Skip to content
Snippets Groups Projects
Commit 02055510 authored by b.gong's avatar b.gong
Browse files

Remove unnecessary default pamaters

parent b8d801e4
No related branches found
No related tags found
No related merge requests found
Show whitespace changes
Inline Side-by-side
Showing
with 592 additions and 62 deletions
Zam347_scripts/DataPreprocess.sh
+ 2
− 2
View file @ 02055510
......@@ -2,7 +2,7 @@
source_dir=/home/$USER/extractedData
destination_dir=/home/$USER/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/hickle
destination_dir=/home/$USER/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/hickle
declare -a years=("2017")
for year in "${years[@]}";
......@@ -11,7 +11,7 @@ for year in "${years[@]}";
echo "source_dir ${source_dir}/${year}"
mpirun -np 2 python ../../workflow_parallel_frame_prediction/DataPreprocess/mpi_stager_v2_process_netCDF.py \
--source_dir ${source_dir}/${year}/ \
--destination_dir ${destination_dir}/${year}/ --vars T2 MSL gph500 --lat_s 138 --lat_e 202 --lon_s 646 --lon_e 710
--destination_dir ${destination_dir}/${year}/ --vars T2 MSL gph500 --lat_s 74 --lat_e 202 --lon_s 550 --lon_e 710
done
python ../../workflow_parallel_frame_prediction/DataPreprocess/mpi_split_data_multi_years.py --destination_dir ${destination_dir}
......
Zam347_scripts/DataPreprocess_to_tf.sh
+ 1
− 1
View file @ 02055510
#!/bin/bash -x
python ../video_prediction/datasets/era5_dataset_v2.py /home/${USER}/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/hickle/splits/ /home/${USER}/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/ -vars T2 MSL gph500 -height 64 -width 64 -seq_length 20
python ../video_prediction/datasets/era5_dataset_v2.py /home/${USER}/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/hickle/splits/ /home/${USER}/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/ -vars T2 MSL gph500 -height 128 -width 160 -seq_length 20
Zam347_scripts/generate_era5-out.out
+ 577
− 13
View file @ 02055510
temporal_dir: /home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/hickle/splits/
loading options from checkpoint /home/b.gong/models/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/vae
temporal_dir: /home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/hickle/splits/
loading options from checkpoint /home/b.gong/models/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/vae
----------------------------------- Options ------------------------------------
batch_size = 2
checkpoint = /home/b.gong/models/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/vae
checkpoint = /home/b.gong/models/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/vae
dataset = era5
dataset_hparams = sequence_length=20
fps = 4
gif_length = None
gpu_mem_frac = 0
input_dir = /home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords
input_dir = /home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords
mode = test
model = vae
model_hparams = None
num_epochs = 1
num_samples = None
num_stochastic_samples = 1
output_gif_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/vae
output_png_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/vae
results_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500
results_gif_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500
results_png_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500
output_gif_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/vae
output_png_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/vae
results_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500
results_gif_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500
results_png_dir = /home/b.gong/results/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500
seed = 7
------------------------------------- End --------------------------------------
datset_class ERA5Dataset_v2
FILENAMES ['/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_0_to_1.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_10_to_11.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_12_to_13.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_2_to_3.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_4_to_5.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_6_to_7.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_8_to_9.tfrecords']
files ['/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_0_to_1.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_10_to_11.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_12_to_13.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_2_to_3.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_4_to_5.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_6_to_7.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_8_to_9.tfrecords']
FILENAMES ['/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_0_to_1.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_10_to_11.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_12_to_13.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_2_to_3.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_4_to_5.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_6_to_7.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_8_to_9.tfrecords']
files ['/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_0_to_1.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_10_to_11.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_12_to_13.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_2_to_3.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_4_to_5.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_6_to_7.tfrecords', '/home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords/test/sequence_8_to_9.tfrecords']
mode test
Parse features {'images/encoded': <tensorflow.python.framework.sparse_tensor.SparseTensor object at 0x7f71880c6b00>, 'channels': <tf.Tensor 'ParseSingleExample/ParseSingleExample:3' shape=() dtype=int64>, 'height': <tf.Tensor 'ParseSingleExample/ParseSingleExample:4' shape=() dtype=int64>, 'sequence_length': <tf.Tensor 'ParseSingleExample/ParseSingleExample:5' shape=() dtype=int64>, 'width': <tf.Tensor 'ParseSingleExample/ParseSingleExample:6' shape=() dtype=int64>}
Image shape 20, 64,64,3
Parse features {'images/encoded': <tensorflow.python.framework.sparse_tensor.SparseTensor object at 0x7f4f87630b70>, 'channels': <tf.Tensor 'ParseSingleExample/ParseSingleExample:3' shape=() dtype=int64>, 'height': <tf.Tensor 'ParseSingleExample/ParseSingleExample:4' shape=() dtype=int64>, 'sequence_length': <tf.Tensor 'ParseSingleExample/ParseSingleExample:5' shape=() dtype=int64>, 'width': <tf.Tensor 'ParseSingleExample/ParseSingleExample:6' shape=() dtype=int64>}
Image shape 20, 128,160,3
DBBUG: INPUT Tensor("strided_slice:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_0_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_0_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_0_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_0_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_1:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_1:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_1_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_1:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_1_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_1_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_1_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_2:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_2:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_2_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_2:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_2_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_2_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_2_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_3:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_3:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_3_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_3:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_3_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_3_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_3_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_4:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_4:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_4_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_4:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_4_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_4_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_4_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_5:