D^4o2IHH{<S^Ot_2sj#_tZ!70d}&
zURpH?Xh}3@ajo{5nA%@3`FmxwXSmVn+3HVreDd%$((YsW|7A<XeXpAxvzD0&&sw=U
z&*b;Kua_2j8*8m$^s1D9`RmQ8XD?n)PmkO5XVsJPxt7Z<F8j*%u1&jS+139&bjch2
zJdY#0H@)qZyZ6#=zWj<=RcC)o?lPaQwl>wSD@VA=t}6ax-mg|6kLP=LJu`Ea-+O%e
zp28<KOmlsvJN<RfjynAQx7PbbTP_{D>aW{#Tk-wY1kNQZum182zM7Ideb4thzZNO>
z*Tiq&c)#)5-}gHlcZ)b(m*018_l89azn-p6b1T2)S^eL<GCBT6h*AH~C2}&QnW?^o
z*K1~M{F2}O$X>By%`SBnhI3oGj$QNnKI2=mn=eO0W$MPnha1j`=zsZO_j{^y_=%$&
z&z75PnX&i&?bt%LJ#%i)i<Q|wHPf^=_tfs_*wr_G@f~;R*!*Sb`(^V_yYgr2z5dS;
z_iAHJ{?U?#GqXJ>9u6tLb>I(6|9*=Q%ZbmukGJM<eKXto{z*dD>q)_PbDp|6O!SE7
z*EVl#eiYm5I`i+dej|OI_2FXiQ&jq9>h%16ZGUD*$<M1N-~07^-MTPFVsY2sENN?Y
zh1?AJ>-U#mRr#Op>}iv$`n5R5e*4BBZyfY%Cs&<Guz7vH_UnnDpMOvNktjYEw)toq
zqk=~cdzXgs!y8v`1f&*MbhXM#N60@(eo?l)(^-|-MsW2^!PM{lvz6xEI4b*#+3#It
z-g+iBF=m}50uSV-F4$nnSrsG9QL4tnRCVz0>^XPe8SP!RFVyD8e&()nqn)$8SI-h%
z(AUZlV0=RHp!-&XhL`2Hy+ov>mZgha(N>UmOfN6_`*G>tYgbo&Gfgj*J$LSVN=WzX
z`FHrVgcf`Bd=d>^ux?_B%C}UGs<`$&Z^Rp~RW<yxjHs9pcjwigbH9$*yxaEq@713l
zCg=w_RO~cZw66PkOpU_a^PA<DX<jj9v{RcnE&jIZS{}z6|Euffe_-tJ%98z~*P){`
z?TYe;6;T}5{)>03=-13zS&`Y^96tAcah+%K@;!gz{{O0*HEFusd!7@gj`lk3W>~lG
zMqtpM4YGINF52;9q3<cbo~$jNx?j@vKls;sYEl0kQ-$}Rt)JQQiMpGuWlH$>Ww$%a
z&dbezeU*gEHnLw_s+a!t^zXZ}v0J6T7A$<$xBq?6j(3)9J0%p}zJ9g&_&*PWf5kS{
zOKc`Ntx{sE+gnz*XOal3mn5&x^lO&iGJZQ(z0uts^x!A2pIhRLj+qa(-`Fv$bAfw&
z=gsT+LfLVT<pXS*R;D$)W=)7VIIs7CN#N7Wxp(-uu69bzjoWf1O1&-c#N9XPw$I<M
zao0U`YTNdsPtpY&?z@V0NuAbF+Hg{5+tDrdZ!h@o{1VixaWIZ&?oX3HH+KX`=u9@S
zmF2uJH{tKSGT+*&2Vtk1Rj;jGGnuV2dit|pE86~Dx$ExxCfqdty-xT3=iPg}BTnwS
z-}f&({I=5WERX!RJ610^_dmch_`sx#j;vYyY3VOxw?1{b^SEW!*I#8DYp(_Fmbv-%
z)ORDX&#DP_d(!R&R-U!cTJINKxx~um#OVuf-aL(%`@LdL{hswPm5R%rUwX;8Z>~*q
zpZnkAi}~!GgBNV~f8i@O@w@z=6=m~w%Js9A1hX?sYQ~8%uS>ma60^JZRAfW+qTUZ3
zLHX@=Ppwv#dTe%RXSGP*cJ8qCrC)E>Y+HGD!|n}som|WLzxfw0nGth^;ZkkQgV5^B
zPd1bmI?9>mzt@m<-yl^icHe%j=8Rt-O`XGUO)qsdKkc!g%&F)5LuvKa_cww=<l7u3
zrVHHp_bTA(lfA9Hx82<<udn(lyg+9~X6Fa9S5kVC!kcz(StlyDb)M*%&o^E*Z4~nG
z=lXVLf1%flXFC|q?|Oe}_2YXtBh8xQ_OH*?w*9>A@*)1R=dGvh{yaT?sFLLpv+Y;Y
z{&~hPR{wNn<}&o}swqhQ=k%%Ce*40TYjZCv^?UdiHl5<ovJ$S>D_xSjvo`9-tir&3
zOJ8rwUUdI%WSte`)s2C1Q@@;WzprF>FMeUpk^Ol=FJ?*9Prh_s`&et*%HO9ye=m#H
zjC-*5PSMvR6a2o*M}`^%GQFIfT+1J4Zdf9IK<w{pYp>()buuKP0`4ZcK6^8@thjeB
z$HGs)cN(Zx9Nzr=Xs%X%hVkFe%NGS|^M&{ryez#|ZuO#YJ1_rQof(JT3w}PEbt9?V
z+_$i*I=C!1>9;R;nfimjUn2@jb%Z>Q|JnWSnP#H>F}8{;dpGXQvH2Z;H|1OPqu)&{
zPp_B!b$8<}+2tn;EvznSGbfkui}f6`yAf+ByR-C$q(LO_GW%Ux^>bJ*<<FJc_UiiO
z#ozXC+gozNd$#wv2NK-h%tG#(TmRbadbnBU+xpcTAMPlXUAep8?8t|%$6Q}rw;gTj
zld-jbY#i|Gfkf)TCF)UfC(rfi<$X0>S5?4c(3H%2Z_ho3Q~i4%DzIkH=bmVzaIoau
z`j->6Q$KYmux30nwcYeP`bDdb?ZnjBJ_qm0{sxo1FJ(Jd@4E4PofxOg%W0V`)5;sQ
zj*2dPZfAeDbF<yoC38dH-aCHxDyOi<f&B8HSH5%3dw*<^Q@PWFn^X5q7XRl^=3D-1
z*A-RG&aP9^bM^90aJ@)9->+Oa?cAQ&yOVcu>`M342~xWyyL;7rz3l1J*WK#%+Tn4q
zU-{*`N=}BklSE1%#Lhg%s;^lqnGyQRW@?}8v^T!%OXf__Zk)>WK#)1FM3BG0*6YQi
z^sY$$sZ2NY8rv-*KF;2dYP?d=+d(0|>ejE8oBkVP=YM-~q^J7tkvDEt^P3mjnu<71
zUF5n_vT@oM#t#+k)Av}iDp>uWaaj7^qWAh5FW0_Z+Bw7J@anjTb&}%K?nwvVn*P@R
z?zd9jBiU<OZlxLDSe0yDzJ2Z9jmnIlck0XUn_4QTllkM>!pe=+mG5-se*YEyBdGM@
z>`d<mzCsPT8+E^*>x-&AyyQbp>E4$g&Uqg7KH4@Xsz{8DL4A`;g7&95OB$Nb>-F8%
z+5T+R=^OK{H%Xh7YutA}etu2u%H;L)V`@vD7~6}@J8c~wWBz7pX0!F4D}85cYl5C!
zyUMxI?&Z20kA7@f^*;L`_rV#P*pJ0($ld=Mct897&SR_3_X%Iu?Px#lvO40D^ZV)Z
zuCJSVMdv~yW7U&Ae;=egWM9Ab|Do4y2d9}_)G6?aa#;RYW~<!0^s5e`#~#VdJ9X;M
zUjN;{znv?Z{rrt;a#Np0u%E);zuJ@TP2Tgw`*_=vdEaE1><?L27w6qIXqbB6n?H5s
z;n2Hh>ivoid@Elsv*}ipnTh_5(>H!xxpl~O#{L&^cY<s_&F^M6z80+hQ0o2u=_)l_
zQbH!Y{W^2cJ^8mAiwfKiU;cM{cDvqWx%1C^xwNE2GFDX<`97><IX*uk*5cvQZsBX?
zL0RP=pUhgU$r3a%r}F2bBmF8%&rTI~`s`5r;i;_lCjEBx-TzNVdreaLd+hOQZZ!}6
z@B4Lnq7Uu~*)8!k-dIuR?fL9JzoZG*Cr-3;zjq_-mde2|(rT@8a>7qvi(LJoxjU>-
z$D{MZxoNeF;(}J!)NKf2lx}aYOsy|;{v=+s?nuI%{<`V*R%RDWwM1S8b893Yy0u>S
zyl&@x2X)c9^XGNCZlCKv^Eyv@NAaiI%At=V|IOlDDE;<$-<8{U3v0uCKQvA{cJPe3
z#f`(6A7a9fD;@vI9<H`uy*wkvN=>EwR`Kuk$DCXxc6@42{P?1}d`@f{&$rvZUzR%W
z&wTG5mFV03|JIpX8yD1eKKjkLPn^~9Xj3~=Mt{$TQ#D17cl@6HKm6teGpp&@x+PB(
z=l{H;_|NrR*-R^G>9=34YzwX*o%c;v#@T5P*PNr81*;b9mfr5i<aU4Ap6R92uU`Di
zeVEVB=GL3{YsJ})TivO=eWi0o7yEHvj=G%bx$a$yt=OHUn45w;{$8BB{JlfQ`fF|T
z*Rv)Gm`tB@cgp14HIr6laP0YWcVZ8dK^U`-O`KE1oc+vqI4skhc5HgFO@FgnrTX*t
zN0z?b$n5cU3a4F}@0{ob`3Ju6{bg8FdU(zBU6~K6*D#69YPNhW$neVgV0vA|7R9p*
z&I=^}OAxUu%{}xYS5Qs9Tce%bs`cLUZ%*sKngyI{Q8sYm?(6-KV*aA{hJ+;FgBp!j
zoK5#q0*?1E2cKcN!a1*`V9WN)>u#@p{W^Q+QGL4$i!-Loo$j6g`j!9Py}8lhPw(Ff
z6l?d){ctFeapx!1m8<{!tzB<kX2a26a&uqM)i96rmqz*Zawpkm#@%bVs}N=xvBqWA
zmd2O&WmobY{&w-SRFq`*0j-FYVcYok?_O;qGfyY*{(YlI!IkH4ZZt0oetl-;s||^M
zsgvJ#vR$q}^XkU6ns*x%eyiG+UiR-S{<CpTeAWLQdp0m`5iNb^@_P4{#cL-nZ@IYD
zZ;pxBgCi3itXiYX*t-lWBEsWhn3v9tvi4Krj<^+fq~_;>jO=?_Jon@mKm8&%{Xp9M
z(kH8(r#i3fnYaFPJKI!?;F|w)R^(qPc4?ZQuFn3hzS#Yy&ZVgj@=ISFdg!>I@Qpxt
zDL>yQi(kJk9Wj+vZQb$v%n9f7>-1;22UL|iGAo~zWnQK)CAcG$cgf~I)7!Nczc_ii
zxPm!9y!o8Z@3)c@+h2J4G1T}c?#<0RR^7*<kg9aKy{+n*mqP3_W>e39weiod1zi6(
zsV!iMcCkpqWG(}*TQZzWP8)sMQ2+YV68R@vS0A@+E&cZ(`}w>CL;an`M^EosbIXA9
zUc8}R%5s)VOY>evD6PM1$yP1BboUxQnGc=Yl@pdO{`79!!-Z3SE-!RczH&T(u|NCh
z$D<SPy}J|?E<3l;(pctb%Z}`-mdmYz;m02_7bowV=3rH7l<;=XLdM^m8O8JdC8w3%
zVA@@HuBNZ$@vp<$?KP|SmA~s)pv-c4=Ct0sazZ}(t1`nbTvj|dZ|=N<mCj7Xf!h};
ztzC2T{6)X6b;sIv<;{Dqo7eN~5m%Xb#asQno<m8UO{@%)&Frhn-bnrav;3B6&E6L}
zCJzpNNwDWX`i14}oQt7zvhMd@n?Lz7lOm(pH{*?b2SfeUER%gYDjwVt%rpJ@Fr2^R
zYxUawC-X$#B&O}F-n>teW0Ci|KibcaT3-3`YWmICODvW;UyQxqWE<=~T=RpQtF6e-
z#y#PtGN;_5VgY{6Mi<|-H~TY17yfd~30ZS+6UVnH4a~Mzrk|F3;+lVKR_1S&?#~hh
zys!S)v>p}rpL$dLKy}B54GIdA&j~MyzQ=H;x9j>&zX}b<={j7;Wtt2(GXB+Ha{j;(
z%Za9EJw6n_s=YjU+Lz_;=50AX*XaD8f|9VjQg4+P4%;5MX0EkP{cV+>|CVEYZxYK9
zt5d9dSPwH5r~IC?^z`q)rDs-Ve3?`;;oPeR^FSu`7jsVZ^?vh{7kU*sEAX1cw$B;O
zpPF`aAD?bERpnk$RK+r5*}Yq5N4<X}=ol%zDXaZgg@jZ(-^ceRE5n~#$31`lepApM
zgBw>xA3NvVfBA9M|CTF@FULj~{6Dd1hB{x${0YB8r%uW?&^zSrd9z`~6A@vj^AqQv
z-ENz~cX@}${Qd@wn2qU6KHX|-WG>lME24Zy;%1(`exBbZuI&*<2ZSu2ZIjy+$a?zr
zt-^-LgkvuRmb!kq^g{2%#g7?Qj$c4aja%3&yHc}19p#>%D%+p6)vAt9%;13WT?VaV
zYucg&pU-W#G&>L#JHP2=Pw8ge{c#Tu%@e!+bx%ds9Qm>}Wz%MM*8OC-JoU()Q)2O7
zW6MjL{T9dgo_4q0<Hz&OhAZn>%c^7B{C1_<miu(yQ9l3FaPOMKiVr4ToVzk$M*F<&
zi<R~kn?1R*bc(rxS%`U1)ODc~I%glOWfOJpsx9Mua_I4l_kmqgu58sx4LVf&DK#p;
zKmYB9bFClra`#mwWvDwy@0QLjoo{Zm;ry$V*)Mr{wWd!K+kBMe`Gli4S~t91xH)Ot
z&E{S2wlwZ`b(A-r_A7sm%s$5bi;r8Se@rv~^k&zhg$&K@#?f3hyWW=9o_h4I=cs0L
zO`l_Pn)<Kjo(|IQy{~aj47qo7;^SNU@;^GN&3^sh;w&+%iy;B5&+mM25@y)3vmj;i
zrZzF@^|eJS9^Yk+?d6@l=$r5@?^{0i9)JJcefdy$^2G-;(qo=khW0hcJmQv@F;;uE
z(f0VCGM*)Nc0cz2ZL+!bgyp)+<^vAf4#esFh+&c|+2=oz{f~=B|G$O>3ynI~S?8uL
zxbj%{{GNK3jc;#GR&Hs0cs?{J@Ul+aFB|VQcd8C4N{6mydzSgX@#UNKLN1=1Ym%8B
ztev&x5%*<g6Z7QND+RXxNHFSOu+)m1sYU6=igQc>vVv1ea!;&a?%;EXU2m>nzw^S*
zQ*tf*B}@+L<=n^T8?hA>&2cEXvf@Pb?o*Fa3+69PcKUVY@K*cM)8GEf^zGf6ykq{#
z)Z^=e?rq#Yb#uw@n%dgG56Z*rP6)_5r}8pYd`vyEug@Uf*>-QZz>$};Z#pyOO#U%@
zdg|E;(e?Z7Vl0<v8|>X)7QCG^&e-+Go~3@)6Q94nz1_6G^V7*UFU=kZ-~2c4O7DEP
z$?f+Zf7^8BiGAy`Wr>2Hk94x=-(=b;WUq8DKE`f)cKg2mYhoJpcdB=bo)bS{&U|Wu
z@&v}zDrMzIe~*Z!TjraKGs#t?+A|$Vm>V&fF`ZGb%_sBzuNsD!xjb)*7o{2;Q)?{o
z(s4TQW&4+>pD*0cz53<*C8o7A^dlVJ3dud-e=ZuH*?4zV$<23N5uX-rzVK7|?){&l
z!VfZ)7r6@fNxyKf-FwwnwlR+Fsr&wQSEURlXJ#2%D73QFwlV2E4n542UnE_yQja}9
z+g1O0>Z=m>=MOG~eMt>BIsWJI@}#4W?*!Tv)N8-bOe>HJGuxt-YVf@{V*lb#cbBd`
zxkB=kjkELKqK77zI%`Z<FN$P*e=*(o$D7>yp5<1tJN+-n-1)Tn|K5c^auPU>`SXX&
z{q}U>s!r}Uws$|Svo>G9Ts*t#PUhjmtGSjrPFEBvY5RVr$L4{T^_$N*=ht)pIDKpJ
zbA!9Lm*oFAKk3KRZ<S9<^N%MjFN`bCNr;H-*FG+wKRLIF@qEqG-WgBZ+N#B5wQ^<V
zL|zm*?Am1Q$M3JbGWY9`>;6ZLE94EEv)+mu%<y}el@n21@~7W^ac<Szb3V6DzF-Kx
z`9R_PrT?=fx8^Q5;_)gY^Z(pRj>AqF@)r`Unq_Vk&-gyy)#U6f)~a;IaLY&4;ioq)
zeP3C(epB?<0zE~q@2y)*`u7%ISld)>-2Xq@QufNfu7<7phZ5BdZ@IH5z0k0|_W8k)
zV-egR^e-J@<^G~S<J}s@KgRiC4FwF3tbF*sUYs#M>OgUyG{Z&Rw{msc^e5Kn?%*sF
zPdm`vP`WC6ww}SyBPXV_Yfp|Wp0R%8pED}|Kdp-jym%qWCgjc=bH2ui6o-ZDK9w)u
zEjl@yrGnEyvSq2Zol7M9_BWN%Jh>+y#>Fmj|FZ4Iy!!$Y&vG-iosYhGA+5!ozx>Gd
z-p)_!UUJLX+)`b;<)(N{|JEWV-P><oocQ)uD}zHq{BdmV`_d;G_2p$#Qe(FEpLu_3
zBBPD5mvpVWTh-MY`P|#|-+TRi;`skrz*m-Ot_mx7&*y5)%wFGqo&QYl-Sho__Z~mr
zc~|M+mp`w>v@7S;F8h`p{`T1PJ+()9^?G&HLrcBRxya9akkhpO+?n)*_{3)WW4^1k
ze@1?D2wa?Bmil|!GsQmU#VcQ1Og&dB7Q12H+4tM7u6&eV9JEpR1<T2oi+9#(iR5jY
zn;Cj)r?&P_(5~afLVu1P{%@^ts+zHC68Af;f^EU)_LkSKZ}O9B+_U1S&vd69KRrFW
zK8pI^H;cOS=S!a)yX&M~TaxcbC)5-z`0#a;U+Xpg32T=eUDc52`#}3m{>#(P7w*4Z
zRrK<Co!OQfX47RE)?R=4{*cKokBS$UJ~*3l=QfL%Rx`i6aL9J)-YMt#BA*}S*->$E
z#)6vit6R53F)y>$x*mP^3g?Ev%997YTTkET(MZpE@6C2*r|j0XXPa*L=y840%H{ZK
z8q%P2Y2oYXam%;ISS#k3hi==#Id|<{z5{jQtKUxPE!M5LbKI!#+JxkC0h4VJvZgN<
zKPftHb@jqm*H<?;R=2fQhAPZE=X2)dBG2=melA?KvO3;MWcR~9!S}^=@3W*G+Lzv*
z5Nq9KTEzJ1()P>e!>;$0i}w^OZhLl>cjt$*%59n(>webHk-nSR7jySU+K$Y~Ye|>5
z7oMKx{_JSY&8xdZ^N%NW-~4G2zJ0Ix^XI)BUjP3%Tn*ugxSM=me9r&t^4zZlQ*Nuw
zEm${KV%x!q|K1&ItDU*}c>BK}+mC*{{pNDL-}Mb2CtUh{tjFTCEzjHMDN1YC&tmoC
z;(p`zUUtJ-g+<xrQ|F4*rYdJkMa6R2RbMK3VPSs7%+luES0SNxjl3swHb<-uza(=m
zaHhC?sa>|<^8empwaX=|>YE+pwr$c%__f7x-`qWy>~BTO)gD*3oUYbdA*PUVUVXdH
z`I%eIHdpcNRrAm_G`gexe5$w5zEU=ZvZb>b-oBMjKYMFJta*@M2b1krQ*%Gg=yd)C
zMiu9MPj7yfr@un}(uJfj2d^T|dp4FefA?-)Jk|c^rnQER%O*|N_xkZ@)puh(t#!Gf
z7uH!XcU@ESWzLPe|L;aj`SNPprjv6j9!Cc}yE>08_VUe>jJyABPWe#v@^1CT+3x4=
z1qRH&A{XJg#wN(t?&Hc><;>i?#=nn#CB}Q6-oIt3%0v0rT<J;|LNrPbZC+S<N#VZo
ze=GgEUfYiIMhq9fhVPkd$1XEV*)!3}+OO}c?Vo#7ox`lS?_8cLyYjN=?*hieH=C>T
z8`t*!x3RCYWMUWHux4|~z3S5FlMj|(7dfZ7Zu6~)(HX%?dyRk0RhH;Kd-(X&Un~7j
z9lkH-^R)cj(b?7MuJQ$J%TE9K=FTR$=<bJ24;g1``_HMBiz`nzk7RRP_h^1Z@s~B1
zc3l%#=K538d;L6tZ9n%}UYxt9$?j&nWUFEI%E0y6cWb||nEvF@p*Q0EaRm!@Bo@5d
zGvmDT9$#PowYFBq%NJSB%I*GOI_<8#dD_*h!cN7hwVzIO{*%=%diLhPXZ87v`HR;b
zOu3-xz)&9&7e8yun=>uR;ydo$iaeAa=I$@G$D`!dHJz{XPr2VHJ27M0$8gS#pT7!U
zJTFi`S%Nb^@08e@d)A+f>YwdZel%P6sFKl*U>?~g>dSh$#6B-~(@K4Du<mmu=cAY9
z{TkP8f3CmGrncO_@%OCM-G4XUvM3MQvE%O_m(<sH6e`5e&lFJGcl()uap8WYwJ!4`
z6RNC_a*M9ZUoP{?_^EA(i{54H|NFdIen=g)3t;`a{mb>w7w*56d-2QH%6YNp4Cx0(
zt15o*^?ZvGYE*Q8w7#Ba*_A(v?<)8=y!fzv+WdPbpUts4db_fSzqxw*<q(Zt37I!H
zUoyT}o2TU(5X=(GI)C@{gQ?p$OVm%EG`)E8ol|M~?#1O7j%gI<RP>eqR+D>@yXM&*
zPDlG<`+aFIytXaZR$R;eZ0FYn_dn~dF81-i99rrA-C^nEmVX6y`8&&MDwdZ^JnR2r
zuC?v?FM|Vl?X_h~uFo|o+W+`7-(#J>nWYC~*q*T1ro6ddrz3UHVdlDR3m2_C%VH2G
zd75XH=Fh`#ZwFnyB6gs<<E2P4|FZahdA}@q9-K51@8`Q$aCyoHo!_5cy}s9X^wBJ_
zRGFaVe=WkVU!MK>xw)8sTBYYo<?@~nwcPKGq=TQkzAsy{fA_|RD|)82$+Iv17ju>m
zwv7A3=Zy}>Renu*X`GS7vN!w7o%sy@EEV+;FVlYQX`laYeb2gsN2Ya6U&2y8t0T$i
zx#xuSJJzXYyz_3?Kizn}HHV?`Ol(I{-NV^0>$LNwzA?UZdSYAo#_M}({&L-0;a6s;
zeExL8!F7F(rfo3qvdPia2Uol69e)zScr5kbreE70EWakE$MwzB>|WE!MLF4(vDpn$
z9_e)j_Eol<zrFaNeqCgq`svfkFLPhG27PvHyYBv0pLvD01Jlbh?$TRKBKJmb*cxzi
z(Wx)L_nnmeeregI%|DV3Wpiz>YBDc;vh@8jcFRSfJD<GmQJs-;d)whleXonN+~dEb
z&H6Zdc3W-Y)k{DBuY2=iRjm}`qw8DqBlT9lp7F=ZTfN{<ef*QyHSeV3-6U@Q>&;D-
z>^~sgzP4m*-A!emf3@MuLVF*7^__gkY{#Ri`lnARAFRG{_DFWU|6blJ)4xC6sJ;87
zjr8X1_VqR~whRBJ`grOc=@zhw7khMcJul-<zSDcQNblFLdo^3PEg^q(cu8sSZ2r9y
zHykZz*>+v+bIi1dPDkGFJF!#a()lmvZaOpPXsGPGW4TVXLH~1cBddI+rdPIgKL2#{
zYbjTRe8dlLSa9?x>&$z%AHNDol(}v7@87*wg(3eO{)V5N{rb_7MGu_I+k2(+OP{*A
zt&wc{bt<rP-Q^7@)2m*}AAi>RH}m_db;aNQ-p$lh+&2Ho{9T;e4&|o(G5B#tF=zYk
z&xcf6yWg$peE8^YXr<EjA2$?z5++2aOEk`jj;hv<ldDQ>`feHiE>`qe?8bL14)CcI
zUS7nq^u`XwM!$n4*BsB+R2f?cT`DkHU>~u+G;&3M)!dg`!mq?ifAG;heZTkYI&-H?
z^KCqnK5x)2T)tM$ZSMzzm%E;x|Ni};gu~qrhhDGBDzrMLth#;2y@wS+;?I;kW0PMS
z{+*b7PBzZ{R@*^WSzWb_=MGouJ+67BW@TG`HP27<LED?<*5{|0Cn*^0*S_s~SCTzP
z{r2JaJNcqRZ|{;$i7j0-|6Sr7vn>%{I^HKgs}cEF|7)ev_A9TR#FTLSYl`sxv>;eR
z%(&&?oR`zAcAvR%QB0G$RL?2EW%uplF-jFx2aZJC_j~a9)tOfoGX?oHx3M3Ly{lTP
z`|pnB@_UY+?;rORy|ytray;X*?(5<*8JRbU*^fW%75(`AOpo1?bKdMMZ>pq^m{?ff
z_~*9z^NRD%dwWgi9Bz1~8Z-Ca)k=<2S4_=+%W39Lxt|z*Gxz(c*Jrm?PSq7uUwvtf
zi2ec=og?o~Ca*|-)NCbYw9h(7xb$>p++Hc^SG(Sxe`#W9eDVL^%=2&e&vf+vE^2x~
zk3A;Wc}D+2$=#`!ADCR}6!W;oZ0-KBR)wL@U9fK1DM7(F_oudU^<Oi;P4V4X+LiwM
zN5*^Q^ZK3TFE;9)-eE1)KgH~)yzx2XtSKLz=DsS_d;Q|=#ANyN;Wl6QR0JMk|8Y6<
z<jR~+Uq72{J8(_)+ZMZ5zjjGWzuI$a=PjmP4-Gco`x-fE*P60*;HArqXPMMzv-&w2
z-2Ed}arOJdvk$6k;*N+PD%f>x`%}ArWnphN=cTGXzr(ti$=df!O6Zpf585m9=E!`T
zJimS4&$Dx%PUc+q{1Ee_oAbHcZ4K@7rk~O8KUZ@2vULp8%jvzmao?B!+<4pQ^z7Py
z8~0`0`S$16!A&;(D_3)=dn)|hVsZcP^nTu(?H;e^<-gIG>of1zmP^MHQcfquafz}0
z`SH>8<?HE7SaMmz6FrJngu6z~oaDd6r7AVX)GkMz`Q-=J7ay;(reExSa(vA#kqWN$
zi<h61Pjg*=J2ayF!PeZ=%Qy22yq_r)2=3!&`}8r8>CuAsaqDw+XLo)#pUeO1*U`+x
z?Dn>*@Ve=7wm&Za6`joZW0LTH3B?J1*7MAbGT65r4YS|BvO=t}nr$zm>!~RB+lF&*
zPCRvY>v==oqDddl&*z$|JN1j)|M!y)&hP90xGrkjZlibh(%s9M4{U4P{_fiU3#a~E
z<}jPg|7p9ia`l#pANIyRl~F#W{r5-4{TlUHf!LeA(<XMSU4H!cTcY;&#`#9C*=jfK
ze`Hd({{OF)1?Sc;YnL}u(wMG&$W`PSGwY>^pWWEx{}^trGT7VxBJ<1qUf0Kr%hpe+
zl8c*ZKTm(*i<YO~G6nBHRMq+Wbgk;)dp85k=H4zgG~>#jC$#P3oWw@gO+7YZVcRQY
z^m0SuuB5Ae(>;3nO_S%YPbaoXM25%9)$-~pDlSO=dSUCu#Q~8A7XOPAyeCyMeTtpI
zE`>KA_s%U7{F1s`_T{XLejl6G{z*FR_*DI1;I`yTcG2+~ZhYUwd8Rl?O+D>!)oGo%
z=H~~T`!b@YOlofo@z_2o{+7bMpL*xnlKds4H1e9aR~YPhymZfp-5<GL<gwlQ{#*Q8
z=c`>|N0sYd&WSLT4Ls+}EHH~Xh`XWY_d~Cb^ZSismd#L<-@|hGWvb58qt(Z>uPB{g
ze?nlhdP{fY<3m%Tr5{TERnW-MV9Cq3wRfD*rlNI(`#|<=h5GxS<hfe@DjoT=Qt;8O
z;zvgX88)0>e<DOoc3BhqoChj9-`CxVRg3$pwOe?>OD}1T4fjjlB^dJMO31ui+%Wyh
zaoq_|ZYV$Bv$uu4vcr0sN~g$$k1W5dIoPh=oEoPZ_fLuY?dzWFXQafn^emL4c=|uH
zJFF9TKP|<YSs(Yq<ns(yhPwUX=cHAV4)2-zem-NIq{Q>5=QrmxnCxGDd%8-O=i#O)
ztfF}V)*Nyr+pY@AKm1@}FDUk6Wr<_a_tXD8e%tKU@m=}(k7S_Ng_yqRTc71mUlN`0
z<5lsaBS#m=e_{;fVgB}~VeQ=W^^-1g7EIqYuj>1TB=>L4UxiZcO%BwJ`>pj{;=`Yx
z_xpmGT6Nv$FdXB0v9cs}yP+qeo-O}6o9I{Hj<!@K@~CuO>|{J7{&+^R0{{6%JR6Rr
zzbrrhcB^P%smz)AO@)04`(s?9_nF1tsXl5T$kOL-CtZJF^XZ0AX3wUcV@3)uN~h}j
z#w)8_4Ly{^cfi?1L7{Z^+#eG-Y#SbJ+y1cq)3OWZ<=hRoPgz++Zo6)!|IOspEIoPg
zS&tRGmoKrenOrW{aJF3Kg2%1>p_x~?R!BZcT`^DKbuRO=yH{(MtkL|}#=rU85xuSQ
zsY~l4qh~0UZ}ydaGF$!fZtHViAM~H)nen6^X5X!S?7pS$r;b_cKIUh4yIpuav+9`|
zYssI)qD2)IJJJ<*uJO!gdfvL?vpcJ()yXVxKaP(Y&#IaCE`PaYqf2Ppm(u-zjnc14
zPgh&Fq?xbbyIN#Hv}5fJ9cw8st1uaz%m+0Ab0${k9Ob^e{OHdqMe~kY-kojjrGCdm
z)g*YYLF3LF5`Si1wQak*`po;j3o}(t+`h7CS((ZX=@*$Rbo1U8v=l7o&CZkEcqmdj
z=-<BQQE!@f=Np~Sf7`se?ZMv-&O1JTn>+Vux)H0jK-ctG+b7<o_SQvZb9P#b)&CED
z_h@0_yt&7Ip4#$CBx{TObJ?w5-tj!X=)o?QRhd})v1NC<OYU52l??`W&R^iM-fngI
z-KiS|VPW3s9~`?cuDNh6>l)~E9S6`(>)^XgviaYad->}w63E{esdQk5X4YAUtdicp
zSGV?7mY%#~%NZTHdsf9o;Uv(yrjyfKx!u^GpR0R#;p?5e?{6Q@n&^K<{CLC5b0y3c
zVFBWKH>>MB52;;~m&smL_c!Qi>cMY^9JFSv=MVLJ@YitSy_fUjr~jFnqs)Ba&#%pM
zE;FdfeQxvHyFPx}kApIy7rfY>Fvhz5yHu!X_F-zRp=7Cc{%?!lTt3Tqql;cUGc0?q
zFDDvugrDiNPso8U2J4l!ZnwGp`Nwxw;U_Hbe{!?cm&koGp8HGRkJBr9+wvlf^>14b
zHr(5-^z(0KbFFrzzS@Pwe1CWJ2><Q>9>&Ku<KD-zdvP~r%2sE||2DJSb#8T5*uKTB
zhnc>6SQ}kfZgpe%?yv84b3R1Rjk9h0de_D@Y42)<t+l`J-3_(dXM1*2Nsh;^zqOTb
z&*XjDXA{@5c7uNGlnwK)oBME`&-;C0O|pU})7$rNS9KJ-TvR%JOa8nk-;ICoraFfi
z&y9N<{C}^#%HBPh7v43sv6SAM>;5{ave&gs`IG89eeTC)Dbq~b*PL4+7aw(Bs^Iw-
zkM`Q?<mybJ<Jw)(4j&&G^zT|1{4<8rX`%VOpyK;2>VN0oN&Nih{dv2~-laASxm=84
z;d_^*&gM`N_F1lS?8k|e!o!Cu7}geU%bq7t(BJ(*5Ogq1!i1CS&BeK!YMvN`bAR^W
z{Is>{jsBMAyGC=}t9GAfdEt9&ru4r(cUApwoZ2O}mFsq;O+zNrUH<s>YZPx@iezv7
zVphM;)Rm{}!?wGhT?%>(JsY_kJQcoo_ntbxtZ2SM5zDC$&J2?sQ<h9fYGVv~$@zDE
zzfsDj7wdQv8|5B-obBGJ$oIx|b}yTNH^b&Fr?%Oxc=GqNevn8*aM+R7-cE+?|L0k_
z#r%I#7G1aPr5nT7tj=TK5A!$g`uEPhI(Ijt*gP@Ky6Tyqo?n}0d*Jwzh8wTHPt$y0
z_fM{K|2@kn#(k-3kEX>m?38=_SWdJ;`~CR?Nu46US-CRhuW2)uKQleJH(#&Jtzcne
z-1{QMmhU?%Hn4Ovzu5F+xpH*8Inxzg+55ehm^Qe;&H|X;?&u>a*mIP{zVQ40?<Jpl
z{CmEI9`h*a5Ph?}u6(KW?wy}(3Kkxin(n==w_7uY<H4gQkv>MfV;&3H?G;lOmAva|
z4(Rc;JhxNe0ILMoeBL^JW?zBDhjK2xoOSW$+!t>-ESf&MYe^_JIJMj>t$SPda(X2D
z^jz-8e?l6*sfK4Jdq?kFvGQHTg6e>e$7EJ-|GXn(K2Q3W>a49xba%d)`{lHO{7oC9
zL%SdCpWU^*;JeCa9SN}w(YDtM!_7+zzCU?%z3)d--A?`~MF;isHD?|!Y5Ke0PWjN&
z)|4QAKmQN6zblLVwA7FMa_iwybAQhJV&ZQC|G(UKs^$9j>F1Wjo%{YjKWxh*heO%v
zMfcy#c+0cHlfCn{{mETF3NoLRi}&%J<77RsM{&8@lf8GJ9)6f&@MQ1l$<z9~KZb^N
z)>-NnvA<q^Zg*Hpe8r9(iC=d_yx+7{($p|NPk;V7*(n7Dk=_Qf^UiIXdfI&ZRKLzo
zrq|L#OxVtOaaf%C;aBj=bIE3=(pXEgGY^@#Uc2Tj+A^1w*ZI4sxtd;`a_+<4$%o1g
z7RUVGpQ|cdCSt;4t$Z+_KV%2*vim!g%CG%<)K+n$;4W|Qp-X>ldDfOpnQ)~;`S4!v
zo7?{7Fn@g>Zdz^k?eE3bn*Il0r&!GINxvqh*=@OJ8WWe}XZhzY?<ZS3-IeXWb2LZx
z`N8N15g}{^)m?XDGwsi%&fRM}$*<63MS5IRWbmD1b9Mzzf4=1Quh#ibwiVbX{{*ds
zsh?$^F}o`8@YIbLZ*QA>>gmH%i5*{8@!LH+`Y+_&mTTr`!RNvL@zK5Taa-{!35WEP
z2fMvy{m|JOcc#)pK0n*D!t+IO{zZACET<<fFYX9B##Sw~DLl78Uv+|?@nqxK3^q2q
zjvE}ZV$M3nDfPa5sr3c!FQx)j*L3a&%M}+|*PCCy^2xc%*CI1wM%mpb7i5Jzo}0dA
z64*cS`M08zGamE`F--e@qpb9I>aU%LUwTgNwOn!GX^3g0=Y*$c_4;=AU*muDwk6>_
z6X%lCZ@AmoML#RAzj}PRIM1&Trk6hZi((4&_Sj0j?Cs&KI6k|odf~b^H-dk!xc&Ie
zozougKbvhd-LEGXUB7)*JX@+{`Rn?M)O#im%#LN$C!f6g;m5S9JC>6sPns8?(>#}p
z9kjQ|QtSTIyH|eheEpzR^&{5=z02?C3GVDpuv&jT?HW&!sL%3s=KB6Ix?ewZ9G$0J
zlesoG@Aa3vV%H*=m-F#|er4jf?C9?&f<DicL@fe&CmoMp`+4^lDb|UPx23V{*}6Ap
z)#pEJ&gmZ7elua`;iB+Q^GpAGZLZw(bvo~|9uA=;vlrZ`X4)N;UgbDfLnlAPW-I3|
zuZa$_izlo<wu@n9qQ`0{f$2q-4be?6zXbi`oML!Jzws6C2e#$n)!KGPr}}a|&|?mg
zJF`+Snm>$7x#8B)%Zo+wi>_H`*Ub1)SN;7pKa1zmDIr$JSDb2yzvy^{y}|tHGuP(a
zx7ICY7lf_U&Unw^K3~mrWV?=^!2S0cp^Mge|NZ2lF!%4x9s6S#%Q@A8&&?NH8*u)K
z#e!oGYk!;lVUqhJ<guQq{d?LQC4~e_(ff*xLANHv-8^4+W20Ir_l`F&HGMQ2)Q$F@
z5nfayU}EOLX!A1B|AVdFrRYe8#C_MyHb;C6Ubvzq+`RH5=RaYOWygO-YZuRS{pj(0
zqTGr1Yfp)re(;)at?AU`bpOlSzI~fN&b^R5cat=?>FNXi=c_V|*F;-PdHGgE|LKPR
zjXB5T&I)-fuk~Oxw!3MyY}xkr8x}kg>XGjA)+-CR(I39=j9uDB52lx24yfk}+P?4%
zK2%Z_WB>7);?kP<zwZ1MUB;kdbH}folV<lf6l~!2xZh$pqjF1JWQ_ucpQ60Yd|9TK
z@A{Y{+51^8{XCQ6V1IlDzt}6bSu-4sJNExy`{=iOyioeme=1SB>70r4x~J|dZgFMr
z((9f)@#cBkjS3&slB^E}*U8@cXd6}^)*0IWM*MS~yvClUpWm-Nyt80;*xK`}IH&Da
ze9q5X!g0aslH>92|9M)L(Y5`C!FBf9r`p`VhM1Q&@fozrZ~d38*0OH$#^@_AW9QnN
zn5PLdMXOCX))vxRUOR2^8TAy&V~c-2zB3~~HqwRRY>cht>W2>s)Shox5_H6U)h5~1
z($*SL&v<z?=8C=FJ^z5lbIT>!TT+j3zUKdItW}@k=305ZXu90q8$TVRr*C`ycUHnL
z^Y&G3pQNm1y{~?L_04A6;s0|bw|+d98u_tie!|{AH}W2_Owru>Dm(nV;E}67)=EEK
z?EJA|#wWdl!gb%v=g1|_e*P_7I^z1VIky6C&3`^c^!Mszk>~sV)*rmEz~<H}_2h}W
zv(`S|Q!+2gxO~GsW8+hAk8*77TX!yhhRCk3|MI!5s-+e%lt)|$mVc}-BVzI}Yn4y<
zakucV|L(kBS|Qr|^V9P@mbL#OgOEQYBNmh=>}v0KHPO~t_lz-FyZfEd0^Z9<?zERZ
z@Xc1M;T6#jTUu+yee0zE`=`x!pPq7=;%Kq^_NlJ7Q;#puo6P?6iLlS{H-YZ6#7s9_
zpPH9*QQPPRm&fL;KTnriiaQ@~Yk%(geoMV3gGc^j_5BvAD(A1t-M2R0blOw4@&92}
z-+d3}e92~<zT?X6JF6;wpE|AApUNBd?#so%IKA!NTyNi=U~O{HcaBZYT`jKmR90`!
z+pXoP$LHNze!6e#{g(33y1G*DCyW)J=6w8f(EI&-;my<T{d%?O-Ieom8M8NMxo`fH
zyWDk6xPH%y(z4QXxwHH0=C9wghKWDBwy5rnRr=4}n`ZCwNm`NKSsVSr^@#538c!zw
zUF)st!=GPC-Ty*B_Qk?IOA6O){%f?Yq5mYqh3%nlbYGp3j=G&w`|^v?D=AsVmn=8F
zi(FIRB>Q1m(NEX?tIOsKeHZ9hr*z`I+8;4n{*-q8Eq~P}J!(~m@IO#uwnoCi(lN41
zP1r&C>WMTKBUOi=d=*zz^G#pI&ingRVj`1&Zqu4CF=`U?f89FhH}6gV-og#S+pQQ&
znLe#%3=__nzUi~a)CJc~RSkDt{?&Kp(n}Bj57LY}UMIEIOkX~?kLC3H$~t`w`v;3S
z$_<;orf<8vPvFe*_A=ub4$URUcjWNxJ=5E?=i~f*3$dESd+u?zD;n<W6#u_^SVT*5
z-sQ!=wyrO(xaIlgwg7AX3*+Nad$(~a{TGY7%(_88RpdqVynVW^jK_C+H;AaRc(fl(
zuD?{EePK>)h<@GY^|of`KNT-`?RqVKguCh5l%+44Q+6vf?8)oDdhW}%`tJ)*cID~C
z3HU5$x7eS5N?f`thryb$f^C!U;c_p}j#oc+R(_buY{oT<KkT{oZmualtNtf#v@n!V
z&`D)>xW6vt_Cl|Z2ef~STFtDUrgppO;U=D+e=OT9gsWccdblBa2Gf^I#r2o#a{4w;
zUfZtaqP0){)6(LNkLK@}TdiX-cb<^Wj^bu+r&`5zH+l_OcU8;9)NV_uVS3lIV9C?#
zhtk51#|K^A^Z$cL>UNK+y17oSM{M5pFhAQq-F|Or#G0Ftt8C2g7Cyc`-B#{;=vj_+
zc?a`fJqnQz44$LRb5NdhbLQ3zqy0KQ(>BhS-?msbd1A7J90%Wn33XZFmd_5CuIOiY
znm1>9N1BW+&&@*@f)sCgt)5~Q*7=je{iOWtRNKts*1OkyPC93}ZnnszmtB4Jx_$Nk
zXO_06PpX>m^P~BRIXwBd%GZ1Rva!2ydfVnL>bpE0+jo{4%5a{Wa7Ay4;hwsn8~**b
z0w*UtUtIG+=bl@$&HF2?R#R)rKX2H+?!Q4X<FoRQb7mVbi@m<}sL7^lUYx-8Ne$)8
zonmAD{BAlp+r6cFs{QZai#bedU#;OtimLQ0dT{TuaE<j}i`uf>PvPde=U4rd`k3{E
z&wpC;*`kL#KG#l(cbA!=E`DU)o?_Q?taJJe`EO<VaeZ5rbVt%se3spf{ik;d%#FUc
z)8^&f__|qpUu5`dm&a;xJIqOMoR#+4UhH`Ovb^%A<>_v@+AB_lmdS6M9{OdOOGWCx
z(!f8f6yEbmP1={y?72RFvH7<4i&9JCwu#SqQ|hC>L+7%LY^JrnzfO;1jBIzD+T5V@
z1=|-so4eF`+TCA)*D8$;Xg-g+VLRdfxl2pVziGaiB)8y7Gq3sFV+%KzY&*%F{o)P#
zwToXDoDVrweDBY5)7Dda<j<+!)Q_L?!L&?YLO|m>%kjxG`oD+zJb5H8Q}32@a&5Fb
zyL9<6wwxr#xof<izf8<EzP*TZo4lkn+tfSOifVse${v5iaQzMBY=6eHOw0RNr*tOt
z?>_wfhe54w9LtBcJL@HStbJS5cX2dr=apu$W9XLpnsI>Jdw#{@b?3LS8OjJho&2&(
zysMV|P_1pw2`;5w)vFGESyy5#|3b5_rZoDcjNcN$!rE(_Y@Y_}OABv$`O>v{o2P@(
z_DfvLgx`ec-`||y{&%b6`tl#Hb)_qenC>ZGE|bswQU7fn^M<vrvf_2$zqKfP!;{XR
zuJ05P*}|K+-pP9*H&^b@%uLNKnv17Q)8k5Yo~f!5z|q1pX@aAZ+Re*-Z!3;|yf;nP
z``7;Wp5b+2D!)Ca$*`9=R9>q7KCk+HrT?eB_idkBPf|O)ch$1<K9@pn=Y+<n1~2lt
zyHA`?zhKS2H_NLkO4Hoy-!4v_P`}F}`LyeLbG~&sThu0P{#@<-X<svI@W1(bul}>z
zHmx{3{hNGjuo1Ix%Hal!%O~fb^ErF*68m?fa-Ea2PyhR1BR1)E>WvN6mUZvW<jp&^
zdww&!pRsuU*R^l&*}Y5MQqvdy@$})!(8!F9_b%8yRDZeAZ{PXr<<t0oy<76+@y@Ah
z13zq-dF-`hVWOP)=KptNO#}4oCfXm+w+{Mm9{6GD{D7Yyj;ucNc(*;*wI3eYPICEN
zJNI*5`lY-if8m7Pv#d1d=f2$#zI4uG`!DCzX80+JdS!-QKlh#2Bh6#Nx!AHM{kaaN
zwy#zDcU8~)_@Qn2@f@~2tIx~MRtcJ`U|>7l{(j)IxEM+1+mk$PIXNyxNA|szZ8$09
zk)GPLndP$Pmsej^&eu<Q&UC5rue~;ROhVPQg!bvXcFDbcagV(!_OxBihJ80!mu$Xw
z@K05RSU5*e#|5!3`|fw0<=Oh~=<5fzd=KJJr8AuTIq47wE0^Ux)d}a$-{G*V+2i;y
z!~f9Vu1=GVFzyQ;3X@pwx2L)=PdNI1Z?X3u0VUog7o_94_I|9~uMzrHtY7AP*7;9o
z{S4x-h0Aw8IO)G3MvW`BFTHo(bCJArHecTbhS!JX=msAsiSuT9+P!;1{=bXneHso{
z>o@+J-eL7XjeVt;g<a7KcBc$w-LluKbmR9$?=fGZvFOdAXOl(pwuLBuYKZ$+>%gph
zbo%p#``%mE9H>>8;P<lE!Rq(btj%r$-r}LYoJql!R=2;iY}z`dqj<iKnA@9K3ChAB
z#4E2nHcan5p3a;n<zRaz{l@`jiQ<a&pKYI8T&j0GS3aduFY3_QYaeYFR?9H{(KMHz
zS}1oj-c|M7!?qirKOViuZ>oAoYG3G@-@N-&xEF4}qj9eDrLxaM=D>6N<<q_-O)ie(
zX>Y12XXn3lcyfdO-gmF7H~-uHE+W>bVQ+_}bMomY$9C@8W4|l%|I5>CHGjK{gZXP_
z-oK!3d+!R*sY84Zw&XC2Xsa*d@2C^Ich}7S?dc*GyX*fmzWm=HxvbgPyQukS+8pL~
z$^6f6sy1KUdhSqM<neY1FU#eRyGxX3UvR(F$?#*%^7NfK!L>`655K<7|5K>=rM~0O
zNil4*eSP{@1+G_2scgF7F4s|A8SR*?w1QQ6S(-~|ar2|6Z=cRID*C!><6QS^LDzLI
z*uDFIbcgv(cNe{7D#s<F&dG~}uYA^%ZuuhQSk{$tmNm9>pV!{_cG<AI=+xQ3-CnbW
z7*5?v{lDUL?3Ql__s;wC=V`zD-uCL$ghfG>eeO&rQu}>=wf~9<f4|kP`+ASBm0@+p
zo6S}=4|iB@c%2^k{>1Y?O$#?ZebvifI$L|c{ins7+OB=*`C)K9O=x|!_4({GlBG`?
zH=f`2Fx=H{|N3_;B#zlc%utK`U;OvMc?sTwn?&+{F`QZd`|77f**`1O3!RN}Kfasi
zrOMU#d{>p(j(ax04*YojAmVA(<gZSLWYzaxXph!!)s5r4shrX{fz?TjBl5xQY{u&<
z=hrKmZB5uJ^Ho&sRjomy{LATm4&j$xwlA=EUQnha5cZ_*kk!G)?Hg1pp4=!a-|qP-
z{P1pvNZ)viJNs4IK77dg-ZbI0a*0oOo*tjU&2?^{E`9c%$gpMSS-GFb)}4xM*FIBg
zzdQLzQ6InJ0yVj*PbKb`IGZ(HdUj6rr}>GK=9AkF{ocW7b0kbH=&NgK@anq?whkp7
zm%c4sr=PleQ?kO_$M^L0w(Z-IeZ~5_zP3%f&GRpDRxx?!_B`BwJg@fet~0JN4>EH1
zOUk)0Y<J(>YqmSK{_ySN0&4g4gk<8T#+s|G{}f>)RQhby&YO{+E}9;Vj2AW0(%Twr
zw5Z>v^d!Uj`;F_~T)VgFd-;p>-{$A44fn5)F|x5Mzj|?(<^hIYejV*y?=D_d{d!ic
zy5)<_@0T}g_kUXQXY=B|BN??550s{*y;-TZc%D{m_qm7LA5HpnzfaezGU<O`0#Egy
z3l`JcH(1+m)qgAaXX%6@mhK(Ft0ujkJw<5Ad5<!+=~JHiYu-7^S@m5=Muo%Ue3@F0
zN`nu7k-w2Vr_j$JZcVcjg?|1i<%)M(zx<eXT}9&L`6ubq8*J>oyJYsgpZ$PA_h+2$
z>$wbzZNg*n*tU9g8?_yoQKk?a_kI<>&x)6NUv8++@bfJE=6pBbU+!iQ`&*B^LNA*a
zz8=pPiFvXgDwa{sw6S2j^r>L;qnXvmwMDeDHcU$`_TTdUtlh7_7n>UFEWce{V0YSc
zX6IQxF583Usm#G&l)k8h7yn-U{(88Ffws9*#M}`5Ac>g|^sF7GK7LoZk3GbK<+A3h
zBRk%e*8V-?E1_HV*)>1>zAn>`CEi@jRmOYjtCuj`PDqNc>zCS)==AZcWkyEXLkF+F
z+vkWloNABz#s5#tZAR`k2|=-c({>a*2|ko%sc@|;>0_$*+8z7K_G%bPwrXarH0U%)
zG<@ux*e|YjS%KkEtwFu-As$W14Hj}Cu`FuECm3>g4^J$%etzffJ+=#Xl-?|uwwY6c
z<BG1;hXt=0=1H^kZR5I_X4&^^>miAYS~?H%^YfUdPhIM()A#(`qS!MVmvA3Gp>uIr
zwZL1ZitBGb9-njS&cbD;y`|52Ki9_jY`7A=`|efVfK9G*KOc{_)_MMR*6*JMlkcrN
z%5A&*l#KH0Z@s+P)860h&-u+38mILA^}G3;DZ7G=c1?enKCO8EdmZaHC%mUfWUhEx
zb0yf;|M`vM(Q2yVmAj4=7u*Qg_vL~8djr4x=aKC(Uur(?6??gg|6Ykf)WzT<i?)Bd
zckcHuon84_rQhb9Hha0?>ASV(vhSTcUG*;4bGdAeK;#vNT{`l4C9cUGj-3xrJ?mY6
zzD>F5!Ik}`n|kv<3tG<%_L7&_Z}Rr#-dX#T@26FNJ#yoIWyPvZW(il67mGPxvgK}h
zbJO<sSC_XJY`OML-Oax2`IK#)j+)U1iT+2p?>+dna>ndxH_0Y@v2^YauL{2$Xy{E`
zuJFev{dV8KTF&)zoIkDEf4saboX0s<_VDd});s;z*;lQaeJe($_V90)9X0`5Ua2Ki
zJ>T+VPkp|o{I~D={^ypdf4Fw&nBo81@q1^>nLRKJnZR1aXq0na{(Yt8y8Cr;C5(+F
zpA7Bn>>p3IUcU6n>FkyL?|1QMm$B}c{@lv?QnmV}71__&7&06dS!v&%n;m=8aKg8b
zSN#2ch~G6Td3&+w^ECEwk$IaLJr{G-*#A4)S-i30XF&Gv(6hN}O&@cZ6}M@szj`gR
zqC0x=ohuJkA57v6Zhk2*yT-G+toWvln7a96YjeK1mtk8EA7A(8+QJRf`B}DAuJqq>
z@M}!9=Bw*3ua|}Rq`4-{%4NAS@l8nK_NSkJHfL(S&Yyp2`GHtNWu-DPbKa*vC7Bl$
z-&x;oXRWi~vh=>_PYGd>zk=EQHYS$ms%XZoX!lh9E6P+g#Y*LS2=h#F-}l;ivo6O!
zyz?!5`x^IIlXvT#Zz;F5*|L}6`KC47lVvORo;>{b&+$sG6)#uLvEFdXXyvMw$xkkR
zQ+vqpYpa*(p3R&oUirnPw*Ivf8q92GYd;8muQzF3>vPrf)%RVF&o?<X^Rn>G*&cWI
zocnISx2j|NBBuJ1DcMit)mypdmb&w)oIMj!sIcDe5A$r1Np|Uf=W!MOUBi>g+;ov~
zdz^(sO!JeO@7L-I{b&8l-}Xr8)2_++b2rH{Y}cQ6cHSFR(c;ZV!xtXkCgT6yJ?``K
z6-V=;zWn-hSF7#2t>ydUynW)Imo4GAC!V+G#F722e&_1y=P{SY&px~Fc2w+s^?R3f
z@8m7szAjJ0qicC$m5kyuyD2(ZQKyf7dzCfuK8HZz#>s2y7sk!G+_Gl#2cGS-Pql94
zRn%%@X?V5d|AXn;OfGzhGp;}Ke)xf*?Vee~i!TN4lY$%P$}~&H$Zb_vyF$=PxvV2f
z&@WkEXtt7}e9VhGXLsLIFxuE{s%jLfVl$;l#(Cxi1HYHb-<k`q*vt>U{b6lvU!~!<
zz?|*=UUtj&UAz1J;FY^~_f@X$-x+_Uv&#EO@N54$s;7Cp`##K>;VAZ^QTe%3ytrv0
zukwS=`)_wn;QeLx)Lcz=t;zAXy17AhXAG~e`CPV-y-nhZ+PNj^g#k`GgyfEtKInZf
zH~aJ5lPB3H6l|Hp7PV*ns{hIHQ({hhFPUIfUl8NEU_S4we`ef=y1yJNxU0J1p!EGT
z>mNpC7B4xU?YHC(Tr)@cf~3;!4S_oMqq^p}y)9OaFIjck$kFQg#Ljbf>|SNR`d$+A
z(_CmJzr&Q5(jD(qa#PAxC9HPeUjFXd^{n&nwh6C#zn|6V`-%G7nQJ>e>&jC$hOUX+
z6MUx1LTYl2&yTmth7*r3x$ieWd9m=7cYe(Ou0K5#x%_dLoPIU4829UfA6_*dP4<-R
zQh!}6E26*PoyL@hC37=bwkPQv3@x3x*0q=U$4rI=R#(o8-{tw9Qt{%9qhadjc-t!R
zwl&P(W4R7~`S&q#o#*c^Wz)=iLQ_}$e)87V;;v}^;?r(wHEY=}?VTDvKY!Z|nG@4r
zsn|$L&3b)Xsej6LjjGV;Oy9O?TeDrVn_D{Tp2Gj#xj&wV@142UdB#TzhI_S<aoN8$
zJ=CU7m718n?S}tG<*yUowYY!ER&$e(UY6#vd)qDkxhC%aYhHd_&2en<yXR9^8{5Bo
zRr&T*UXC}{4++_fDK?Fd=S?~maq|34^IQ9mglLs6-}5v@KJFjap&c)tEzA!$+uAC>
zHlx$s>0(vUtKIeImo)vCUEz6r*`!CNrKMpVk6dpZ{o8S}J|beqNA;)srG6ftVlQ9#
zw{6Gt&(Al1VbOC-I`vrH+G1U1n7e-==b@FF(I?i|-rssEuG4$^68FIB;@x^>pW5V(
z`u{Un&8*=cF!O-z?dL9cX7Bm$Ix(WtWqAyPz=3b>J#A$Z=6KqKN))h{X4!q`6St{Z
zFx{?IVY67{^`${QXEe3GiG+t%iFy|rz17(E(DK$QV@IYpeA9p0R0w@KEMNZV%;f2R
zCSK?3tQO-5w0Igh?~CpEJLYkJp1j{Z^(@yPlMeQOd*lD49855jk9oI*S&7|JIO)^O
zKy|OVhH8O3{cP`cPD=Ru&dVp|w~5c|zZY*U6Wd+7^z`-j*XnYL1bm*WU1{Fl6A-tF
zH|S%{CdmrjBlDj-yt16!SE2Lq!0QLOc@;AE`yMYkTi|je>c05Hzt=lMl6T(Nwd0(j
zT*KYjEcL~^-+s8bvi97mLuWI+8}<L_e0j>Iq9n^Xp?<>VzWajFs-<GB{@F2EDeGpv
z-g9|h#vTcuyHdL!6mQ&Ze{TNn)k`W{)|^OL;34<Qmc>ns=hMaw%S<;N%`$%Y{9@JT
zA5Ze^vtK)!7T>i#zHj-vZ&#Cr_^Ykg>y@kRv!6WgXtcWO^Syf(?^(9$<g5)=%kzZd
z_q_ih?CYm-UH;Mim<h{N&YoRu#Qx=`h4B0Q1Hab@r(e}O^Gl^}p%$b60qLJ<8Y-m^
z<=;u%O?-ED;#^ZJ&zQ=*SwH?}x3fMMw^j7X(wn^A+nwR2-t(z*l&>FJc)WJk>7NF|
zr;LTJ-u<}9cy3Vnixp>?=H6w_m2z6@B*rm^^;US{zn51!PuI!nXWjnRQE*$);`Ewh
zmwuQ1>6PnnJ#12M8}sLNSG2CIQ+(;R-iQNh+%w*YCHCCX?D<-@@xk-LM~mx!*e+B}
zPbga&$!L(y+j;rP)sK_=Kd+Da`d9G%sSUPGb9Q^l#BDk}`Ecxm<zjakl9eYd?pd4V
zU##Zp6g2nRG_&d*_c*_tkq~fx&%8)#S9n>7Pe}92FA_V;Gap6@8!z%ZG}-+A*&<mx
zqlYVYo^@NVaq&jl+Xow6?zwKz(Ks(K>utjVml<D9J!AYe`;MccU5Z-U^oyr|dF-sb
z@a)eomr@Qb&D4LtQeR%_&y(MGYVW*j=cC?8X#B2tC45Zu?YVeM8_PvZ3A?htz4HDo
zec=4dm9q1^>yy6wOPUt=hCT8#t>vDadvDi^8=+6UkK8Z&E&nW4XL8NkdnK70XYbMz
z@A|H<J>%2%(`@giJpaG*=aY(`b>}YZQ2J>XaQv0pYR(e9Q^w!lgr`ofGc*4B`s4#f
z-%hiwWnzbRe_7D{NL4Q;zW(P53-(#UEBtB?e_;slztryf??&C;AGhx?^!85SpTu|M
z{*lFRxNKui>iSwPTy*8fzomRS9g#=kB4(*yYp`qmb6fmy^xyw^GV7*X|HYpf_3N;7
z>n8a>SHI|}Z#T7FcfU4eRkqB$FQN&P7aVw5`TFIqX~kK3KY!1=+09u!ds^1qX%pqn
z&-s%rc$58_vVi{NU8i=<IpH<&@8sF+cX%GOwak)oo%VF2)6uE>OD=PMT+d&8b^6|Y
z>ZND;4#zhOE$x5sciD2=-v?Eu)mPvDxO~RRjQ_hd-1o5imsEZA{x!w><DCih`|iBF
zsj^t*$DW+GN<UXz=VN`rWV><6m9j?{^-uq(x{|d0T*<FbcMtuu+jGi>&0q7*fBlEM
zD@AO6iqy|<KKHANEykO<LYDpcWFHlUzxxbsN*!HuU;ZFptl6`br3_oSA9e@n@GQ%k
zC=#hIcK?0l@|CL_<Nj~=2)ru89wq74o5o-z-LSJjChXr~$5IY2w_1~1ubTILNMluT
z`mk2IJ%^!T!^AJ@v8DX`wbvO{{E2<P_v&X?^U{KIi)Bn-e7lxbettI3TM?;}vWjP2
zji-07HWv80lTWiQ<B;(c)nc7ecYPVtoK1Fb>u$dI@wR8yU8Z?u{?DK8H2(VH_2vqz
z<11ZG+_Ne9_}*TB=FY7f5=H*qku>NH{8u3y!k%7ra=reG?GyPT-lQ*>-+S5e+r3L>
z*N(BRzFK1V>vLZI%i#RC$Ilk;b)R<3b>@-3<~*Y1e=DcEv-dmncTQXybSpih&-nkI
z@;Q$TKK%OhczMN|`t*Mcg}47Sdf6>GUB@eyay5SYxsp@coZNH8Uc8aZxGbP0y2;^z
z&IMnCzf}UdGuW6?R|p(`t>yCnNzQ^+hHE(|OpcwoS621VEwNCb^JI;V+WQ9v{9L>|
zYQ?+6bcCb?LPOh+GT-BFIJxU>hWyECJ(fwOb#rITwuvx#`QqVg^BH|D^Szk-{G|39
zYeovs@LS(-`rqVQW})^x?_N_g+s!NA$S~dC)9cal;?eRei+-!iKRthLox%6tAA7Fo
zzn?K_O7?}9Z@<?^SvbFH4i8ZMYBZ6@X4YrEIVSu!!){neDb_u^t)?cgk-yDl&z`MG
zw<dCZR`~F5o%_X_Y2Vd8$Zb|m*(~MRH);Q=e2K5`)}501P&B`$F#0A(PqN%UKaKkI
z7cUoFn<$Z*BQjO@QH8JftNbt#zd1MbckS7+P4(Bm&%b5=Ui`9e_dcx?r<3l!*EzNO
z-M`O4`z7WH#qIu6aCd=g_3CqFuRclpTu(agaMRs3ZQuR48xr*=U%S7bW2#d2p85IL
z+lI9cDQ9mb^VZr~McB@~etGUm!{)1Yw-?;`e)6N_@wqkM9&oJJT9e!R_wK4B@e6M?
zRF7;;7XFlC6r@`=`RkL9y>q9E&snkcNUX8mM~+G9OZ9Cds>RkXJwEHyI+@7=1s7(`
z`+B2C^W^%xlRuQU*D=L!(l6M$=}Wh9Xw2NH8=gPB`Ipb2$eBw_=<ts^FSj#>Un?tZ
zOWtHK{_)~pEo1ml?4kUW%Yl-1o*yQO|5;k0=uoYjVZFqFQL%N3&ZXVuKetVOv{-*?
z=|Y`rPbW?;;gs8KP<d+UpNGAN=LStXQPgNNEp$7(VK0-YXZoMV^KQvDwk$CEskWY<
z;rfm{CswS<{TlLgO0npwf?Ju&!bfh!o|wC0(#sDblawkC%}fbB)spye$9cKWe>4x+
zybMrscxc2mMRy;ExryEVz;!CyYu+9_$u;AgfLpodI)iTY`<7a(82vuaOf+FUSl;xq
zM{h=t%K9m`$Fue}{d;G{{IPOHG*jrSltYo4ubv0Y<319&`tFTszkWY>UEVz}#)#{Y
ze0Q<P`}&|)b5@vWcZT|}nwOh*;m2<omP2aG-PxZVaB4}O=d+-YBW}j(TZ|Rl|6W{S
zc-D1v9rLC)U0j8mc1E44+2a}UdW}=QQ_W0S$?1pREZP~m)y94834;SkYZYW&#H41k
zUhtjcIA=|lkX!Pi4O_o5HvUxzILLGGj&w-Rw~ZHUrMBMSKl9((=XSxW>P4&`>D<@*
zyfyqfF720GcK!B~KZ{pS+xlwRJo9{sO*|&r8y{a@EHbNmT?zZ+U-kJ?yRwqw{+`<s
zF?r_ay$M?$u0Q9{@M2T0)!{Wel8<PsJ7yN2d+@^R<n6a__pB*bldr85^Z#z~GUw~b
z?``WHQl70^-z_DqC2+R$#EoT~FDI79+~4_Kf5Q2;-S6_>rth8+`SrnsR8Rk>-&Wmk
zyjNyFgT2$K%EijQAn#<#`z7xe^SqE=s<OQN{Pe{!JRj_PYL-rRlbPcw@ld@raAwb*
z#T8vuQ<EP|tdG4KZ;)>#n!2#wBznQ>>IIR3D-upHF>Z4Be&9=FWJhXdw1N^t)OY`T
z3fE*$2&d2gJ11iuxA;R2<=$)O59yyQSh&5HX@VW&ew{zAIpvw$7Zf%q91vUdmt{}R
zHztq!YHNN8FIt(qcSY}7fn(3a3Li9t&M~&&Qe0c?`2A48mB0VY_F63uV|+Sc&C{0U
zEqe359Qu5$N~qK6|E7gsg1(kkEET^~8rF4M*S|aOkIikZ*=sue*XZtK(oaoqT43^(
z&znzw4_~*)&13rKJa2!R+~3EakmM%$NPYcBjswBM%S~Uze%rTibCTJfIg8)_UGl(f
zD&N=poL4gy^<?DcI}6OOd8MDbdIq<>iJ^(+o{ofHqRB#jeqQ}HUgw|P63f2xK74x2
z(sgrWaz9vKnNhXn#>A&gsq(KTG1YJs^k48?Aa33^Uv*XX{@-u*MW^0O*rH#WEPT}P
z_W@3u>XYYlHFpQZ1<Plj$(5eB(`eRgzgch9ULS9JDgEN-KCR_>M)99)Y~$Wpo0l&B
zBlF^M?&L}3cH6ccF|}Jc|GZ`8<NrHXu94s2l{`Ow!p_#~6Ije<-k03BKWEFf454X%
zw{`oixBeb<S$vx4w41Yh9rIE@^*nE!_t^A9jQ2zl{i_e!?Iygpu0AuNF7VO!`_fOn
zj-3oRFY7K-y(myiKa=yx#=A2-ns<DfQ@4p*wl8a4X<OVXN1v^BO<yMc|E*N9#(Up4
zzEezp6!&krxb!2Nj#63DoZJG<VD3!s6RGFq#mg&ye4F?v{r@fP7pne>{d=YyZoQRL
zzh=o_e`ax`*@6poa^2R~79|Etoy~pZ(YQG0;K^??Mb~2|=xvi-Rra-7i=pz;2b+J-
z1dkfew0X{QiSy6rS$n2$ns3xHd0T(CS=UGBN35B-99+R#8#EmZraetDxwhot$EMk=
zuTC`Bga!qj)I0o5Wp3AgY3oh%_sJbkl=&WO9P+R#J+{qs{#&1xTP)FDc7^d@tF+c%
z&0C}WUibd${rs=qH%)s#?aq}&^LAFaO-nxfx}Y(Vv;5hkNs1R<rC*!=yuoUJVAA!%
zf4o*McU5GLAAEVF@8?S1bMhWcT+^AD``EZE<}K1PJe=LM*+Y2G^=OwR7n@?aELJ<@
zm|bx{vf!WupLw#yx)~FeTxrU1NLDF%aJq5oeo^;|nHtt!R$ESf7P-@XS(W9Dp47Ym
zxi#M`&97V8^M}dLm@%>VZo<p=zEyK)vo2?7y?5^WvVHZ@zYb+QTem29zwWjdFAI|v
z<_SG4{Cx4<qj`D1eX`!4-aD0R!oU5zs_QBbdCz~+Em%;nu$is(^Z9uUmv4$a*KeGb
z@PemA-uV}3r2c?8cL~4GdUg+qbcv3Wl{#u){vO@4MpWtJ>ZFYpt6ft<w|3rmqWi<u
zhE1WC-Qlb0jGxMDLfSp|$>rsS{V{r_eBmy~mHuOA<mMSY(MV(p+&$TkW#NLdM%QM(
zZf>JRhq~R@=eL}5V|dPRVPb~o0qwZuhno(^wO+U>G)2?tV^v~?j;Onwd3PoMvCb(v
zYyRYlC)ye^KD~AKdClgjD~zLg88nR}-|jgxBWLZ^_#IUj7P;H$@7o}9MQsgp!QR}U
zs*-b$qTAO^n8z4Z`u~3AN^38>%_^6S#aG2_+~lEB@*;jZf6Vf4yHa1o?Fiid;{P<!
zxS(q{o-K-Iyz0ul=ZX3Mtg!uu-}o=NxY6LTmAK5ibrXE!7XSJ9s3%rkTF3Bo{}sQ9
z2Mijn9NV4#C3gF-h`$@=9_>{=+!MjJ>dCy#L7T2WzxGzye}@h4`b2S$e>Y{7?f$-7
zv?r2b?r}kuRLOgPK4wll{;%$5!`4lH;kCxEKIS+GWUbQC+j?ta>6TmVsav`Bu30BN
zd&keohfRYebQRvZ&c0t<cE7YEO=HR~ri*rmmadJ^>E=kT_AvaaJZpg+`$WcA&ypQ+
zg`r=M=DE#%ELr*B&lc18=_bqSs#cUq9d^GNW|I*jH)ra<oz}Gzw*ThHFM9f~^`(q^
zY5n23$&ZR9^O#OBXs+Gd^7rlOOFL`q6?KiC-_&Mbo4Mr%kErjo52lu1|JvV)H{B}6
z>G^;A{<L5F*KNstxajZS+R#&M5iNn~`!Zv8|6Qr9GB-M*E%Ex1y}8LQowi#)H=SEu
z9y?9*MQ4?m#Y)euj<<U2P73Pz^HoKzJ}kFabFaXGuP@F-u<ntYBlh-@XZlaG0`I-M
z-t1VdwRb|N?Zw|SVs30d^Ll;5<M8>@_1cy__U~p7RImMW^sDpHS970wH62+lp4zRX
z5-{acam-@&KM4&#ug5ZMWBJ#~wJFCwW^q9)Khp{Ox{N&)ztodiZQreat&_xX;QG$V
z!6NS~KQ&+8d@GhC<KO3dACvyR<7&-Iy{5wc!B?T`=ryD7kL2txuby-7$e!}U8z%gG
zefRMcmq{N^2v46s_gNQ9#JA9uMg1#Nl=qa^F@2c*eg1-v@e@=^UN6*Qouhh}D>uFP
z-XGQ1T}KaZ?+?0F_vQEEqrc-tx|Lgbo8;8LT$$Hvm)^|x<iEi&&J73O@3Z#Wc#=IL
zWn%ZF13RCb(|5GjEN}MTTkq8pvG3`U8|`Psn4aC=U9q63_x}gWg=dbW^^4>Qec2Mj
z_WNS<Z-qVUXJmL^5}P>7gEy;udwkT}V`Y14<>&nrxNI)JFM3b<9g&VM1FiE6^O?6!
za4s=?_cwaAOg&p|CEt~*eS+^}?Nw)dw)$VE^?1@|tGe(5Gg_n>Cd}E=>S4sSsrX0m
zjX9qa{bp?l?9Ww_bttv|d|9D#1>eO5W=|p()~)RBntR{&>7GjgFXlt0^%PPX=WSs4
z*EcVIn?ZKv{Z{r{wxU&w64RyUO!k+X*RV6$<k@kt*6Kgqu}>GB>vfw@tfn?Ke%H#7
zo|etbvMDp_pK+h@pO%>SwEK>MweylA?V82b=QiGLExsAZoIFM4)0Mq^#bN%3R6Yv%
zJpW@PQFONQ@06c;nv0&>MfhuYHmpl+5Xy9)P_M5c?<qa=;gW_YUkjZdO);uDU&`G0
zPupPsWT#W&|873L^Ly`B*Big>LYM2jK5uBL=M&Rlsq$N}yVCyo1?#?$@*6oHzRGQo
zpA$dd+%8g*Ka->T)Y*c*l}+vYw*54EKJR&tx6qmDFVDQhpDg4tWG?=qo%XF+*J1X~
zT>`1M7Aot$Ei2>SA$O~K&AC*swdeP&ZaQWj88Q3PyOM~gZA(KgT<p4d{m;D%7qgk?
zO;0}ByFBBvkzJS+^WPZh0^Vb#cVriAmFeI}R(nv(xMgw4zpnPl7j2%jNwwbz{Ns1!
z;7$$kk|jN7rtu{#FHtB_$`E@YH$yJE&1G?Iuwi2R^__XkHf$H|_|)k8sB4m1qv`sH
zzc=>r+1fu3-}qrlr(?vcghaD1?f!l;VvHX@s2ul_Zm`*V?JbA>)rnoMjMqLIvwVnV
zk2Pc1XQm=#`l5p^glR#((_f{)yHk`WF)mwQwd?U@>!<B%hW{HiUYrxC;ug}8e1B;B
zcFBe+$9B7)A0FPlE7}~jw$EjPdGB?%k4|6Hc{^6#`o4LMxxIbZ^h*ibW<J<jYovGR
z!awdNqo~sJQc~eNKi@NaKXH0>xo5D`63_g(_w09R>7ILHSoi-@AgB7rU;M(SYfASR
zN;Z^zu53=$y1KAtyN{Qi)2CzUoJ*>`0z41Qw`p9;J5Be2qmLKI_2?V?eQ)Q@T<JcM
zrReK{76bO6gOdf?9P4=QeQwJ7+$MX}x6@-&R#@@k;OF)ElD}9RCYHYb^!nG8gL=US
zYjrE1%0{)kSd}0Dr{KyX(1bW6XCfEVva5YT`^A1gIDYS|?ZF)}-_PFuDZc*XpI7p0
zx2=`#yRJIjJVM#+Tjz?;VnHjH*4_;FKfc#q{&P`n_uUtH|5NWvZ<KlZ^ODNG8@X@!
z-Sk#wZJRYG_t}G+zpA1WgN1&GOYdnluHN0BU2l-p`Rm_G>(jmMH_tr}YUA6#ZE=!>
zaOa_u7n{<HTCE<wP*oN_#h`s7{$`wEyR=EQzsYgKBcfki4zDb)*}r0sN5%6Mw>kE{
zzr!@c`Sr6|W}W>ypI!Ivo%`=dv~E>%@C(mg`Fz(U;a2uJx9-d7z1;EYXT0CBA4Ss-
zZBqMqe0zVF%vG1x2d~_G;?B)~z@Rp>V3YPu#sjlzZu(4`F-t@-Z*M`vr<rl*vNvWZ
z6r9Vx_v_-+fA8YzQzQ2=DO{dCz0>Sx(e!f#wM?&Aa<6_eREx~wulZ{F{reZWI?>u6
zXV)r=T3=|MqrY{6`#gmmBHpuOInVBW{z~!G%8fG~d{%8;u;z=-^OVVx&ux?tKko5+
zcel*^FQ=YAJM`CF<9E`!MLtHS3j=R`m{;>QTtTW)f9r>NlP=Fa<3A@MnB$KZUuo&h
zc@^*Tq>GlGW{p{J>w1>?w&!nuUudg-b-KKJHlvT8jno&Hc`XxJo@`Q||F8Xq#K)(W
zAC|nAuI)?MA)Dz`TCr20;NPlU0Ts_b%zgeO)8Exb(kt82zts6TsPUr!TCZRd%@F+3
zVDHfiMXf*n#Y-7Pp4hy4x0<b+wLmYB``|9t8S1t_9@VifnSA|4kJE8>w-vrjKlE6)
zY_My78Gg<u_}TSdhS2LwJHlg>=5kG~Q;f8kTq>mVo@d4M&u(F*{;xEDU!VJUYsZp5
zAE!2pimHm9UAI%>V9BkfY}TfiG0`fF2c|iti|^B5H<({KfB9tV*i)}$Q@iHgEj{&m
zpWMp_pO4G;1is)-jm@(Vuz7yx@7}F_J9p}TPdK{G=;MvKi;r#z?6cf-#d|r&v(<}?
zxewl1*W%|}lD23^^Gjv+BfDN$?VLF|_6kQi!`x?O*1OrYAC$Nl#8z&rsd%<qs{eOs
zoxX*g+hafZxf<urc@>F1{Cwc+WBvFW`g~T4H`GeXixyt){<WH6R=gSWIrATSLB}4r
zK3}K!GVsyE+iqHZ+NW5b+rK%!)Hvwq-}7(ZKM+}_CtM{j!@a}d(7Tn_RhjPiiJRSh
z?Q|qhkauINlH(eo%k`h9uE@#|u>4=N<3!#s<MI^gqZ_JMyM+9Zo7i%v<7*(BV47Rj
z6E>Y<#jp?d+uE$Bvsy?T;9t4tfap!Jbm6O}W{K<AA7;hRUa@wIQ@Oi9<-hnvuDRJg
z9eFb~qrOaMUUK|%x_nQ*c%5ts`xLW3f6|X$v`ST-FO|Gzj-ke_gN6$N#chvSb7y}I
zH5ZQArM~mc8f})9AGJP(x$$~cX)`<QDQ%SIH*;oReAtTV<*Q%J!pdhVEHChVWqn@V
z^lt9W)}vOhrEiBbf7o{`uxzFAZ<*DGyYEE*J~|^NZ}q)%!Rn5;*Bw*6{cPjrZ&#n*
zxi6gJ*OT(a?fVh&-G&#JTisl~`Tg4!v!<`B*wz2;`}w`As^=csCB{0p+wA4pch@7t
z<$iVPemI>dJxS!^mgP$fw{18#=~VU-xkrX7`yv&Xs%L)gl|FiWT4lurpSfXnHG3Qx
z13V8as_!+PE6iILR$RGTdr^A(m#nGRqjbUoISTaTCyTw?bve52!Oq<_pAQ=5Jz&(x
zoquIQ^`Et&tm{~oh&8MY;?T1HU%2ka$xvsJ4Ktq4KKR&o{r5Y<zK_noPwk%1Br@%a
zn9cq~{%-}09Ovg=3Xd?@InBu-yzAFqo98c2?2|aad6jF!rO7#s?En9se;fSEj^o~o
zto{n7^{W~APIu0#u2oCeE8ERb#U7kduJ)_O_37F3pJTr!+8&F(T;TAB$#cHZFPBYS
zN_*Vj+`oUY^X)rN-}N0~cjjB`%P%R&KW(*VyZ)m1$_=`$_pJ8U)b0QAP;};pZ=Rtq
zqEaHO)?CPr_B*uSE{b{H7o&>R-vbX6MfSx>|8IJ3veHRoy}rJ{_ZZuVC98jV#m-6C
zzr!S6(>SkS&+7Qt+&2}Q`Ndme?i4oV94uZaW+#4Wa`fLlCyicjsCe)wuxQ(7o0S=I
z^{+cl7spNedTrmHdshQj-(9hM^ZRD~`?`4{9tv~0pQKNWnRfrQ-Gi;q)5PDKwtk<K
zuuK0u<Lko1ZHIoZic8Q|cmAKTr9qaD(JJ5s$7ZMM7jts6*T`szZE`rlq{!W(^Prv8
zpnvy*58{8rd45z+%$9xGE;WDEgk}CrFXzmjuzmKtHFs{j<Ilg&o-eu3g<+R=DD%rb
zYX86On*REY@$<rxu+FyP;~5uzt<;p8IlH4+vFpt4$G+38`UM$}IdEyFyxWyC!CUqD
z#Ct7U+OiB{WBab};e4)rR#*0U*S%9&mgo1__x$N8uX?%pty@LK>5Q}1`lnf9{hEb#
zzB+BI@OFWehq|u%+*p;``&XZT6P}--tGL)^>z7b($@p7u-^iI9^lX0g-gy1e^{<SW
z1rtwA-)*2(rkLTGUsuv@<o`9bCM&k)+MW64>$ums$!YA|_juOiY^Q0DXQ>#P`E;K9
z_qpD5a{7wdJbS-iW@2QzEXsF`!{W>V(Q|f8Ej%AA*x!1&?1)?PI{Qe2dsgew0R57+
z4&H}!4qdvS^G&W@{d3a4T@`M^R<akq7S!9U5PrP8t$QZ_5$;$Xr<mJH6M30V&X)S!
zaCmNBVDN?c7cJ75IuCI1olVG|@&AYN+OXMq>4ldAuFW+xuv*}sX;Avx_?q!nzXfkS
z-v1XaT`O3BH=u0VWwmb)H3h%kJJ<e3tcIJ%izB>3*W+o??L~J!=g*x{Qt@rVj!$B%
zWBJ#ARb2mWg+>3~1V84YX?8o$IeuhX-nH$vn%Yy={q6V9h)pUnP1;u2{bP@>;ri>2
zb_@+BmgiJX&F$1Xd0S_%DaVp>GevcV`m6cdtFGR?*&5Zd%I?oAMIRL(uMDNV6I5fl
zZL>AMw1u8N&hRWePv7~6jGFV0%>^<G*08*Kd2&x7e^FNQMnT>HtK~Z!8T%|Ff12AE
zOYGBVUGOcGKR0mCysL`cpLQ*O{B{5RWfvw)_<nn~44e2*{(qaef1hJJYTP*IKF@j8
zjyp5gG?kxB=a{K_-sbIG=6}Uo-6yPBy_BD4_Mdx`<F)^Mp1I)m!yWy}$;}EY%zyIC
z6Rt6wv)Mmp2h){P)eF}-pMP0jKV|uzOOEP?TODk_Oi!wLReiMjQ0K$_3}=k47|L*S
z=1nn}xpbHOZH9Y!)tjUa)VFxJWo(q}p6=A}qbRsN*Z50F*LIebpLUu5F8H+T!S?7W
zu5G{nJ}y3<K4am=ch^@;e>mM=VN;W5;$(xDx(-WNUsnj_zpd7K{c=v&?vuOb*2-_G
z+kUPrFYV&w?(aTrvyIxBUUt9WJ$_{K${p8(w%D=j_;*R<^%Xci+&5AAYinO@l!A1;
z%pZ*xt3K>Dm({;;;=*rR6;Ngr0Od-@%?*AoZZ7KQSv5~Acg%KPI^*-z)sFE_4{cs7
zoX)Q$*tn#lf-6BMQ{bhiwZy#jf27~)m8AVV{wVp>{pg^*KeW!wc0Ofy>dM4U*#jld
z)=fNAbg$(~eX#r1>30_|`uHKt{dMKJwe{7vIi^oO_B&pra_Ovv=c@nj3@_YJ`FeZs
zRJ9i?_ui148+kwFpy_0$lv6Rj+ikYR&aA#H*<O6x;bhH)E)RuT+iRNsNq(NsnW|L(
zKdL(VJUE8+!R*iF@$VXY((cR+Rz4BxzrOHZnOlpIpw4{8rng7WWn6IRJw6>&=a=@1
zu?D<5XZ-&Bj2YFPC$o0;RsLCLcuGysF*hizDE86mA2wUhSO18=u&hr`WN}W40oU~V
z&&|WdI{Y2Cp4@4$|8)KTmlmm$bJ|X=JTrOUg`EeY6&w<(+WUGLyB|yVTzSsXe3q$d
zYSN!sPq-HU&0c6MBeXI1fmy$e+A<zZfoH8}A02SIzg=?u2`=%gX={%8FhA7b<Pw+O
zbB<RbVXN<<LkTi*8$V28FHcd|{>K;l?w506ZZ^xFb2~h*f1Ar$+gB3nT-}-Q>e{_-
z<NYZM-d4Q7)O7Pn^Ym$qvEBy*Dt?@=|N7$TymP0E(r+1FDC2(iV`7cT_o;86ZoPW;
z+K=;nSNtAx*Wb77Pe|!Dy8XEP%lhfx{?GZ#aD3MxU&E#KCEL`w7AJiBI(wOQaJ<Sg
ziEmwVuYHVbKPK%MXRi5Y@i|W0RAGy)&AKdVcMp1+8oa(|r>^Mk`2C^hi&P!kl}}6i
z&d*Jrbo~VH`|b^~YZDYo<(u03^$&GUyTyGXI5z&*0biy|n|>%AeGvJ?yL-a-n4<aB
z+UE@?3UAV0c)iE`yriV+L0-=9)vwoWH~w8x*8BI3*NluUt5X)w-Fk6q-Ph8!Wy!MJ
zUcUXk;OTBxeiJUuNx#%ub{{C2_n2v3AUDh9m$J`2x7e`HGt;zr$u;+L8sqQN1$vfR
zvu=NToVKUies+n_%hzj5UerKp9f3^_2XqXc|9yS*leAd*8DlYnw_2`CZod$Ie`AN5
z>iju}FaCTo)&KL+$@XW0yoA?$V_CNUL*c$j{vJy&E#Kbbw>aF)F=E?CwgdK>EK-^!
zelK(uzF4|oX9u%~uVqZY`kAW&R$e%2Z@)Wlh4!S0wWa;{O4gs-+G6$VO3%)Dv#ch*
zznb0XWx<j7gHb%FKH$aDzt>$Y*J;>VT@?5t$2`k+XXL}vwhNe#t?hpwFefx+`OW|T
z*JuQM`6MG7%A`AoJ3Dv#>gs}XH<skte%{xY>+iSr$6@LAslJnM9b@|_v@e|hjzvtM
zb60TF%Y8lP_tswGv;FogF*Q$b|Nj3GMLb9QN(-!%-l`SzP3UVEyqd`1bmjc2-S;j$
zKM<uLk+5rV`w{;)Rmv6@e>X0-VSHdVPtVqAc8P-CR`c!~9Gg2fZtuFe-<+9gaXZhQ
zy8>0KRHA|=bwASPd8VuudWlhjU8w*2hXszGE>+3&dQ_fVvefHS&=%!49j(`yUEB|;
zt$E7Cxun=($tI1;j6cS&SmxWj_ni=NQS@?iYwB|Q<@;WJyY_7R-<kz)GmN(QmwCMX
z{3et+{Nk~D=jN~c#=qrmQzm=R?myp`E$sfkVcCy2xvjm68_K0V9G9D@wL`yZuBx`K
z(*I6w_M8bOKKXYnQWl?0yS?ITwe*>N^};3oMlvEdw;f#M=UI61YsV8lOUDc~d0x*G
zOv_I*ec6|E!Z7u7riaa~xrN5!j6Vggme|L>yxr(oGgW;3_jRv+RqvIU$Pgs=^PR^N
zj}<{mb{tELUryn9t;Xu>;WwFe&9$I4fBZPAS)Sc3`=g=!e1bjWqU9@p-@lSpeClkV
zH(v<9)J63@&t`4(RNuGqF;A~{N@M<J)(O+qtK!1K+9GNe$VT!_a+F|HZ@oAz>*4(b
z&Jynfv9)tA&hS~+{J{GS1A_yDr;B6A4OzvE^JjksE~rqP%QZRWlJjg6Ri8~BiV~X2
zPIqrFdj7+=*eGWC{UYzQm%APupWV%9<@IY)62r9mn%^Gn9`5!b-y%fStd`~+w7GV-
zGFe?{f{>4TufK^<(yuChiLWp9dZd;qv%K_H`MR`v=DRw+kllAHpZ%3Qe7Sp*s6y<u
zzg*v4f^5BR$;N#yujvt3Fu6Nwk7dWo=lA#czv`7M{d?i}foO$_1gqbxHZupYKj^KQ
zQt|#qTgc;63GT-$cdUD<WiUNmXP4*#tE73Qe&uG;ij@q{cpSG1+w?P9%<oZVEUsGQ
zBy#26$7$S5bv%=PFrBz-q`7iIeasE%z{EDQqTceWs_`Y|OJCovv7J!HlCeAM^2c9u
zUv6CXjPvD1z3Tz_&(EjqX1|noa2xaFId#AG3Tn@a&RtzL)#O;XY2@Jvf3_5c#H++5
zv^ko8EIGH<<Hce(?i(LAXFv4c=%%R4VaZo6vywAo#^PhooR2z*y+~m6%dor{yLt1P
z!$0r+OZ@dh<>Ttv?mw=aRX=dp&wuK?DNh%C-}y+UT|H9DOm7|Q_Qks=DF2&#we{U{
z=lRvAnAB|VJmpb{T+1VFc=Bw)+{1HzpAl1W?6(QX>XW{&zi3)z)wv@V|J?s{sCfUG
ze2Mm-X_5!NY(F=}kmb_1+uYKR`KOv5_kS%PY-we>$S}f;BV~T`Rrj@3)u+;;ehX_|
z-^#jXZO5(mjGm4qGep#;m3*Hx?TW7F0{4>XcW-WGWSs69=x8uE*H=zZe@1tEr_oHE
zd6y)Qp7x)xAcX1fms3lp=Pz(!dU@r0`*r^2LmOUq9CFxtSJt+~M`49;;=g4JCA~gu
z^;Q)=?xomYws>dc#*O;l8)nOGR{tJYq`Fk$KJWXf?Mt`js3jbC*m8Sr`n~y$1)H68
zdt&t5OUxQeF7>TCI{#yB@B4-3lJ+KkJ`BFXnZhqu9P*elVU8SkhP!LiMf;{FWwJKs
zS?hQkHW{<7e{{I_s$Tz}H$RTwv^r3^+fUx-fK9+*lSk9j&wbc2FZLwcs?UnHU969k
zc&>T7sm*EHRbee}6n#Fb!80yB?HF5L+11T=?Ukc%-!Ka`=`PEkzK+dNZs!D+`%GV?
z7EV~b>ul{&kw4GQcZf+GI}>o}vf>{5FCw3wHC_#!A2mVddbr~C41ey*H}9F(zwj;o
zrR^CeDdaOty~=;$p;pd_lg76q-?hwpS6p!W@brm%DyR0C?a<A<mUr2{;a^Bc#1nyy
zoi}F$AJ?i~!Xzsb^}%>iWVpU&#qmjhe*X5F-R|qte<rT@`iiA{mqgFneXzzp-n!Vx
z*QeiV-OeO^4GE9i_jiB!@#oE{X^qD0$KL#ZQhhj_ExV%eP4R|bD?;=AZ~A1{J3Y19
z&TygklwG~e72E0W&&@7~^<|r(bu@U9OJ(nrwu6(_GF;BzpRr)kyge^jE-$yzjH|dL
zJ;N^{KuyOy`Jc1Z<b;zq6bt1t^O~;4m48|FUi0gUgT>{0&*(aTd@6Tl?i?kL`oll7
z{Jj?33cdZj#P)O1yS4M4vt`xnQSW)9;ClJS!9v^0Zl~PRq{hqK%eJroy5a1zKZetv
zKM9;Sy}4lSboD#^LLKebcgx0ojGb1#UjCO1sPzbHMn1@Eti3jIL51X_I;$@`#Um0<
z{t{|=GEc<ENBB%@puWT_$(ya&ayM5MNUZv4QY*>vGH6}pjqOkN@IJU~BkQ#9*M2MW
ziw%prHCm+RYsSSlZWFiJrNd_Zf=9H+>cXwRSC^kYu*X8|Nyfv1&>qLOjo(Te_&hyX
zGsMpu99$DW^X2Va+4Mj1Z&rK#zx(P@<&Q`c9&bMV*rL-Lrrv2Om%q2jHdK;-t<El!
zLrb!6f(C1=t!Cw(y!Wa{jr)bRBKI*B7Jkzr`3eTR>dkxYS8e2)qHR|B#oc#$Dbvm8
zdeYoQjq^4|PH6wQHZXs4bohgZd~#K5H!WcauFT?`RDS!zgWcTzJA)7GnEvwI^PCHN
z3<N8kdOUp!)nuo-t1bIJMZ5jph5dUt&m8cp{P)=Ap=RgaJ@3v>4D6cJ_V}DPd#C=y
z%O`E`PqMFSj&--)pWbw%f3FJj(zB^~Qft3A9I%<(>vg{TR>#Da9)o4#d+ZNcCWQa9
zZOqwjZu5E4G}oqAjWY7@51!w6Q?aqcDnQmrud!>ESb+M`o%1s5--WK|`!AL;-{trE
z#oMpu94k4))LwSP?Zt$pjt^A|jvV>2ZC=|iVMqkdVHDHYw6t^WHP3Vh|NVPoOP>4v
z`4+Y4d+(lJLFQ}kEdNZ({^BF7w`kv52dnp?OU?zPU%knrpz8n9_4~U0oAr}rQaZI7
zj%M9&z40%0?~3(zU!D52xOC>yuen`swWV%P5?iOKSNm!b<MN*M8@8z~JC@Tctk9u&
z>E2vcW}#i}|Nowz9I3W=gWjCU8_n-5375R_xxmgYWU+$##>8L8R&N&avYE~`_ie7&
zhv#wMw;8tlx?C)FfZy>f<K=y4ZLccrT(?xmi7AL_jjb-r>zOQ{53Q_a)wEcqDEP8r
zAD@<xf85KSl<FVf0@wY~n$1vTyfJM->fw1@h8zz*ZVBvc>b(4ZU)F|-M{gH5=BYjS
zb>(}>)o(^0KP^ALJIDF8twnjvCT@@V|J_O-Uu}A<r!Xg^IqK)7tFz6v<c51bX1=P&
zyKvk2mle-?+2(q56<^qW;_OU^zW)3DpVhtuzN;*)UH#+e=HD!`5gOd*-m1UbY4zm8
zGKt0Q|6iV)v$1{ms*RR&D+Co++lxP2X322LJUKC7<;JuX5wP}Z!)(?WIVHZgEen=%
zuG%~!diBD7P*xMzbYNy~qr~KmW(P&Wx+-537rZ+VbLHM%*{KU=G}k)azu~_3M}Uje
zc9x|xFE70Ew*T7i>eb6kx7}*m^)7O9@bq85&UM|?&s{sCe9NYt$_v-BODj0;VtD`6
zAa2$zd#mNk_dc>M(Jb*=%v!Vey42OSuG(PTgsp35?G5;HV_n5gE}!{p7KOdc{JhHX
zvwW=KjjMZp?XmgUtZg)vW!9e>rYi;YmHN?fTt7a>%RkfVG+kM7ZHb8C)t8p*w<?-0
z?O1NLWGDBQ3z^F<N|st=?obw8IL++Jwr^$pZA-R)Uc3KWWFxo7^17-S*`lTu^QK)n
zciaEjvq)iChYeBT<+n=rv)leXGv8h=Qzq1iZ;InXHFnoNABJ0VKdubX&YziJdgjH9
zLu>AYJ!8658TibhV7>4!nHp}E(jdt{@2ay7^7UmqH*$2_@y(hpr6c6$>-%P9)QcEc
zSUa$DN4#m<HZ}Kov$6Jv$%puOTkTqOz@6j!2HX;@=Y-Z>K5Q8KhDpAXvG2s<#jC$d
z<ln!X9De8c1@UhuFK&-Ll%p-A&0lf8;_BsEo`UL9-EB9PUz;l-ov`(8(D$1UcU`$9
zR&mr>Pg+*q=E;iay*#0OkqRk|d65V5*4xj_y?aAf)-KF0>csr=_85EZSeND^U)?1O
z6PTQ<F0%!f$*dF$Q@_r>_n;<A>vy@=KbAk6v(a<ilBZ{kKezvvVJnc?^EbWe>#LL(
z-(#(>u!dA^X0O~3JvZh_k<P14y{kA5Ij(&BV9PF_^850d837?*{5xBY|6l2tZ(}d8
zV0F}%d(4sP30c*D<&&D5<(@^k$}$#S+!7ML^?j^Oz?`{LpZt3BvUeJ&A8q^h%zgG#
zQ!h3rO<h=U_~EyoeSa8iHmWaw{P}Eg)Uo45Q7s2{P5JVK<B8U(3G%ybHpzolX;yp<
zpUh&;;B?8BJL7*@$@;nTR_C(cU%YRt)0*GOvuAAHAoXsFukH1U6-yx#tPf@<GNv?c
zTf1q_{-wv{%Kw=syDSWZMuly9qgkT)OCFCGks*J#y_mm;bEob3R`ws25488QxUbPF
z{JMo>&qdxD+R<-M3!dB=c!M{-YtbniLk%^zm?@(AVS4Yx@~!r+UE2P2qgL^gi$%q6
zmegG|%`9A2c8~G-j@M_x4qrQTY)R)|rR<<2pIH?g-=>-!Tf}LoXuC^R_tlZU+C{yg
zXV-4)aQiUne|i4XE$z?0-4;)KF{|!q>{9#M((4cZ{cL6q<vQvOSt(Go@Yk+eeK!8<
zf1N!pJ;{BU%N2efj^ZO_*4iQF(LL5(vxJX->0NQUga7gq7nfO+oKv6Oc*(qI^<yQT
zSpnMdx%$U9FJEFPW-&SO*ntcS%dcyG+<I<i69bwTn|dWmseQSN$$@*lcV8AR_*%ML
z|JAOjcb&f^Qx@Iv{mJY3kZo>mT3XK5sPKH*#k=~Kf8HB<BRpN#uHe-Bc!S?t?AOQ1
zFwXyUqH$wVgd5L+%^#D#o-+yz%@=HD%sTJ5>7<H^=HBP;tB(X0%x3rY_*(X&VsH2J
znQz~mT4$-cpMUSAr4zQ#R#>-NU`oAEiB999z18wx%s($%qH)FeFN4Rus>J$_zlz>f
zS5KF^(O>hLC2;<)KrxL!Z<6Bk)>h=_7P8;_+P=>?Xu`6z2)8X24}Sh-ye;vA!E#c{
zl7r0VeBzNFE}jjtQ<9R7%}|Ny4j0-D8XUZ?u;otkA)|M{t)GOO^M~*F;{K&SaL>1=
zmAc1mkH3mJx{m3qN6ePMC91AF9MZ45K9{Zs4cC5oV(hQ~asOT8u#M)QnI>$XbLd55
zZ?)_ZiGMxkGiL?!9Th!$sO>$&-V4k7{X#NgJ-!$HyqT$F7`FSe^vm9q_?7ALdB1I~
zmtMTksi%9v*1%SWE4pOPB`uTYShH)!tR>I8%$#S<*jm7`r`~PxSru{XM!uJw3<6($
z?48z@TxWX!Zq=LnNt-3}=FW(DaQ@Kp-EZ4p)qmC%s5-XiezJvL-JhwYmsaE_{FwCJ
zv3(shg)?q)NS<vfcG+)thrby|@@LMAQrki$)}$U<XWz5O&Y5Aw@x?0}3x0WsTh+4M
zQ_j2nPT>wufIPzu;RWjsUb)`0vw-uSxQ1oxn{P1-x628H#m?h!zQ5>D@1nCNlG>){
zCSLz>`_CG2e*GKen<`)EpL@G|(V~k|KhM28Rh-DkJV*ZawtL3EmLBvJJk|C0a<l15
z4?e5y8@6rLUw?Sz>3LGwYj*s<vxT?Z&->)}EC2Gt#d_nnY|pAH-8{8S_~ga5SMl@i
z=dWG9=~Zz+Ui0<u*WbTiYqx#X$ypz+^Qyc5O)J}9w!|^^#^Gw++PgEN=a*crKAyKt
z>E_Ci!i%eK-k<uq|NgmG+yBg)x$^iE)5<$R4;EC+yr8cYK3C?e@Qvqvs+HT-T%=xo
z`<wG7d;8w`*6Ei&wq0BsW*vU!+e*oc?qYY1N>tZvx81R);8FB$t@F8=n~j&<v(r#q
z^QBkX$X;@K*{d|>{P%a|MbAXc4mV%6>7v&>2H9&*w_TA)etJIPw>-;*Z@OZ$-<{+C
zeEz@D`nr(X)w*rTeI~DFsV{Yxul{`OjA`e@WebAhe*D@~@c8*c(WAy25)Hl7T2kjf
zKJ@jgy|wz=?)md)uX5qb*n2(d>&9LO?s=V_>))=}!BBJZ*K)m={}=x8`&s0F<@^Kp
z?Y#BxkChj%)6LBYu8`7`l-&RLt#tbu`?<+HTh2DESikaO#CE-oUwc>8eck-{(MD@`
z?JL4xmhAhy=}hpcLxs<#6gS#A&JvtyGI7_{Ey|nLs+b!DFdh43)z2cNeE2KF<*Vs`
z>-`n}%=a!jdbzwmbP8yl$r^@Qw!PQ(o_xj}%4Ki5{<PgQr9kym=^qbgicG5iaKf<e
zW8_Lb|FWeH4F3|Pt^2QNaK>`3iZfH@UEkMra`HR9*t^YFtxvNa??3YL)H81Lgg=6}
z-o0LIs(1SgTWjAGwT3f+g0c(UAzAK=v48bhncEZJJ$)P6%Cvv$sSST6=Wm(*dt%(r
z?0tISMFKMzo=jH1EC29Vw}{<(#h3F~Z>@dVKQAcc(rZJ<>+GdYPlB(P?BZ^5tk>O>
zyY9G*A@^smtHSHDk19D$KHlb^`|Z%7FBS(?Y#Hxa+}(YAd1j6`|FtD|e|sJT?<dcD
z{qfk5%U62yQ#Yso+PwJdvetPU&*`MzzW4p(UTrIF!Ci=1sJl$Q*S7s&lV&*WF_CTi
zqp*opKNkFW^l|B-xDVxt+0~+^_Zej$iqBXr$8ckA_fM^unI&5fm_1Nik$q%q+8*=A
z9hIw%%9o}uZWmv-v1hSP0^7CF5`}XTR-$)vnsXQWZaZtYLH~{LZ!zJCZ0Ab!PO3kC
zy4_^^O_A>_ma=CbXL_)=!NGEB;@&iq=nG6w?$7q;+xz+O9_jw;TfR-czH!cF!^juw
z!$o5MerXd{j`Y*z*U_Hy=g_rJ+^Q4KWivDj%RBvCEc%S`)h4-nujS^=-DiEK!Ia_2
z*;vtZ%Wph3Q%bCtZe2e$Eqlh!vvP?Wx8}EPiFo^vWyjA|X4?xlFKHH-qMMx)d`)9|
z&kK)_LHCnPFRJnCZd=hEZ(sFPeCJo;ISy8(FUmyB`*tOJzI*dt@0{)Xsc%J#=Q`{v
z>Mh&*?QQaPVT(|6kMqAO*URk9`uOivafbD)hoyIz|8m~=XMVbSog&BTb%($IyRmQC
z&8#_pKTXV@F28cFR%6Mlth~+cd3qTV*1wl7Hs54#BhK@2`T6ia>c88rT9!_}>R|QT
zt!lyXYtv$WK0DXd_-D=6xa_O$a~r2@FAwFbD6KxKQ@8z*GS63WzW(P~u|KN(pBQ&^
zbex#|zxquU=aS&?<kg`+=gnVdxQRc#>&2UIrp@Bds@>C@Ehfa?-0C&y(Bq}PH+CP@
z@A6vpDvH%{fAZ0{jp8h?HKs8z=gj^deyI62quwO7RP6($<(jLnah#t}bKC!;$v)c!
zvm<xDU9xE6fs8eiFC7Vv6ssyPU3jAG^ON=2B?l6GtbX2c->DZRJuROtl-IGk%Y5!B
zSNVUHFQzg-YUWktitXDaW&52oW&hX8b_MmN9^u~`<gV0LnM-><zO86`Z}s_YQ@Xx-
z-0+&P=j_j#6J~pRpKmnh@BUl)_TITS*RM5)KAL`BK44z+r&(gt^}spo$FDP|H~uL8
zbG`kz)g9(bfA;LsiPv%b(jXFlJiYJ|$HlZeHbv8)+n*DT-mq}7*gffI;tIAwuDUs|
z|E$~Cv%LC<2ivPO>Er%w95z?BeR*Dc{H<cVX+*Nl53@Cl3*?w5L~(|G33WYi`to7-
z7ncldWxA3@eqH^yI(u_&GymUp4cl*Wi^lzUHF37zq2>o|A3JByi=VJ_sd(AO%X)v`
zJ6tso>)^d?K5wq%-H+jmFV9-oeRu1Dng8Bo$kxA*{G6kCW;^$!(?0nX`=qR-AwAC)
zodrjk!d9QVlC=34$Ktp7ha~=Xt&rKs|Fy5X^na#kx~>h!1Y38G&y0#nNdiwZh142%
zZrQ8P@WJK498QM?!V`XK%YD94FsHx8-I_s1`q|q_a%?5KOy_j0v>lR|-hUHQSiJR6
zM~!1y)#BG4r4ed6-`E6~wYTru>3Bmq%VqPv&>UBJjj77Nx(n`Fyx8e|qvS)^^|#^Y
z%0JIf4%Pj;GRbNGi8q>m7ju`3ntyvQoSTuGu~F*Zri~?X*SeQgRQ$2C+W-GkciZu)
zb@O7U|2pyf==3Wy))xN%x9Uw(b=s0NZ~xr95Bs;SeiZYH`KZh3+{u&o`}Fc=|4N9z
zy5sQEXFj4Q&L}PCn7(+4-AsF>yzk4~vzAXc-J)v0d1uXzFD7<Nn?ASb*X}Kdh`Fg;
zd};bS{oHrI_cvd?GezM3+(~=%Y_GNS3M%H!n=t46;;4JKUoWU=Y@Yeldh+B?SFb;*
zZu<QD;Y<GaFWRz?tL46z&dvGt{MqDU&O7-hvL((JebJe3Zj$$%)j6qe|DL@?RduD_
zZmE+^KW?4-w7K3$S#-7uU*((bNgu1dW>4Mub;6MkkKX%x-!oIt+f(>;<1WVc@7CG0
zOUwPg8CiI*|Gbx_jxxuet$!!K_BtlJ(ttndw0LjF?VlIce389zW`k}_uubjLi@P*C
zJUpti{&ucW*L$`9`pvx!vVYzhn`J#(<nwFU>SHQz_SD4$`FTy*SbA~as)OhAHD^2D
z-m_zE&8@e=zeK>bDr?Sz0GC(O?yu3<C3Akyyep3DW6ggsdvZQ`Z~O53>LvFsO%G(f
z-s^oU?BWeETMNw?M`!=z63m(FAH3c)Te$bw)`R)StGREw_C8v-|4op^W%lpgVUs+K
z0vD^lf1O>wJ#-=W%IqxmkiFU0>Yn~MnYQu4_UMP9c>=x+(^cQ^h+_UNk@L}P-*WM%
z9pd&UkI#>LxM;%1uxSlH?=Co|sOuxIBB^Q+6Loc;;D?jXf-bzA((UMX@8g|GYS%0F
zzWh>pr$@e;alxx@@6Rljmx6^3u?W|TO`q*5{Nsgf=d|h#b>*!JFYE+r-aoma{=$nt
z`_rq}jLRpU^VGd}%sFaiyv=U@E6QpUJ3Bh=OmYi+eWzf@uZL?HiXMtSVCh|X{(9N&
zUF`d8KEL|-<JZ5>vQLwE-+jICB_pwIZqbI5j~M@d%dF{ku-(Em+j`6HFL~kLyTAXO
zc%6?^EnYRhign*<*Qb2}ytgL4EKq4US2ge5MZcUKcYd_L<$7_hT6EI8$OY92Ra^@)
zXP-Fy<hMke$B#_u8;pD2#ohn2w%73C^?(<LHvA}fu_5{3!pq8AuJ&yR`&(Eqdb*1H
z-l=^C`&Wr_%e4!v?dQ5YyZtMt{_mGPSN2|Jb2%*i;tQ8rT^IX=qpvJZ-K*KLf7N~A
z_$lWO^-ZZus(SMA`g-|KOZP0Zv)i|=;>_;dtC#h@l@9&#ue@ztVEpUO??zurgWRqd
zF8<e3vhF>nG3!#^{PX8h|65jh7OqwGH0#>G`rDR>;Iki|eE3wjGEkIzlTg=|qIq7j
z%Rft09kYqre=b4f{L8poD?4{zTikZxB&64=kiwX7v#fS@?82ur`gVRyJ2PGSNTvUl
zu4TK_HhyIYw=dv~iMg1!s@1J<-_rRPW>0dJ*5LY@yFk-K#Zl16H*zt<H0F@&vlpCx
zc=7bkwRijOXH_kgo@H=6*KMv=fUHuvfW^h!X<i!-mN<2^YNb4>x@Ov;$$zYC#SV?c
zY&o9VWP{v!eV<zyK6;7@2`En3_v@UX_uM}(xq_Tt@mBmYs=M>=lxeg59Ag_l$^W{O
zCw22w-AymAulchqs)A|%roA;M_C8<oY=iRu8#P5d2g~>UzsmYI%uC?UbK~H?xKCB*
zQome!wsqZ;^5V<oYb=7SJ9ow14Xm4X<?-5klVAI#{Ja$Ys`-0SH_PSsGbYts5x&}$
z!@otBYs1{D=ASCkr>L{I-rrMr;?ZZLWYOb7#k0@;IDa-(+|#!F{EOYc_{Gx-7Q3C1
zmWr?X#@$}m!n(K4Iq$Fn=Q3ljuZL@|21hQ7U$e31YV6_ZC(QKaCl_q0-?Q@H$<n3M
zFPlXu`&2!>b>^(j>;AVluRYq^w@q!SyIk$}WeZ=g=sxt_c%Jo)NwGJLJ}hp&D%{Mt
z?DysLZ*Tj>({<+B>+B0TYx%!@OHb`@ui2NbB|m0TsFk1A&3F0smOI&Nf2}YUaPt5E
ze&76ZQQOxW4<>;I9=-J?9#y36s`hVK8aY39!q10IhMZl~cfZN^?+@Hoy-en<)keSV
z1?lSx{-zlDZ{5B@U3yk@ZqxdN4)X#7iQLM|H*fyn2wJst^Cq3+XF2?(h4vg{S=~Oj
z)9GiVSGZNG<yv<(=?kA{>gz6gDksnp>aHEI{_y5|ohw&Ly~?s-*}M9+wNuiX$DTEr
z;n&{3k=xKSEj{klr|&1szGXd-`JP%+^myTByHKS~E&5w4nC|brt6aJ$Vr$=p>ieKh
zXYE`g$9Pq}iY0L$8!mRoCpz_=-1lCVGme+(WyFCSIr`jlKQeEuOAIW(l{Djy+j8@c
z-EvhYbgV<HzdnEdE%xV!IX%tqu2&y^{$F;5B;R_`iucVH!t)f*=h{^s67GMz@zgYx
zs%`DY2HR&Jns8Y8>9Zr=Zbw9SsjvOGq$;_t`o-&|%0g3)PFY^mX!AT<p7~~3_^vAF
z`!nuvJ@_o~eV>@ZoQmi~<s#vb&iy$yfq&14GTZO1NO783Z5eNWdGm{>xlffJsHXh9
zx~8tZcJ2MppO-)OlwMtaxmNb_60^kzf7Kj1RBJm=K}RB4-r6GU;Z$j+e#WZ9B8i($
zwjTL!Ht&T%lVr~}XDeqhjxDS&?iy!S>^WrZ)vw{#GV{vo&$6kqdloJ0@ep|4Q~I`2
z>u$oPb$5h)yUGP@BX%%+R+;x|)#<59JhM4>J3X4rYM^}0II2j|;eAkIj;=`OL;23i
z$JaP#@AFv`yT(Z4*%fIqvF_LWQ#wBB%++W=U*xTN=isDLC1F2S-L3P|uRr`)HnD$p
z)|w+cHaGQNyqHvUKu;|F;gdss+Se_G9iAsO8++W=Nz9F9$p6kL!TXE-T-n^ag0J6Q
z{$L}n#_;FU)|vB9yDZrx!tmr^O~3CCC(b3Ov(5cHwEms#I#kv4lKu2-&-z7Y{%;dE
z;Y*S%3cqh|tQtLS`8)o6>1{Q?irFW53M!wkw&r7v|2;>s@ON0_sl`jqP5Sle+NXo|
z7TT)kqbuuwz4D#@yHf4wl5(^2=cTVrd~|g08|C6(y3-x3)*m$fFL9t`=?&KA<ZoUI
zbJg3Hx|r~G@LPMXntm!RdrJ_<|C+KTYQ1s)|M0wjv|e?B-unCXd9MWBST1~H=>L8C
z^S>K)wQ;fYZ;81_+t0Z6J$t3MoW7i7y+(EMO$Qeam5Y}RFYp=v`!ch3ce2;1!kg}2
z3uB@}G!K4h3yU_HT#$A4p-4vUE4#|7`q$MvKXZMI`7Bk$zvj=mCEr@M-QBA`y|?TB
zL%t=O=gyN;zHJlC67ytj+2+cxGcVmg_p7tl&rqb(q*ky>*}F<O{#B;t>@JoCA3c7o
z2z@1`EBnkd>NV@;nEcI|2FGtnm;TQFf6roCz<j^;th?BkC8sLrMa|H-6;<47`-xZP
z`{B>Oy%=8oZM|p5dbzS?&Ac|oL$Vy*OrL!7oaDD1UYtJtg7ki|wY+N|ZHia#O05q%
zbotLqNr|4As~aRVB<`s{O4)og&noERo&L$iYhu3q|FP-KqmP*vL&JByeXv{W+ah!Q
z{Pi;)a{BP6h|iC!f5hbXf#Fd;`;2`HweNO+v(Sh;kze`w$daaYr`R9uT3^fKGCQjD
z%eH;IzFhiFBF8)nD%+liZvCR0yFukJ_o)LL9^Jlwe)rGD&zHi=#2NlS2`D|bf1lQD
zf9byu*4SCT<F;SB?Q__>5}V|Ezb-|eV_sD$?ogGN|FVDoUxkh0#uv}eX4>PS+AOfG
z;@plD!|Wvaj|V>gZhls~ILqjZSxTdyu7LzYWyuo9C(Y-#Yz{cRdP2&JcdOZwtF9J0
zndXKp+EJNT|MAoEpQqFWrFUoMwau{TsbsbF&YT))^X-{n;g$^#r!d7{wQW0E$l+su
zG$(fcg@g8i2X@Ek2W^}zsJZ6X=?4!CrM~EFO*;3U%OG9v;LbF`w%C`~ws{2_?Fe)+
zJDhF1J@Sdpve$wKKCPSW$T|1Ku6rsy&Wjc*N6on}rh7<CtvS`!QzB>Lj#+&9UmHEF
zr25-4IT-#4cHHf2E{i$*M)W_sm)bJ-r$T$mu7_XD=zP92Z1&wBFJ?W``TqLLif217
zRdqA{`BWQPJbMynZpr-<PUQvj;ttg=@+plvQ1a^6+c{s}X7^q7bzqtIur>E-dMVqo
z-<QL;Ep)wi@9B)^lNLR-o;=z5?{4c_haRmeM~<Kk7e3GUthdI|Dkk|hr(oN$#}-Z3
zZ;7TduuD$2T`yg&wUp<rm73|M2dkTo>D;gm>pLia`|#^KSFe3rbSz{|zttawx7HVz
z<S1?UfAi_ui;IJmKdfQ*`}1;h_3HwWPK&6#KF#b6E^}6F*Elb~Pj1&WvjSbomh*)l
z8?QaTcYi&Xlw0xNlb@G-FY(`S`)+m7_nr{dPY3tLetn+MA9(lL%nx$*zodWv`RVa@
z=7sC)epS5Nq4M#=+gVHvF{|qS|56NS3)yMM{^_i+y#%w=RE@J*mgPS`#+PU7-4|x9
zp6y$mbY|~L-uk!qXCD)I<^6txgZaTPtyVsNf?cA0rrwwluaLG>`O1}9odzLQtG)|b
zNjp6{aPWjdxNJuFiQ^}>F4(kzV*(?`oX_@q_wv5ovOrcS{NijW&*^X2Hw4_^Y@hu$
z>FvBAL-sopde!_s{b9R&b?-ird!qYiPv00nZ+HI7kD2e^th1Ugye{2rx*zk5f1P`-
zMfZQ{sjN-^sgQq<b@}4^v*#{c|AFmfH>ddzn@n5T+J9zSD}S!|!Tjp^yoF2e$4@e!
z&AI7>wx`{B(1|iDol~o4*E*e_aa(rd&FgBm8`CG>zG?g^xoUm~pM%xpH7guXOk_@;
z+VeGkosFsEc}dF$0n5*Oa8KC#W@)@t(cit>mwD{@CHZiD@Y8eN?2HQKmR<9L4zN6b
zDj#bYQS;>~r%&a9`w!W6><;(v?h^R_<E6zx6}KnfLcGIE5BJPD(0U}F`vz;+{}WG{
z+e{9`6)$A}92RS5_g>KL(xw}IoXRgac-xE?oDEgVtZ23PemyO*a|2hh@Vj-vK7oDT
zr!D&?eMxMou-azE&+j=dX1!HD<iPLzuXJ`UM@)Tt(lno4GdO27r$5?L{I37i+o+}A
z8f3n5S=FyvwMwo1_@|Gru6?P`(+Xas;;+lP;jY&24weaNAI>>XTHJG&E%x{_SF<^j
zC5szs9zPMQFAraOU4OcL?N<{q|L519UgLlM^WjOs)=8Id1zYAn@2L`a8vT}Um;81o
zk45FD_DD`#!FAxQHJ`gef0<8@iB^fo*R<KES+C09etd7&fiKs~!_Jhwj%)AdU3)EW
zowN0dV7nie)5Bak_Pm>ZntOR&*^;s*tBY?}>SkCAXVlK!+O>Wo(~`|TZgW<i+nZDV
zp7WDY)t{v=P8#m&Uuj;yt?_$U{QBstvwy$vzQ1dhbj79AqfG`8kFx8pa7s2`RgeGr
z%Zlw4|1#g&xBq9x*$I3r_kUHr?6%64`u{o?_IjSVxphHF=YHnI)whqhzI=V;RI6&z
zwL|GG8pd436?-~5?6|qx6KCw{cloZy)m*D#*Z$bQsHOhy*8uN)W9M{bF`mi@mxjp~
zw+R>j2)Y=uKXi-gHRbKWvn@8<=&Dm?da+;Z;o2`del)ZD-nILBW@_^`W#K;(mli(>
z+xF%2vI&Vz@#5ykzjvB!F;cl4Y<Xdk%i-S#)h07kYAv>#9-P<bv`KAs7=tnIoX)Ae
z*I3lds}j9`M*eGhx$j%>-E{x={Bup*m2IvH{<1PLJ02`K&$+Df@4idVcAS))xHZaI
zZ|}=b&uwfCy0W5_)R($nKcy0RCE58zmS1Z2m6KK8M{d*!mPXmGlKdo{X%Mm3mVe)u
zi`T0kz25xz-t-+D&7~*H+t&U&*#6U3{@r@J*_|8f{#Ty7629*A7mjn2v^b}D7<$ae
zlbtRW`TkwzvvvD7pZ|9w`nB=3$ldohWVcW9sTQk0d~8X9<i0PS`+jgmo?kq{<N3qe
zZe1rA99RHaOyZRA%lvKM{K&@_HRrLOvtxevNric0%!^}|lIy&szlU{g4-?gx_lBo<
z-?u*{g-<R2Htc;RA`mMZ6wBr{En5EmLKWfUif<qO89r)@lw73$O0Z%7^H++GUcZr2
zO`N)#@%ijmzrqS{3tQjyZ+|h-Z=;8~f5K-;DTchm$#?RY|65M(RI=1?Rxtd}{LV9A
z3nQn`LlvQ;JCn4ZZ7&dO{LG=sQ}6oLM~Z8H*<@X=%h^>*%oWGpTzvS@mxJe=j!zU<
zZw<eZ#Mi~`=WVsZb~7{h^_hk6_wZ|fzp^84`S$yEE6>MO`tq(hU3}M4N%GmY-b(l5
zLJc*E**8lrgzuP>bfxO=%hkK~ZU0^w`pl!EOFRAzciXX)YHjfopJz<pd99kwj#G~1
zaVg`bPn+zw9hW-J`f0Mc{O@YH+uof<rYy?Gr`Xp%tH_U?Vm;kYvBtme!x4?-T%Q>W
zcV4?A_LzT%Tt@Ai>E@Dm&6V``uTovR^Q_#@Khr+lyx!xfCDGig9BCKveTn*UjpIdX
zGX7CXeX9SG%g@fd5UyK$r)q_iU~c<Xk<zIzyXM~wz0hx7U#TDM`ETV`t;r_e%wFc2
zz0Li`SyumN>;FI%$+NZ>SA2L>{kWdL?7s7_>glX8lePEnx+=9^cw(3jM`^LkYFpuk
zujz~SJ(+RiXHfTvs2f_3j>PSlE|UDNj^W{&XoI=0n#+_7e!3^AN;}(fuKH}eM`1$5
zNxpd9xkeYmzIV$#e_(q(a^seaqeTsqo7G+%(qFtzEHhhsV?f5SGs}Y_O4M5phV&F%
zs}h_Up?Tt^U3Q{;;Xe(A+uOVY>v_Jt^XIf~-PhommQ*2m@XN2T=qXQ_QZAj}ZLhV%
z&h}o&zf-0&PaFK)!4}haGUv-Y$y)pRlDf7{84vjW$I3sO&%XZV#7VaeG*1OE_Z@FO
zuP^aqs``bG+qJpXb#BHROWwRXd4t?`elH&Tk6)Jl-F=6*-$*iQlhI<nH7D23o)LW0
zsG)XGTFxa--L?z2&Sy_pQ5x=fSs<NB{bOa>-6IG6-!kss7I4UOdTsuZqVM1C-HWy3
zd0sN-#67#6=QS6u2Te)XZrk>xscF_U6BRF`SzoNY(~@p<?vF7_I9qwn#oA}yuQ$&=
zReo5q!EV38XJ*s8d=>kJ^D5Mzr}=F2^^cjdvE=5%Sr4OrS-f6)I9ygMTW(81+^^-;
z-m`<Jy;(iavpu|^!t9By#Y0!UKDo$Hqra#B{0iLhy-Qs5SfYQw@zlSCb^p(=cW6@o
zzHyp)gim?h1pDW&CoNleR;(q#dD3E@tL47FH+^P_#J^eob>52K58iKHd~IRf|LflD
zIkBAm^W0SIqso?~841bNF1?pgk$C!~;`;Y%=A3d>5#P8*I`6{P7z>@e^Tph3xoIMY
zK&On=*snW&=)CAWo0s$1->a0eKW)ustF5^Bp{1z)&D#qKF5%~A=zPz$IJE9*XVqcv
zU9Bv6`y>ywnDp;2b;&vWWWHyEhL6sdFN<<Me#>93758f8hjUBAnI1o3R=pxFxGY@g
zUdF|f=~iZ^G^8ea-VL^|xq95JcJ1X*`4y9EOZ?9q`F!t)OM>|Nb!s#BeVh8%;Bj4)
z)yJ|+LC2!^Z=KVBY#GmdP2K+9^^ZSa`S*3RvBB5EKR<HsL@%_teq(*;k}a!TZN3N2
zyU}9sYV+nvdx|DM%5S!q5x;(q*}KNR-N!TzSp6>l{ipHK$@0E=mrR6JH^!B#2t0o8
z_dDGs_W0VPHQ~=^e4dom`sh#f^B)Q24KEJm9jjz`%XQ!#!)4h6J6KK{&QIGEeDw3)
z*!D%IS8iyIYyURypSj|th~q!CCgq9!d%x}B{OnA7HK+SmE@kkpH;tZJIcb{J+^LEU
z*0l~_9<FWeP3D_2lfD1V?MsF(ha+#!-&b~R-JWmP)}2y3XF1{Tw@%Htxo#;x9b_&i
z`*LmQDl2U@T|04}OzD!^jqeI7&3-5U{#Tf&m(~4z-I*BS45{hs-aPc5A1<@<rFLUU
z9>aw9@s~vYzCArzd0s?xk!$$dRqNTK=0|N`{=HN1*s{~<YBe9H^Ddo!Y2nl3@=wp-
z|M2*t{HaX`cc|Oix>-&$ez;@1;@1Z=FZi#mO__9*w_0nf7=M-Jxy*AVFLTXoFTZ`6
z{r~S)``4`9`%YXpY&rFvk9&n<_Wf#+=w;&D_H{C>+BkXrZieHN58YTEb*7%-w%OF&
zq^JA)99L#7+7Y0?jKRh~>$Jeyst>D}^R{=h1Z-4@og~YkvM2q`hZuF$o=qYN7mPBa
zA52Z&zLj(FMmg=5Tsjh34D-yHD)gI#YZh8G1l;_l;@$W3(1z;63`>#=6Hkg%U3Ou3
zSI)R*ebYai`|PXsvQ_L8{nDm5!)PU+jntu+jB!_Qh%INkv{Lj=-JJjdpAQu`Ut4nb
zh0I|8xLKrSrkc*pcw6BT9nSf8FACP|`#-r!Gyl!1IjW~GMfm9bn(MmLTu^7)b6dHC
zTK`|)%)EX5*|lDGTiZjq=b2{wn|OTFs>%Dl2+!s!6!KWkb?d;TM=!!H*R8y=>)yK4
zxe7D%+?KeAEIB%JL*ga9^LcibJPyB>GF*&JUhG%E$hYU^-emLrI~0mfK4n<TxJ}G{
z^GcumOyOnrb%8~42TG=2fBW8E->y9&e@Wo=nvWengx9}aF=Nrv{!-nroIZ}r($?C`
zH`w0Pt~jCHz1#c$<l4n|6eeCTxH8fG|2h6odH-w@kE|BHv^ci@FZZgmg5vw<)e7e<
z$ys&6>WOT9mh;Qh&WmoF!+!rP@=vk8TUEA1XuGoQm5C8q-`OkTKK^NaYkyNldvD*o
zdj&~PCcpO5FJG+`wdUIg|Nbj}>+9{!A35u+Ik;c0@%O<Og40)+sk>A*hyL96BP)Np
z(VmxYB!j2_UbdgzdfLrv*M7};*WYgAcKpk<kHQ*%H1}lW{wV&S&QQ;ONA04|nIA8{
zyK6gqI`m&g=%#CL(%-y4GYckvniZy+nH&@|Bj_5-DXwhoy1#!WbT9iJ+HfMLLSuqn
zbKJhm?&?}qGapQpJF;Hie%*yUkFVms&U3rjx|~#;{w6pZR@`;(U{MLk+8A+keY{P>
zlNEjQ+qu<LxQkNyWdyVj`BWV}TvzV5wxTdIj(vgMjM^D|dU@XxuPl%d6i;P(a9^#c
zOZwt_p`(}M`#fv+B&E&TQR22rVvg9_hL@sSbsufkRX@6Wb+=WHSca^8sruu%zbOY3
z4j!4olbY!9^kEa*!4Bq^UzaTH*)>1v{{AlqTm)qkMgB)jQVtS2B+CC`p+}mxL2Z8C
z+8@7vA9qQOZMP5X%F=!O`);{=*FVj-J0AEJ|Kh$iXOmomdG4|F<a<*@p31E3`Z32-
zA=dKcpZ4=?@li`w-BXfdkN@?0|GYIhT88xuFH>I}y(jzeYviBH{F{#)xZ<JOsP*&q
z-KYDq+dDS9sdqkBpMAtxzkaa+=l6wKKX?ja+^Zy(TbPGGaeumk?TmHIi(2_f%DrD%
zE<0L3tk_{%VyJlMMu*2KmoxpB?{;6FxM0)Dx$$@Z?%IBS|IP*XudTj+PVxMfyqgPW
zot$^-=jHIN^H!<N&R)05Zok@b$B#xDifgtkZ+Pa|nW)zwQ@o{-f6o2&hsFEP*>~J}
zeZKBnr`fK$fIl1mK6p|P9zC1;Rd7{S&!uDkzEocR=%%VP&FbE+#1md?1NUw6{a$@H
zwl-+J{_*(pugWuyUc4pPaZC1??;`hb@&2h(dJgVTU!L5v@5bYjsatP;y_OZd#9eFY
z<*UmBc0*QEI-FoSxY6PJfiK_e7&+K2^p4zjo$_(P28pht3qFJ@{oTmxa9Zo=wbpZ;
zSDj|x;5`2!lH2j55|?1lD)FxsSIwkda=QJtxH8;*!2N!y#G#3cjwOlnrpbJ?GrOb}
zm9<u>YJ;yyc;wOl+soxX<c6mIx0l+MbGll+Pq^QrtjtNs!^w%wdWw^VKv<?`?*A<p
zTwgMH&EA)~r8fH4z60Kk)25k;oYG?9P;BAxaC-FSDbMjf+4p<fQ?mcee|{=_-z3%O
zNlz7jmwyU;XZ3x~^P2ZlYu*>n`)pk)QoGLQx}BuIV${Vas*#R<<+Gl6PWJTs&iTZn
zf7w!znG4UnJ6}3&jzxI>1i2HF4mV59Ii0p#q5g~Gt3K&;H+h}BH(hh3#4nxrIqycC
zse9RydQI!B|IMc-_f1+H_WQ=QPfy*otv6dVwJkc{f9m4nH$AmqGyWRcnVWdrt$e=c
z#$Ls!tuO5>dy8*2p8oyYTc)@yuxp=d-#NwO>Bqhm>wa9lS?PDxLJ`)=<mczZeQwmQ
zS@rjecKfU8p9DPTZTRYadGV(mJ7#t!2JIGg?OrwQ-Ge_?&RXxIC%)78cxcz1{r~s8
z;wsBH-jX%zdi=yqPcLcyvZ~Xps0#l&Ri^XfmGr5nPt5$sXI&7U>L#ara&GsHkGk21
z@|T^M-2C*dj-^W8T-n*D`A^@wrD1(Ze94RRX-8SF&s%BivF7z{-;0SQvH3FuYhScU
z8oK^cnEG;l`|Hj3t!`Fm`j@!!ePUgIciPU$ub*wHi@wTqb7Gz6s()8<i<3S}t+@Pj
zzWXb&P17<Q=PDixzP)FsTFm!^_p*A2V;5};xnw$#Z~mH>9n<E%ELrdV@!U*L#pmzm
zh}V}~T(aSpv1!|h<K-bSelm~BMYo?_lJWU&c&h8w3oOgz=dO6?ack23O%3OFzMu5N
z|E1jfu*?2OOD!i}dcm1rS!_Ay*Y4vvmTVvEA1ijNX2!+mwm6Dt&X2k%dv9ibhj8&B
zEx(rR-Ea2mYOOYHQwleJQX{8-Qadwj=D%rQo@DmBEv|@pCy=s)A<A=BQ+J%H{lb3Z
z+MjbQ3>E!eB@4dMP#4sg9#&wv(CYS!;%ojg8b!BK6>Y6=Hl8k${4;Z^oaUkvA2rp>
zCVg5mLFVY6H}&UE?{<v|d2hMdSV`60QvM2Srpis;xaXhuIZE9;btLz(|G6gxA0Pj>
z){Hv2@oeUboQT~uO*`J7?wcI@>f}V*@~kN{CQEAimi@eTj%#V>Blqnola%ki>z>ee
z)cxg>bIl&L6MpP3&z0GeDPLap=i|RfCDnb`U!DoPb$Pi<$NG5dNSiNzJRg}A$gw-`
zy)naQa&%$ISN?sIG%Hk^=lqaMa(_QxjzjqVpTB~$%-a6Xx;1I}+u1Y4ZWgaUmCiNM
zv9HSSjnDHbc8Wd|zHQucZn4e|;}h~Lo80o=SYK6`cCKr(%qtttR|#`p-n-Ww_%Lc^
zv(mhM!W#u0o96y0;!@+BwR*+l&i&$N-|qeW^l=2=q=v3%Vw+ZR=-Bc{U-kSFbLaOS
z+Y^gyEp4qG!;hZdrSjao(OpW*oo|=5y|v<!O&bncbk6;7Zn4Juy44w}soH1nK1qEv
zvFEc|+dS)_i<Z6CUt{OK>t4HYVN~_`D+{~&eu&yCt3CZx>GEjiEx(uUb&<b5c<d>-
zt9?l+tKx{degD47J%#uFc27R1>CR{Ox3bi4^TlQ9{)-d2Q>tqd_m>n^*!D|3O_*LU
zvu<|%`$a8DJNG<{_?1+g?4KZf+g<oP+v#&9?@|^BPLEvo=H2_K-$C<MTkW~~{n4#0
z{^9?ho>8e&uTtDz_ViZdMe#KC*U{~dX7z+f?@cwD^HbBK^~<gw-$glD92i&}5^MxM
z^c%;z#_1`#%6w+lwJZyHcxcf&VHxQ);n%xF`%Sr~t_;m6l~p->$#-GJ+$RF&UDxL1
z6|R`1w6p&DZ!;@}uV1qYy{CjtnQ7vb%pts5etE}SzqJOX!C(9;b{GUa;d0--`bdpd
zmXxmg1*5h&i4#W{x#dIHifYcUeO0;QyeF4td8mYa!oSDYzL=l9nJuq>^w<6Ob8;e1
z{CwG{{$BpIQQ67obw9Rd$}ju>?+5drfM3^{OFyTi|1yrha&AF;(#v0ee|2@rpX6U{
zA2qF}>TCS@JBRopclP|7X1`OX^8eZYT66p=L;mmG{VJ(wS^X2e{oBhnZQ-oHXLqk+
z;uD{K{+0iZz7pA<nD*Z$X1XI|#hbUACLQ7K&)Tqjd0E(*Wxu|Dxt3m88vE>axB6>U
zU8}xznprQp{(b-0mb&Ed++#+1HE*_GU-BmU*Yj(UGKzoq@6MjKQSHy;+uGBE8&|y$
z@a6V=E%Wry-mMeQFZ|?h>3wC*{+dVZ=4W~|C(9qOynks+N39B{ac8_txAd+@mGbIc
z`}meUoLE!eb<=ZUt9)JVtF*{N&v$zZan*frJgItRX23?Z^n|*5al-2^$A9OW{&bE}
za_P4hzdwI9W4C(#^Vp#u#-}eUow|OucWb@=4&(MYiP7(?zaM&AbC-Wg*^zr*$K~Tp
zADVa`l1NSe<vd;Wxn=hG>ci2Q$@Sl77|qvzy=ptl-)XY7C#M&)Tcy9_EsynmxX=DZ
zgz?j@d-&_(l%B7z&H2V8boIx!J#Sx5`ZDXj|1aNrhW?-XQ}!u8pXF-)ebL0py^HrB
z-Q9Q1<9CMmbIZ64{*K}^F41pVrp|q-di7uM?<99CcfQTdMrukCtJ<~Gbzk?IXr^$w
zF0xkdvv{uY`twxTCgs<MPdJ{Qxa<3b$uoqlf1AHsDdf8(w)X2$&mfWFUo3k?bpO4&
z7WDj)O8UP2IX}+vzAVd}__+1#q`#IcA0}yK=503qea|#A=t;eAe{s26RLk9o=gevx
zrM}9V@O-*-b7N4-cabb{_59n%121n<p8NJj*@`f&=DFYQ7#=z~*=F0EegE~XFGZi&
zSpCas^^)D^zblm&#JDZyuv6V_QF(cvKi@TuYtHHIX6kZh3l^R-b8sokDR%Fj6LI^_
z{LJs2Rn>P-+R6t0)4D(Tz4N=3Zic2u%j|WRO;opOa`gMW;&Pkp%Q-uQ8Bd;PS<_UV
zoiR@_(BD7W#A=FLGuL(D$X`3tR_<sI_p|X4wuw7p|MGt2`XAaqOFieWn=2YDBrbJj
z@Af68)8;+h^nLoa^nbs%^^`xkSQEcn<Dbaf%^lo7PP(VYYvsSY=V)Z;TYn>VYX`G$
z&DP0Q?~32oEY6s<`*ZC^Q{`>tMdx|jUx}%{-YWZCnm17Ul9Kk##{Ca7oUOLTR+a5f
z_WNU5Ui!P*{`~RHQ`F+_x(cr^I)0>d{ge3rXI(klgOZ!9@14)~QeOT&HAdZ2eA~a2
zqb)W&>*8;GF8=UprQq?Ol6vcJO(^_&`sekF#drQK4L<bx^z7YE_fBWe@oBW*y!o;7
zswwZSi{>`R{z@wLbLz15czDtOl>XnprYm<xYb@w=-y9_1*QY)E*Y0ckraWK8m;XJ`
z-RU>8_oQwA6T`1cy;Irs{_dpJ-)ho-Svwc{x~=NHXC!&<&zGFq6Ir2m&7SbJKg<z%
z#l$GUz$j4QeZc-)N&BNv%_m#b*tcb=S#H{|XS-pwsmR|ImTs?q9=deW>UpcvdTrrb
zlh-Dt#;m@f$;0RRW&6R&^TJ$?a>VjlUfil8nmkACj7aG^&eKaO`<`E18T_KxeoCA`
z(@8ci@g<z4Z5r>Md88h1x&LmB$@MK|?Xs(1c|7DOGP8AAZDdm!>a}il+8rC~wE34N
zo~zs@x<%#GC9x+ywrgzqmEB)7u5FUtr1CYX`tvfL-(I&bS;+Gq&zT_ma*wR+D#N|S
z>CdW9$J<o*uQ`17_g&R(sd+^i4-KWnDi&<9oo!eBBqQm0am~idJA3|Zv)9d;d+K=Q
z3vGJ=t<Uc?^A`I_DW<8mik<8$`DtM}@AL~+pIbID*Tjw=kL3T<_VOaTh0&ZHs`vk7
zJ$(8m=ef^!J)awgf8Dvn8Taq_Vy?J-nzge(XWoyU74l+F|Mp^`(y29XHJX~;@^4&9
zZ(Yk-q;loD*fH6OvM+0<s0XXg{a$Mu$sY5g$oEg_`yEDlnl?B7#papMdiguu`;Eo(
z{yn<-*HhfXISU(i?+j5r>t?O|bI0zb=6j~5ekt$Q{M`Gr^wmzADIb0WZl9aBc<ozR
zoz;$hpIC2RoWwkN?u1RvR$l8xFP}Jhx^t;sN^ePc?hm1p6Hau>-v8$%GxN{u{qqHX
zpT8kf-+#~4cg+%xD!<j4i!O)TZTntp5%TlN>YqvKCogU|X7Re=|C4PxyM=#qUaGlY
zdnoO(T9L`Vf2Ug~HBVd`5@UJ!Z|Y&^*BSdh$VqmZmrwBye34^fy}rJ%yzfCE!<zOw
zVc(jchkhRos(!a1qsp>S<zIXCOKsnmPkkL9P29bS)!nH$YJ1kCy*d83rn?mx7FE4@
zdb#O6vtN$5_r8KpD*aOmC;Z=N{Qu!}KhMAyI(t8Dw$Ayty|u*u)VY%UTi-t|;|{v=
z`}qI8A)9*U-rnOsb5gR?o7w-myry%n$$ZmWG+*HM>n+p!i<b+>CqFOwRMY+7<n&+O
z^X_!txo5rJXKGjEt0z9wt&jgu@R#d(lFlS{T6j_P9MM(ljGaV3Y1L|OUfH%!{Py{)
zr;bi+i9fEpIr;nEx;2SAO%_dfa^Hi=&n_o^n?#jr(7FYJlTErlnONRj{<+5K_4d#^
z3h7JgbZ42nhaWBef8w`IY}{noSm((L#5d@)tuC(FI90uNNoVM4uNONf+u#4aASKrS
zru=5(*uEk^uItxy=OuA-6^4Efc=P+|tInO<|ExCXmHcFEfAiUeyh$tHRHb}9;{WB#
zJ;m$qwSULQ&MRtIcRhq}Nn-k!-1XwmPNbeXeL^7j{=BrG4-}^DyHWG}&y7d_ZwB8z
z^X;8(=KP(1mKqzU3YSDpy|`k=arb-o?{D2#@A)V)|McY3vTr6a2W@*|o<4uYyE9L|
z&-)sEI(_|qqo*zV*LUilVY|I8u-4AiBcSI<+NO{D-M7v)-~%^*4&<}s<S(hy_S5Q^
zdp#@a@g`>tr8(2TgymeFJbA9tQB#x16-ymwG|y@^Yq`IPH|eSV$--4;UJq-tO!pc_
zSG9S3{1D7L*WdWC`p)Nq3=CFcptArjr}TgMwo=)B&%7_^UoPfebYY)HSHtp-_<2)K
zNO`AT<ZjfOH7#m3bL^*!7oS~Na$fULv;C<{kws58UKKXqJ?l>U=f7VzEqQ$E&z`so
z2|ZK3%W7o%TZG-rET47yccmwP<+mDN+jBQ>dM4J|n8ujf@|0IS@_AlVzeC6J`@HSl
z%Y<~lPPW$8d*-?Dh377R@fQ8EqM!4Ym@SShUbISi?q@yk*VbRQt~<PJqu<15QLhfn
zSf?{N?p|(U&X&(k#nWT+f3M6A=Bt}>-=<&rR`~^&)ivi@IC9_mhySuOJTpo2Vfx&y
zccyOpVC6n<zT(~wzxXEiEs0iH8!Q!>@PD1C>9@G|>(BJ)7TevQog2Td)Be3|di3gB
zyWYqxsHmNAT70SAX4k3l*;nT08?=?r*8LN$xKwYliI(HNFBO&o`8yXhzB9DE9}yR&
zd+pM!n#<aa*N*M2{5r{a?%RDSZasHr?lZ7<o3!e|-@lgkEUc<Z`<rLat$nlLTlg<y
zUnWiKOFPemZ<}&s?n~p#Z|&XruGY?p*0sD;x!h*qvqduhE~|Qc-h1|)$Mf8Ie?FgS
zHS<hsP0cyvoGF$!YfH$?KdFa}il!B3y?imv<o;7DjR$i~cAxJ~70Eq!?pvfvu=4Vm
zD#eFa?mC!IzT~r+ce%}%=actnse)P&?}{D;vbKMky~I~9d>ZRVmHwEv_1=9_@&0pF
zYwj1^s#_9v=|p;aU5Ullo8MLPZ>!ILo;OqAZgg32^!m@YW9#z2U4FmLb~n@JT6zB`
zr<6tJTxO3C;d|lp{YhNP-`aig7O(2-!c0m<R=yRU_v!m=;p}(wb>H4Ee4it~%~oN-
zrY&6wq1={NqBXbYI$fT}XTpC@&Vaj=&&w*rWXFMz?p}q@ip_m)o)296Bq!$NIq##2
zj+5ifVogn7J(*-u+g$X&y4?4!&ti?UUwY=8RW|)8;rk)(Z+ED0xr|P>zgCukpGu@k
zyVXfft(#885}w|vm7lD9<MtQd&1UsHYZC6Av}0S`yw@If@4DCe$4=h#^<S^5`G(80
zPA@#?;qB(*Vcf4WWq$E=zs?utT66EIKG_!GytTI3euuBkr%6v`Q=^}zZg86sajDhq
z(5Y09c8diMFX#JJe|?=V8e6>Um_el6U9WF5gBLfi)|9H&wqEqK=*d$3H<k*|Hs?Rj
z`El&_(QuiyIp>$m_pdHsy_q&^o<von+c7H^*~+5#C(L-ae7Tk;_xDG=x!=7XkNLgV
zE8bk-qBlRTX^qI4kblQbS4Mlr|9}0(bYoV?C!K?4+;?KL<4;Skh&^}wu3`1ZWn0#U
z|M~icW%0>Hx1%l?_0D~HFK%+<yMqt278lQUe{HtYuJ5RA;gXtnw=QNWzJatiKA1D^
zZa#mXRr^$Ox>9Y(rNf=RS*?H5zlo=E+iu)4;f%z}<(Hf1pE{M_yDh*>+jiFqkJA^l
zN}hPBN1qjGTJf+wxN6D5|4ZFv<2L?&T=9H^T3y1`%&TryE7qUyxP0}`uTN9{%A9?4
zj-xj!H-E>R<t`6nmUcR<pZMKc-}i~fwC|?UE1VmJTo>&wi8%GsCUR4nD)06^2L!K(
zFxq|2S^JP%>6N&(?aKL|_tb4kG}<rm!RlqZ_=ljG8A@R<m$u|Cx%q0jM!)8D&8j2k
zI^@6P&6l3-{mEkWQ=9ucRSNq*pPp26X7bWm%kGp;T|YT)Qr)(tB?sNpXFa#tFJtvw
zKX;Sm<x;7s6V>0A*u6XZ&EvlHxowl0@4T};({tHWyMJ?<Ew62C&7Muc(>@2veapHl
zeXr;s^WEg4eTKWYL`7R$`<y>tqkVq=yq$>*$NE08=Pf$uJ|$v()U`L?RJx{?uD5&@
zefHtiDfe!NT4wZStXwef{~UE|-Sws(eGb=n`U5SZr#{aqn3R?HHh)^jLlIZ0$2LEg
z8)>TA3uPLe+2(iL-tgasM$>zhhkX62{mmv^>3KY#p|4MGO~&*mmN(e^bo`awJGHGv
zmNa+$ugSds^VGeAo;C;HL|dOeYCS*nhUw#-75<+z&V0!Hs?cjOq4d>GhIuSgmPYKj
z`=;dC6Yt+kgn$3+SGx1Y^P1oG_YzOKZn;0tobH)8MOxZsuXy=o+i#*>b?GM;cV{$3
z{_EQu;#bsf_t$PxlcN+9la%S#fDCJw=8H;?LcXX@p11RF*u`o2Df`1OUt6ji$$7hY
z&+JL^5wk)L$WQuv;bL*RjU}t@7nxh9F78|Bw~$Z#HGBN|TTXm0*G7MuJ*nuC^V`b&
znX<9}9*dq-U)*HM*Y`Ai|D0LB1z0zzw6ROaPCfa%Lgm@d`Dbf8)r<NsSkBX*|9I_M
znU8hf7M|GuMnALvq32ccWi6(DeVm4Wo;u{tveEkZEb?OQyyB<7Dm*WlPZjC?FXP%B
zEp*6J?y{!r|A#&Mw@96d(!2h0LdElQE2E`)pS?Y2V<FFeyXM>fW&2MHZ@&~{T9|uR
z?Cq<bj$8lcxz;A0FWtE1#I)+=xsMpQCOz&{{OMWil@f4#k>@Y9@Wsxyd&_scX`G~Q
z7C%=sw)!c*lKBOV#~$-8r$<(~+<lU9ySC=j!N~K0SNcL%8lN}YvHM!hndzT?*yK;L
zPJ3fv|M~Gv{cUA4w+eQ>Z#BQ3vY>jxTbVwys>8`DpM%U#yDLZ9zhVP-3MA^7%8oz&
z5T3h!;-77+JU(@|d%ug;{%iTFaO&pubDl)~sOJ+{u=feC<g=5L-h}On_<8Ws5#3Kd
zFZKAU_jE5h@@C@IV)d3?y<syaN@U4%XBTr{?6$r9$@$m2Nk?KXy_)CrPng$rdHj=8
zGv9U3oj2=SVBO`AbEhXdE-`)gLF}yX?pbS9t~+<C%P(!eEfrVp_e5p;9V_c)0Xz2}
z&3MQAelfGjn``ghgg%pg^Gh~gBu=kC=mDSDt0x&3*|T{>Smhjr?D#&Z%HG}f@U+ZR
zqZ?JVJJmk%K3-q#ev@sM!!@xm59@T5mv1)d?LN2t*uCFnAJ=~BR`I{)b?a+xRPU)6
zC*g^jr$3~CvR~!jyU9iKHtte6KIdkV(cTHamLIs{n^60%>66*XiKm6@Ha(uM8Y`x?
zXy@x$vEK!2{?0n7RK4kKcD&~#<@L2uR!^qyoVaND&6(%=YQk4eHaR-=<iSY4sGIz2
z&IZVpMO+WOu~0qJ!0d|2&TqDbGv_#+VEcD^&XNmX*X0+Uo&WjwCA)bYpQlaLRO(Kt
zx$*GSCIubaPw(HRKKh!QYGQEKEb_TS<&L`g<=f_YCOkd;+WR-#CeKRViba0)YN?en
zitpxFXw6ZtDE0droBVvwsoe1IkKR27wGcjUyuxjLS^uG>VAUmY_190Y9*DifbMea=
zui{I+i_{`Q?3X;4Y9%Nh=j1<O`lZLE`6)Y2Whqt5%3Qh^bxug}om<*2t?s$iHsM{X
zlh=JK{Q2pQ<EO%Dr@Z;XzyD->^mpE^LkGo+D&I}JcjJ_!^ttn!W}h=jExh?h=Cof#
z`PUP_ciL6ntvvZ{*XEhBCr>)oR%)DnV6$$~g{hTuuJ27;Z!`TVcjmb{ZBH`~C2nan
z;$L4~{oZ2!`>fm1vwzMjRP~-ABX{Nbze%r)OV>V4-Zw>k{_B&Q?9*y`?DXD#`!efG
zIsd6|yrt4ltK2?c($c9k^q1LhFe|quSAl^^fr06tx`EW3mv^+vrhLnoZN0bjhWnMB
zWnMcRk`(t_?>|s-Immq0ocDS;Q_Flicpg367F6|o-I@M7whE^!D<s#1mR_#OIF#_}
z>e4W|?`412{VZJS|9;h~Jq~?m{m*Yb6FE(+@YGJ8_%|hbhp%qyob$B(l(o^%8;?S2
z=V!>)1#~MN*LJVbpD8nMe`4&h=j%=t-V48yz)`Ee{$<kBx-P#sr^$X#cXoXVeD*o*
zZ*#D(=O*R3rMB^V3O|)uM!ong9~1Fqy)y4~&C}jI{F9vYBK4=V|NX?g{?iWg*ZX3G
z>+f4{+I#2a$Ad4IS?)P2{PvPT-0#}IFRpD8edCetd+zvpub&PpKhOX2W8!=5c81eW
zcScv8o_$*Kg<Ns?^@yAKfst|ke{$pIzYkN`+CMkfCQ~uHtXM_c`kT_>9OdIbXRIzM
z^y}GC8roITp{`N;XMVct?#arxuU0nSRd?6fEPP|Ki~csgo2OpK@LMFQud@kUdYDts
zY|mCjm&Db@A-iJs*fg%b<0&9?C)?aTXSeshE&ZpH+uxPv`G|+>`b=>XEBO08#@e;$
z-2d~@twB?MI8@%bX)@#Qn-k$`@wb-S+`qcZMo?4!@kzhGGnc7+Y;cTTwMe7-`H?rL
zU;nbYIrnqruD7q^&S>We_L>~j>pzi@DgJKG!$9>pI~Kpu@U@Y7(Yb5R85M3v+jKwk
zGaY6<CYP@2bY^%iNl&>}8nlmn%jJFd6cv@Tp5ItCXTBrf^imH;<u7w2F4T7)zc$q|
z+aciti_YCNm!<r2%3chN0t}4}$_r{b_$(gBO8fYgoSdkarI}_u^;6F;vt>tHZp`ho
z=Dsadkumr6wy*DcuVh-?*y-ae_;Tq9!>ij$*-J|z_tgFV_w3M*JIC$T?wB}t&+e|M
zx0BzWSo%9@d5xaF=i$4l6?b=6`U*#0pYwUm+?(N#TE8D}%ZlE)cZs>&`P=iB%+J+}
zKXl3P&5ga~fq%}47iF){dB3g2_Lt|+L(}uTZogi|Q*O5Fl=l1Ii?*1~ysqhW@z1Z^
zne#uWZ1`jte%dqe;J4R*Ki*vSWc``GwFOhpSI*bIU9(fox$ymlwfb*XpIwt@?0Nn9
zsWm4Sds)t1)TY;YvGnG}_Njd(&&_t-T9<3gu5CSI&(pK1RYm<fS5@o0S?t$0x6@zd
zzUYhG#P*-z>u2ehHY-O=o%kYtLP|@qmnX+UPJ`t)pT@@3?9Xu6oBrIW=I*q)XV2bw
z{L*n%uuJLw{mX+sxXf$zy?1LtkJU8Q^f|c;CaPT0yB3n(zG+SDIqSuT-%i?bZaHWC
z?c<aC{(bCa`k3G4vj6w1lb2f7#r&A%J?Y4P53Tm4={6@XA6Kw{|6yzG-<b=0eD+VW
z5;m)OG+$6uWx3)t9luTCXBV5cS?$~<{`23@rhS*M>-tTTdBwe=xyOBOfmOw}i+0lw
z1l%^8^7OUoEVp@+)?Lm?T>kZiM_8#)=heXXQ%vX9{IUJa{dGo~=<b)@d#g(%thTMu
z$bH3SI!pJ9h1Qwv=R2)!w?#ewf2}6)j#TmCwx;#f#!ZvdN^>o~K6n%UC;#k)X%_Qt
zz0#kwYC2!(bvvWlz3ZRaMpu4x)=j%RWwr9>m}v`xUcXz;caO#Y)b@zAKO5!NJo2=j
z$PVsuFVJrce4p_y`%^>pKevZj%TF$>YgpFx)As4Pg%i~K|0(cAd=g0AQPQ=q?_R#8
zM1c3KuFp1w%O-DIBI4>NBzbcF#Fk@SJsc-zAGy`C{qOD%QImvT^4fmB@b0&my2!uy
zJm1{V)tZje9tX^O)jhYy&eg_iVSdW$)8ePT``=WFQ!0IRP5-IOt))3rO8tAUhl-m?
zs0TiGl(Mg|;-4S-?CIB9@mq^CFJ0WZr>->Y$wc*a^GfD?Ig>v5jJJ`izrIe6fy`Nz
zw;p!qG8LqAcgFm`f8(6+gP=E(z3mx>CE2OzRow^IoYc+qf1d2+IsNF$hg;{&E&u;D
z;jYb6<GG7>@Bd!+QuTAtp;i00YAl@b=}Y!({p%?j@8;frzvfJ3Rr%M@9DVos_JvoE
z{M~b|^Sx*D$!j|=zcZQCWa^TsXY;b$<7CI2=idrW{!M*bc&d_P&(pJJ6-E7p`!-I?
z^7*yusN?PX65MaC!ms6BS4v-7a`WeD!98o9uJw^Q_~&)?GH%<KQN_Q$1oYn6ar~)&
z_a~G2*Oo7ByRClw2~&`;v#V48vWmUZFD0LMRZI$Wo*p^L|K!VwOA60jzQ4^rC!yWI
z|Ha8$pSU;YPF|?>x~G#vDAL5JAvm=8?dy}*8){m={9ZOm%_H<#_&WY=wKA%Iuk8Qi
znjF1dh4;|cjXiwUm!CfIl(SsFe|mFnd`kbfU6<C?%&OUZI6)|JN%HxLFTJal1Wc){
zm>bZs<%vOH$jyf}C!T)$>OGm&KjrkxFRfvLLAn2pUl{G*ZoPM>`1d(0rq6j9bDY^c
z)nl<=#of^767Tu9EV-Ov74dlK!ri~^zo^%X#_C?1Ep_vG-ZZZ#_1`WoXtMgh_R7Ka
zw;ZcAPdL=-U9ezp+N)z8{O`n_^c!KR3mg@0*a?|kE1%K)^ZZ2f(#vx^-&L3F)y$aj
zxo`QlU&6VECNF2_u_+F}|7NRU?ib~Z!f?00`p+uGI2}b-Y?<c!?$yCM(2z4j<3`_v
z6!}-}O>136IJ;l%xbd}OUBt)JR#lb~zc#FP+}13=N#DEw#O=a;rFy=Xj2}M>=iXJn
zLFeU%qnf{t9jaRV@2oKIOKE@8vpch*QtrK9=WBcF>RmhicVE<5b;T#Thkc#<y=(uU
z#ZOOKtGlyrwcoh6C^BxV;$iX05q?jzKR^Ed|J~<RuVd!E@17gu=QCql>=o&wt!l}a
z4$DtJH{0AXKKfqH75j>tEj4!YqHmc^oz5L}i2Ej6_{)b=PquHHmp#*+|N1QJ`xf=D
z&xfhx@2#uysf^fgYR}#UrXRm1_g?M~-MV~kb;h~VHUAbzd+R@DFP!(}vw66`)q&sp
z_UjaWcb<JJ?^(uIcRRiKPse0u-mbJ?-k7uHY0=;RTPo&Wndo@k>F)*ir{4ADxu@<;
zTK4XO`?e`-wN{vadckq>qTc+o!f|JPxHKQV-m-s7?XO+exp?E}icb%}eN*<+^gFv;
zZBA;r^4Gcj?EX3DU(V;Jvv|JUvy8cD`0?Z0r(#mR0yXQ_Y4^EttB9CO<wi-RynkDq
zw0}Zs-i^9zX+77orL%AO6jpcNE{d}big%d!`l|lB^)tVnnKw)OhupK8H<KT$F4w$h
zwY&VwPW@HlubZzdQx=uwEI-$;{bic@^yepSkNrFrlx6tx$m-C|+wOnZFFzsU@C^@{
zxpjXpaR#ZZma#p1aqYR6&&yLdBTqfb@myW;>+C}5(Er`LHy`DnXPZ~<b#GhMq`PyB
zmwt29_*VNatZTIui}j_Ok|)o-X-=Olv#E%k@#UQ^wIU5`_4*qJF3E|nvUd5!vUkQ5
z0j;p}{o=R#uiRYyODl8Q<|BN|5C1f~y@P$m+#7T3ix=HHn78fg!RX~)=YFc2%@Mt~
zEx#^o%7b$qlV-)G=TA%u3JN>+QKf$`qw}6exxdeRVeeYn-n1p_q~yCCxzmERJGS4_
z3;yz9?&Tx;|I4CebYD!hPkhI}cD7bpkjwYp{&%m#4|V+CvdN|I)4xBzpK+hN`fc}I
z0a=}Q%PuNfP5S=uBGdC&iw}EuiuX^L;%!|j@!`mJ&5uFzO?=-ai0({I`E)U0?Xuw2
zi#YApUNe8WUfT0!q<Y}{_P80IO#!}sZzuYQO>({x(VHUuf9+FGj(Z}{-u+Fw9(?C!
zzX=P=1DEeIueMrFysCeH&OiOlHy{04dcZYMSgEP%f5Z+8yQuQs<=gtMWj|OQ8`iy5
z<Da?uxgS+_=a|7gJdPZu<nuGvEfr^8+h!~AWzmAX#huftoUglm+*+0MwdIcU)3=f)
zk*O@!SARav(eqsvsqy~zr=AyD^VWQcNKF^>%s%qBrub&~ljKWJ`WDW*)-~;&O6C0J
zr{(qht$u%6xJ$M4t=jC0m6tp3-8#8Eex_*X0l(yL(tEb-e^;dV$|Y^hx}Lc&t+}Ef
zX=pjmDt)u-<-|POxXqKlt~K}RxBT#K<*}nS3!g8ZAF^g?e{<=VG%ZWs`<53CK4!Qs
zH@SHE>*dA#7h9GFwV$%(7Sfv_;;Cu9W4rjaGnN6j>N;d^Gv9m>(AnL)K29Y5%bR<%
z!uFNj=8Ar%U$ZWCUw01Q?z)o9Py62QDOcQjJBjz#mIbMH^;0C(+*r1ry#4p%W1D+%
zlMmjr-raRO?X&wf|F}=J-#>Ic@!z-G<Ye9aQ^!5-J^gW?|LWQ-Rm=JBYXdg<-ShYE
zyYPOpnEyrdixciv?c09-s<e9jXZxe&1!`RZ`Cq1OU;aHpUYV(a-IZJVe)frN*JghC
z#_^vaukUWj&6#=k9kr`J%cyI`SZRmp%oR;Pf5`8o{9UE5*KdCRc&hWWZc=gH*28x3
zW`{I?{jJ|`?p{;2#c!5G>C-<fD<#Uf>gGg0E}WSEUPpZY4r}*Q|Ksn^(7SghQ14t&
zNzsHb)d;3!(W%wz*7*LparDT9f14i8J+FPjQu}X7*Ze#4RG0a8PMGtwc)43=>gHS@
z%ceSI&x<Es9^M=JXrKOa@7b^4pA?iX51;gGt82_wjlixI)4smhn|r@J|BQlqZO4s0
z>!apqWKC34e0nnF=@Gv__wSo$AFwI<w=k~iw%xXN^=D5`de1brJ{_q&$J*rH-PFgw
z<#)+tPxqf7^X|^fPq*{WPg&9xVQxQ9=Jw~!i<Wxk{hamT?=qh?!T%0N8)epsZk;hl
zM5|Qq<B^R;Rt3ufHyGcZedOZD>l2RfEoYOO`r**;qnCDlvd@3#`mQJO_|I3}Irk;?
zo~zXAM^~#=gie!Q@j7biHgA)X8Vw~A^_Nb1B4TfkI_2)1_G`jU=38bi2|uK-yP8dz
z99(aFa@YIiadVp&U5(M{YiD0#T=e7P!ea~74#@JGU#dPAaNStyl)QhgEbpGbnaxqX
zD@y7=xkQ5clL8qmHt#-`-`vn<(5JdoagoizOB$YHXH!MG>o!Mh{l}NM(NFz$(wk{3
z*QCwf`^`3PeMU%@cDCH7PQ@tGt+fwCZcXl=yH}<0p^;a~JD1wm-||!QXUvs}IxtE9
zR8XYZ<4GZZSN;}PpBwV7{{Oo*zTClI(qF_zp7MVheL1P*>H4*&E0U{fR(vs&Te9({
z*8Js{SE%0B%((E>^s&aH<6-`Pro4Vq9JIJe)+Th{jmrP3oX@8%ZJj?M_kpVKoSP@S
zyLL_A6QdY4_2IUgg@3ya{rmcCi>>``ol4=;2S4mQHAQCK+TZ_{cFzB>uU>ZT{FB~Y
zY43h~4b^O~i1oAIt5a!SthB#5?eN51nV<gs%Xwm>Qlj~hYnqzy<nx<!D~&Fvub;ok
zaN?Z(p9FWg>ujFn#QneK@Ra(`c2@Tv&e}BP{QW$?YURND)qKwvZ`*3SLFMD*`RnE+
z^p{ASFpFQ;x&1p|_B7+<ZGnC@^VQrlvVY3$DfK(^F8BVKf9=}tR?q7{&6=W|y=gJu
zhjIryd!>lGe%C)QX4U+)d5?a?>k9u%E=!{=`sd7l_xWAVj0a}<mDR=lCreqsOtN_?
zHuXn!todKN{*6Z_3vD_a=qGcvYRv-IdsC*Jn{4_iW$7Brw|^oZIGsAOP(k19Mq|a@
zNhylfmwtwnX`K!-nJ4n~=f|qTNpp_Pv55NV*Ts^){8#dmgm~v&m)`iKdh(Z_lsC(<
z4)NxzxqW{1G^hRfF?NA>_M07k=aDWsL;PgptxNZQ7k@J>^*FKXw$-oZ;@7J@A7wxJ
zGHrcxXUQCk%G-9rU-xa|y&iAXxaIw3Rdsvs-37Pmc7zqVR8MlZN|gBbd0TBx(7Vjr
zE4ZeuO>8&1?<_1jx%boG&nM#jIHyf~ctn}`#>aDfZ(DB~PEwxxy*Omg#@qicpW40T
zck9;^Klh#Y>z$y|QehwO+^1Fl=IP}<FGC+y`}oQDzSFBL)_<qAQatt7qa}~GeTq<6
z=iU6a@BhNs8BQ;Ms=d&u?d-o${4#FVi^=Y1WS6E2>b}mn>2q8o=uZFIC!3ZSt;~s*
zjS;`@Svr|_l5UFd?PF$LU;E4_%KJJ7g0hIgG>!}1DVcBb{Vev}?pyk8p)N?2fr(kT
z;J=y4hn?GGv@SaGKQnoEXMR4zys&pl^G|K^$~dA_6&T^*o_<l?W#Vavs8`Khdpdb8
z&iE`p>uZer^w7ydqHhAPJeU3a_>{-xKbB=nHY*1t_8->rE_?H*?3`PbX4qTJyzbRm
zulW?y%Z?vi`O8hp=stVF4!`7kMen!xt~tB+X-x5@LsK@m={0ljP3fwut}MPTzVhNh
zzPGh6mz}JZi=A6zVR>io+w!yh+Iw^J&TL-mwmHo5<Kmn(+sih+m41Kj*ODa{Km0uS
zsc*)U4+_m|D*ZOe2T44c!<p#1@Zc-P_$N0`o|(Uh&H2T@OMMYtZDwb^_%bKnK5KSK
zWx@QgEoE>0F0hvxD&NeD+hk|y`O4<@&D?7~J)2zDd<JEv$G7)S{aof7J^kEFb@|6%
zmP%f{di{yZzM>_2j9=9>8`ij{x#(^Uel=l-iSD1B9I^L3J+D7;3BUYYH1G61?ze?k
z4qn)4W4q^GrN#$`ZH0gD#R<p1o4bcS{eSc8PpXCLXWL38=U=wnn;-94owLR7?p`C~
zpii^z>^ITezwhvul{QTddjFS2Zk#G|@!z9;m%hzt<C~o48C9|O%oXW-_wFSZ{cYT(
zQvbTWbH8O-%DY<*xsBDkcWX{6<+r@OE@Fa4YEo|M^Nu^a)WY;OysX%*?>Wn0$Ns}|
z^3{!MW_y3H3h}%>%RedI-e2k0bzgCrQ!mVJ)*ZSz^JR{4y2|9E{{L!%YJUGRDLi}B
zbCPk0Uc-4y*I(EE-+8oa|HXK-f+ZiC&(9UP|J$oA!qEI)9`k3N)o(@g>jLh$H#IB&
z-~azi;m7XvY27y^Cw}9Mk35tjZMir8Mudv#;oIN-))sSGe!6`0h}=(py%x)wJ^Q1I
zUVPrBUTHC3zw_<&0G0Ca^QFPlF1jAKn3yu7cKi0XkA?ql-Rd*x+sRoXdHTmKwPvTr
z?)!emu<-uZmq~6>Ege_BpJEC(l{lg%C3!%ZYe${mPm}Ej-t=hPc02#S^sL*-hvx*}
z?)EpiVVx^ubkQw*``QV5>c{uYdC%ji_Hpll(vKO8;U_Ow2ii?nelqps)SStmDh{lB
zzAioCca@!f_te*$TW`7uH)c%>esJ8?X=$J8hR7{G%i^2wDY7(Y8kGNw*1f{H&dNQ0
z9$V2od9k$5dJFkh30l{koqJ=i`G<s82X}%pnZY@Z3-=H8AB&uPF$Ub>VpL~pE?zeK
z@a~t4wZgW4KBr8->6hAnW?87L`+D|`AwOAuy*N>MFS|YT&9dp!ocCJSL?^Qs`>aU*
z^20&3@>Rw2q8pXFth6V^HQ0E+d9c5A&XH4h-s}mnn(;Jwjpw_p{WURho1D$nw){U)
zo&PpPHuu+!W9x%mk89mh_Nrcz5pl|8-R;S$+Z*kkuKTnp_V3n<*A_n96@2)n+s+zW
z%U{2>Cx13Hp4K~YY4Vd|_Fdm>b2TS_-W|Pbv+vCNx3eFe-Me_r@$02!ZPxsEQ@8zk
zaIcRy_2wP9%A(k(b#dQYz9+mheEy3mF*<dtTwm><8|{q=tJj<Td$+at(a-YkC70*Q
z_V4+;Ir;0CpNFnL+H3cI@?X<O%9E5%&Jb|R;tUH6xad|d_FL?;*vTCmlxNJ{(xxXa
z`9h>D&HjJf#4isyQl~LIGz+^b{p;%CZ_9b#X#9ID%{u9vjsG$!*K^U=wDMn<yqI2+
zq40W#y^AcPjO}@u8!yjBJ%5uCu9kmm`}{d;e$UTR(S0i*C@!kQqaqvZ#aH|E1dIPC
zjo07P{!WX3JAFy&lNAL|cdm(*6I-;9)2TLJe){o=&+q@eY&g;7y5+oECvJawvhLKc
zQ`x+a3|CDSzArBG^DN&dwn?3Zes?Mr%M&c*r89nC+&^!TXSL~~%PV6)RknYg&075M
zoI?5c=H=fP#Vl)UD?EAYW6k`XS^MXd^XRTsa+=b?6}*&#S4%AHbG4F(s7v41fQxrU
zozJfgG-=cJV#&<N>{V0W(bp8_5i;-7iZfScFM3tD?Q(|q>bB;{T}vGVCw&*n3Z4AI
z%>43|-m@isn>o1hmT+J5^j|Zj-R*0s)a;`!FMqm!EU;-iu}Bk~7_YFbS!DF@<YSXl
z9Gu{Btppoki8;o$jjGaD#7|VlI`DsLWieX&U9o=2h9DRI*0i8knbPfn7qeTuLS0Tf
zRn07ry=5NstnYDEymi{-io~50)}QO%T6z4KA-~Dp|18I@+NN!tb~MAc{=5Cri{Xp3
z)qi<Jo#eZ>=aJp3;98%er!(G8UYYLBYI))3<IGRnd1Y#2g3dC}mSHphn%%eDtUz(z
zw*IXbe_oFNq~c%ye7DsxS-HQZRT@k4V&=Zj{Qcw3E4Rg83Pbxo+vUIGd|u3cD^2~U
z!`rMdk3T!K&(Hqd@Z2IHQrXYuQp%kr{MX9AYzftjSub>bUH^jhF>5Z&dfakSNZn*r
zMNv;X7q@SWmd=vmC5q|0c0T_(&27(0ohLrwlZEF#<KA>(rdpTF^;NG=o{ZB!<kEcO
zkN3aCqLMlMR!dUu#HgBxfWqyUz=l;ivrk@q@MHHa_3J0zKd4NzIPvgqtmu{VtAxzg
z&(OG}@nGS;rO`4=8d_ydPVTX}l*e^;?V(9cHoxAiohw?K!nOEt^O0SlnoryUJIgao
zWbC&qG%|2B9H>w?ut@)P?yK6y2e)@B*>u(gzGRGB`_t0(qRXVmx1ODo<1GnY)cqz?
zMC|kDFXE4qFNT@wscn(ueZjtI;i{$4Qn&XdO!^m7|H+(rYV%`-2Ym-Bb{^L`bSuq7
zdVaWVo!ca_qT|UXZ|2XIRGpNRd+YkV8<%g~QstYI`AKE<^~yt<T+`J<zdZ97+uxwT
z%ElPKe*Nl>_XjcxA4Kj<iraJdMd4GQ?b~l99@R4IHrqJqiLm!<Emx+mMlC)e-RsIk
zcb?2FlUeNL(sJqZRwveFS2IL!1+3;!ow+ArZ>WTpj>yM6!O4%#goyDsN`B>&XJ@(r
z9>@>$X{b57wXk3QzI~VPmf2;;MJF$L9zE~Gyr+UUH|i~)`epm{`StaG`;8}++}?7f
zpzw04@Y2IJw?5pzEz4fNewMs`UffJ2S^KJWoaag;Ri-I0Fg>1i;K7`iJ?}1f-+lDS
z-K)j(v!t)Qgj?)1@yTw=?w5R)ddylePvhgm)-!ikX<c+$FSuM~t;g!8GOlmm2>Kpl
zE1dJjRZ5iQ-$uRU=|_`p9=&Ajp<p1wa_q3J+y1h&U*8`sS$>Y?$eCLEU8a+lSbJ<d
za5yCDvg5fg3nhK4KYoe$c=`6*&jQsc7loxGvLrMGIDS;smdkyONYxcE=!ug4sxozR
zgBr)PBb!WrrsgibcI}~Y&R*9yQ|3HSlluI!TI$Tseb>8;9YKrSeS|jzB*ZDWEY;RO
z^wm4^=eGP^C1rDVG^$nB=a(-%q5VoWcBh8=vVP;a>bB31=uI<L{Ske4y4tq}296$g
zrsncx##?j!!?-u@3)2xyZ}ls+w!C8e^19g+nOj;_suw1)@RYVrI<|PKoy_KSn||pR
z-St}Z;*WpoQw|y7zh@?QX!>_(omydCCT;o6UVPey=0~5~Ew=~?TCpUWs9IcdzkF3<
z_O_m{)jo@Z_6mN!a`BJD&MG(as)z61OjlL7BP)NVE_wRmXIs@%3tRea92d_tIp?d=
zYbcaya>GFA?YCuq@$1){mn!J{$7IcBK5+XzbN0i{X;#bq`L8mmax^f2P)`)ole`R%
z6ZYH6uSnhKPt@7}ZTIz0to~X5dQF3mOh4Iq?qwG%=Op8~s!zODSs!__#K6#4H>e<e
zyUlxFftMT4o$Q(N>G||+oP9CwOyF^r4X?X?Gqbv0eiM6NW!=_1fv?9D1eaC2t=N^u
z)oa!~=a$IQyqT)HtqOPDL+99MO*s`1nRU|fzE|F-&mz3_lfKC+dQOi@-1h9^Cf^yo
zY#XP4QF-X`UAjnwg-OA{isQ)tXGb17bWiws@4wu0&uO-7H*&XGzOu19*xgg@vH0h*
z_$O6`Z34M$lTB30gng~DluTRmidOWyRQ=w}Wq8S~$7T24m6xx-e(fuJLc3-Ccctu4
z@&7;A1+j1On*6;<m80Rn54C`RY>sN{^@XRlRK#BP-JGO-+?;ny-4?+M5uwo``4=ND
zFnG)O?)x|Oa+A8?^CQA{_0RN)31?2p`1XWDP&{)sL+$;QY$s=~Q+)o@)g&wDdg1Mc
z>yeTZl~iu12~Mt-3TkI8`<bZ6(QrUO=!1XCuDcFJEArJUMK$kUQqh|)>El<z?7Cz-
z_v&5Eb6#E*(UkK$9dSy?NnSTM_{lCUkFJFD|6L+`&z%##7P~yYzR%|QkFre9z~^C}
z%yUZ{EhkU>cKrg2kIlUmEozC2Is?w8w?ti6D6<Q5`|Wn@w4<mpxbO=Q<}h)$`uyNz
z-xJRMwpsn^Pk&p@)O7#ZQzFG{)??D#`0D%DUwk2Yebz!BzQ&n+bJ=P=txZ>eq4A@~
zgX6Qkr@z(9n{8{Tw4!a!OX*|E%v&m_2yN(kmhtxD-NjdS)t!AbF@BTsT;oqEkAG+C
zEs~yoQ$<{OaqhV<dn<b%T??JOWRu=Jx#aR?n;*ye6)ZcgvVjGZYWBM=vwmE2T4&qd
zqnDWbJ~3-w&vxBy=+&|^$n$5b$<@#=+_|2MQdz!!&1!X9y4XNf)^+A&?zwYbI!!RX
z=QZ)E%H2iVwgnxj>@^5{va?P0%e0`4)-Agt|7X8&5}!M}H5bw(L?+ppo}P`h+I)1;
zEE(aX0t<r!ZqsirvWmGk@#M{CzC61P9Li+a8Xu}I5Q`3d9qDKGA!svq<g}(wad$tz
z=tx)@xV$Uc#k+cY#}#(YYrfMjH_7e&a_pMkK_h(+Py>L0$&9J4&*wYC>vLbzHWu0T
z8yj`GmCf6p<te1rG3TXOwAEE9L8qvgG_hr-k*!h5OJl?q@+~Snpi+>%rL6PG3%$gE
z;EH5x%bV8^PkhH%mcj@MB4pRs3t7bGURZxWtnTBzN!#9h)8jr_$tAclc>CG;5@r96
zZq2k7KGe)65?S}VBH7D*yK2w*y@hTj%Pyn}cr0#d*t&M%#=oBflX~LMa9sLwO1JEC
zN1&e#tAP8w_?-rUlj=_TT<mU}rZ>^h@mEwAv&h*#7SM76g$v#ZKjyyF*}JLO_WPSf
z*_*#xCGdi3Y0y?71}KB6?ZCBSUC#LG*MF=Qce;sc2X$W*`WG2^d(P=No5_=^|M&1t
zRGd|}^`hClyi<=ZD>q$zo1N|!uF-W!i=#(_acbyR-Pbpn=GEt(QDHu+eoE8h_UnlB
zyCr4Iig&dviOZ7e6?yNmXwsB#^YbQ5`@s*k^nlxf@7IN0kX->~A7EyhFJj`K<ZiX|
zwhnjDm-P2@<bVHM)zgujHSc{_kne8IcN29}xqoS&O4``Ly!36b<@rONznHd4tzZ84
z<DrcYl^vQk&wcy0)Vgi`v4<-)4VP_y-~BZDNy&so=Y7+nmsol4JyG>CZF8uMP4|hC
znZlq-QQ-iKo05_26Hv_p3loTIm}WS5xfDdFty%Y`KRal<-uaFV?d$BeF5dKqfA^|u
z1(%DwZy9H7ydP_9Z*_Unl^hfIPaj_|bDvv0o&UFE!VS)dqkUKGr<U?n@2y)?plEsY
z|Hh>*61%FV7K(p;v7p#g=c&?izQvI;`Z<E)zTINRmz9ouIVpH)q0Ob>eDg1En@|7T
z2+C)TAAJ~N|3CiuYVli0Iz^BV+8olSyZ*iT^IF8e54((R^G{cud;9-^9`h-EC3gQd
zi+V0U+;l~DPu$gOcdchDzAx~4|HRZoNFO??w$!hEBPWxx(}Aq5Qo6l&Z@VoivJaaz
zO-4;_U$NB5tv9%yW_+Ic(oDV5HFo)m%vYt$iu?9fZnHC-R(>%fM}6ZoD<M!_)X<>Z
zSSUYpU6uEY)$%X@<-6<L0M{wsS$M>zKAyb%_ZkZ(a8d-XZ((#_uxhFGkA2^}!s`kb
z=CZA{$=JboP%8BL*UO9f_mu?Sy7y1&NRaMC#aCsOK^OmhdZ)e7bI+N%0%dk45qG8S
zZr}H`U}JLjIiR#=zPjCfbG2s`(N@}if8M5xcmBC#=E)s<&GCNGf@S=!*L?Ta>+-9;
z^xe6IAJU#bF!8{S@@2+-OVs_ttdkCkfgKO7PN9{mKhx_t{g~oSyNrWx{?glQ`eW|o
zUhnz)^h)b@ocp_E(VFIaIU8C&?-aC(ZrQa-rm|Mtb@%3VR{8PPyh8d$;yUir7&4cZ
zXK#_6-?K*a-`d(Brxg$S=V-m=eYCi?HdY~&*DWySs-nfBxx)TVRST^aOwD_umdptX
znmSd6^j-0Viwuy93lN(lh6yzNkmCO9)=b0d|IfWw?&O{OY(e_Bkn=Yxb`{o5h`O53
z^0PfXYLfMC-AT{$s+>G_S_}QR%bQ$rah2=$Sn;D`yHB_ORWO*t^5WY)+1>jVq*}a|
zS+#4M+#11;fd`Z7!nW+?>I?Y)ReQN>aP;=t`9)7J+;rP*EF1XJ%P4{!w3sB0sbJan
zO$VefLMxl8{nYKp)eI5y_ZNP;m(Y9b)WjF{JM_L9)U!2xe)MnF&!TyzmKRw+?fP7F
zzIEHmJ^5)#FIs0@P?-F=FsJ!Ifw02bl4AGWm1kU*nomCb=llBkljl4(6m(p)V$<QZ
z!m(+&O0nvej{31`jo?O)g;2o^mt!9Vwm~u`GI>n6aYn$}jm!70n$CY^S@q>_Z=Xj!
zD)zp${Ig}I;jT~B;;JIMt^T`&R=Z>tha1W7PSR3|i!nU3Pf%gO#&^DdgU{>CQ4aVU
z5PI^GuKPJt{udt4e@#C-amsX;@a?Wv=__)az=IV6Cs-c%@IP72ejB}*IMB}$mhKiZ
zc}n*Z#*eGlieA6}`P$D^SKhX`gQ@1FTgxM_@aJjmY<m<HI_K%e)9UX^CWr2+TH@H{
z6~4OV>ALPm=U`Fv=>O{<+a9jT-*R=TU(`#@kK5M&y4aL<G4)mW#mp@;`Kv*mPPoBw
zV5VKpia<e(2+3o5xlwWM_t;4rnr$A11=U>GSHJ7$q@(BO|GRemp2hw>M-#5?*!(a0
zzviwrTb5dP_OCzLXI1dH=Y>{zW%<1PvyYM^3l*16Ufj(4?o*;YN5g>+f)f^B+Ij5G
zW!}T#hFPFI-N>NKkY3dgYcPWwlxz{@5CfC5XTsm9n(Fg*)_s~Cyr@}wryN%R@2?Lz
zpO3EoSM+U~-e%V>oA7J@?RT%<yzR+$od>luH}2XrJ>T#4Qk!)#&o#aus!Ha1A3agB
z&aaZ^HE122@`crP?;>YLPCoQjve@7B*7c2$QsjXLgYX2aPX{owZ#7f-nSSlX=d;(#
zt@N)gazFa&m0562;bOz5FJrG-+Q(Zy+`^ObdgnCI{;oaKwu7p%_3KXmXi(Rk+Hb9I
zF0=Cdf@Ya3se7yAnr?G?%{_nEjH96;oki}~XHP-7RiZza=FXdYVe;$s@?zcn-^06K
zu`OPaH)XlshgBRYpyjBIg^mp8z7!VDyyyYSx}XjmtOf94-htT9kG=H0`zw+^eqO#W
z(<HLA^6Fg9kgg{_`b#uE|GuZYZ{zJG-b<@mU;4+-+EW<i<l$bwHAj1c=bnHo)mOhP
znegr3r$-VT4GrH}`oy<Y)C<iG47#jYTVo}u4sIS=C^5{OURbj6VhFn3FZvxk7b_l#
zl%899c5#VRyFvK8f-l>ypR3<_GNJFur@B2lm8Wk-8cx#b+5IW|_O9l0SIYbzK0bX(
z@#CXD;cv~b^06n5sl!Kp?{&6mAp2mV4WNGc2B97G$uGNtP3KPDzf&XeyKVO)ZsjJ^
zW%0SX^TRLabtOA5Q*XKU@sqp!w3woO2Q50Ud)`~R_GSiOU5WQinR{W&yqC9xuRqzj
z?pdvLpV^Yh|8{-ht2#cX;>A*J$3$6hbAf@yK|z7zz)U%T>6~s&$hLzVa6$M&Moa7B
zv;Q}&&f=}z;dIS@x|*!-mF(C1cdHoAFQ0Vhv`bu3*4s@j`zJ(KFU>vu@&92_+2to^
z{u6wEw`iMX_jJ#Pw{{&3{^Gj_+GqmTRE;hSM*58<TNAX=a~h9gqrASq-4w~y+QA`;
zmY34i+C=_-zSdK|*x6LgZrXzD+q6FIh~>S!YS#~i(}(7)c;)f@>(XE)fvM*dCV=uV
zxQ)To$DnkX`Q4<03Ft8<A;on1vXay{dxOWlO;OG_Ca&ts==F+#E$G(Bu6h4%jeo{$
zd%308`>y(%&01@9uIBTa7fk(rPHG6lBvM%(_{gV<vpfUE5^|Uw@DQFcOW@K$`TCn1
zU;0c|`6wj3Ds!Kj^{!*_PZ!A`nc&AXp;FTC)(UP|B!FG+z`$}$l5z4;CBEsUK2Ou)
z=d5xM`DL;DUxpqupn6o93YPURUER42!)6WTfLm8YvrM9!u4pChtM$9{GB8c<3{q3M
zpVQ&l(d3X59Owm%!UNsLFE=Kf`{e%2?(p->s~yijO_TrB%pho?+>(2a@d7BnHD2^S
zP?h-Q)-(Il4}8AMdrx7(G^L4Q&I`Q-3E~_T9tpQEpML+<KzkD;N%I&owy^RTCW!E3
z=mIAQXmKa-15^?G$?gzC_Z@N=GjQ-2GFn`2moVr8Rim(y0OU$kHb)Ip!7}!LrY_v@
zq=A*uHiKc#OJ^R5qwp+=)p`~?MvKeL|D_hOLY<53bR@R3kAwA*r;97VvE9g^nvnUr
zE>v{YO}Pcn4(EUb3ltKz&5)KUOJGKi{5>iR>AR*Mm;lb2$G98k>aSh3c-tp-NGl7L
zMBtt{X3J=C`S@R51_l%tUGQG8bE&a;gnZSaa{g-|)d?>+w}?jFYTqYz1?JFZWrp-$
z4}LrZy9ge362UB+=II&)HUDTvi{J$c4V)gL1{=8W#vlXJI*tP~-*s_)gG4;mfCLFP
zG%?J1$t4lNiW+5bhrk%&EDwCjU0HvFe1YMAgrI{X!?_Yxo;BzN7;3s{XwYUVShjmn
zQ!Pq(!GZ)KxIlp+eHRm37<zjI%>@TU87(fy1|GBn#VX9jSl9sq7qqTzgoK>IG>#Q(
zG>!9r9PZh8MP&;sC<**=cQ8F!pSn;5the!_-vOnN$j#U7_9pEqcK8i)go9dwsFJ&C
ze$PVHIWLo6BtZt&A2>KHt@2uZ(dOv}erbg3!TzaJUl7*G-WqTbns`7agf;K63!Apj
z@^a|4IWx`wE=*qI;9w03!~_%J4N;*t?|pkV%gSYvz!m-;NbFzGZ``xzYwX{&1CC&)
zF$zdDG%P(mA<^*Vp4t@o1tBuv>iK~Bf@@O8=OiCePu~@_BEI29)r0#NG8j1;92x{Z
zJltrLf4Z5M@siaHaEl!7PDbAcN0zlqGw)(FkOPH+!+~U`|MI8QKG<EkGvyBh0|SGn
LtDnm{r-UW|NT^zK
diff --git a/workflow_parallel_frame_prediction/__init__.py b/workflow_parallel_frame_prediction/__init__.py
deleted file mode 100755
index e69de29b..00000000
diff --git a/workflow_parallel_frame_prediction/main.sh b/workflow_parallel_frame_prediction/main.sh
deleted file mode 100644
index 066b202d..00000000
--- a/workflow_parallel_frame_prediction/main.sh
+++ /dev/null
@@ -1,28 +0,0 @@
-
-#!/bin/bash
-EXTRACT_PATH="./DataExtraction/Stager_devel_N_24_Bing.sh"
-PREPROCESS_PATH="./DataPreprocess/Stager_devel_N_24_process_netCDF.sh"
-TRAINING_PATH="./Training/horovodJob.sh"
-POSPROCESS_PATH = "./DataPostprocess/Stager_devel_N_24_evaluation.sh"
-
-echo "============ Parallel Data Extraction ==========\n"
-
-sbatch "$EXTRACT_PATH"
-
-echo "============= Parallel Data Preprocessing =========\n "
-
-
-sbatch "$PREPROCESS_PATH"
-
-
-echo "============= Parallel Training ================\n"
-
-sbatch "$TRAINING_PATH"
-
-
-echo "=============Parallel Postprocessing ===============\n"
-
-sbatch "$POSTPROCESS_PATH"
-
-
-
diff --git a/workflow_parallel_frame_prediction/setup.py b/workflow_parallel_frame_prediction/setup.py
deleted file mode 100755
index 207b869d..00000000
--- a/workflow_parallel_frame_prediction/setup.py
+++ /dev/null
@@ -1,11 +0,0 @@
-
-from setuptools import setup
-
-setup(
- name='Parallel_Workflow_PredNet',
- author="Bing,Gong; Amirpasha Mozaffari, Severin Hussman",
- description="This is the parallel workflow for PredNet",
- copyright= "Copyright 2019, The ESDE project",
- version='1.0.0',
- author_email="b.gong@fz-juelich.de",
-)
--
GitLab