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_5:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_5_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_5:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_5_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_5_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_5_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_6:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_6:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_6_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_6:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_6_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_6_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_6_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_7:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_7:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_7_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_7:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_7_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_7_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_7_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_8:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_8:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_8_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_8:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_8_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_8_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_8_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_9:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_9:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_9_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_9:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_9_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_9_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_9_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_10:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_10:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_10_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_10:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_10_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_10_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_10_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_11:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_11:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_11_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_11:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_11_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_11_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_11_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_12:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_12:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_12_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_12:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_12_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_12_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_12_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_13:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_13:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_13_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_13:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_13_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_13_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_13_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_14:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_14:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_14_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_14:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_14_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_14_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_14_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_15:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_15:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_15_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_15:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_15_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_15_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_15_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_16:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_16:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_16_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_16:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_16_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_16_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_16_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_17:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_17:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_17_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_17:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_17_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_17_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_17_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_18:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_18:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_18_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_18:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_18_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_18_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_18_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
DBBUG: INPUT Tensor("strided_slice_19:0", shape=(2, 128, 160, 3), dtype=float32)
conv_layer activation function relu
DEBUG input shape (2, 128, 160, 3)
Encode_1_shape (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 2_shape, (2, 64, 80, 8)
conv_layer activation function relu
DEBUG input shape (2, 64, 80, 8)
Encode 3_shape, (2, 32, 40, 8)
Encode 4_shape, (2, 10240)
conv4_shape [2, 32, 40, 8]
latend variables z Tensor("add_19:0", shape=(2, 16), dtype=float32)
latend variables z2 Tensor("sq_19_deenc_fc1_fc/Relu:0", shape=(2, 10240), dtype=float32)
latend variables z3 Tensor("Reshape_19:0", shape=(2, 32, 40, 8), dtype=float32)
input_channel 8
output_shape Tensor("sq_19_decode_5_trans_conv/stack:0", shape=(4,), dtype=int32)
Decode 5 shape (2, 64, 80, 8)
input_channel 8
output_shape Tensor("sq_19_decode_6_trans_conv/stack:0", shape=(4,), dtype=int32)
input_channel 8
output_shape Tensor("sq_19_decode_8_trans_conv/stack:0", shape=(4,), dtype=int32)
X_hat (2, 128, 160, 3)
X_hat (2, 20, 128, 160, 3)
zlog_sigma_sq_all (2, 20, 16)
creating restore saver from checkpoint /home/b.gong/models/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/vae
temporal_dir: /home/b.gong/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/hickle/splits/
test temporal_pkl file looks like folowing [[cftime.DatetimeGregorian(2017-02-01 00:00:00)]
[cftime.DatetimeGregorian(2017-02-01 01:00:00)]
[cftime.DatetimeGregorian(2017-02-01 02:00:00)]
[cftime.DatetimeGregorian(2017-02-01 03:00:00)]
[cftime.DatetimeGregorian(2017-02-01 04:00:00)]
[cftime.DatetimeGregorian(2017-02-01 05:00:00)]
[cftime.DatetimeGregorian(2017-02-01 06:00:00)]
[cftime.DatetimeGregorian(2017-02-01 07:00:00)]
[cftime.DatetimeGregorian(2017-02-01 08:00:00)]
[cftime.DatetimeGregorian(2017-02-01 09:00:00)]
[cftime.DatetimeGregorian(2017-02-01 10:00:00)]
[cftime.DatetimeGregorian(2017-02-01 11:00:00)]
[cftime.DatetimeGregorian(2017-02-01 12:00:00)]
[cftime.DatetimeGregorian(2017-02-01 13:00:00)]
[cftime.DatetimeGregorian(2017-02-01 14:00:00)]
[cftime.DatetimeGregorian(2017-02-01 15:00:00)]
[cftime.DatetimeGregorian(2017-02-01 16:00:00)]
[cftime.DatetimeGregorian(2017-02-01 17:00:00)]
[cftime.DatetimeGregorian(2017-02-01 18:00:00)]
[cftime.DatetimeGregorian(2017-02-01 19:00:00)]
[cftime.DatetimeGregorian(2017-02-01 20:00:00)]
[cftime.DatetimeGregorian(2017-02-01 21:00:00)]
[cftime.DatetimeGregorian(2017-02-01 22:00:00)]
[cftime.DatetimeGregorian(2017-02-01 23:00:00)]
[cftime.DatetimeGregorian(2017-02-02 00:00:00)]
[cftime.DatetimeGregorian(2017-02-02 01:00:00)]
[cftime.DatetimeGregorian(2017-02-02 02:00:00)]
[cftime.DatetimeGregorian(2017-02-02 03:00:00)]
[cftime.DatetimeGregorian(2017-02-02 04:00:00)]
[cftime.DatetimeGregorian(2017-02-02 05:00:00)]
[cftime.DatetimeGregorian(2017-02-02 06:00:00)]
[cftime.DatetimeGregorian(2017-02-02 07:00:00)]
[cftime.DatetimeGregorian(2017-02-02 08:00:00)]
[cftime.DatetimeGregorian(2017-02-02 09:00:00)]]
Sample id 0
Sample id 2
Sample id 4
Sample id 6
Sample id 8
Sample id 10
Sample id 12
Sample id 14
Sample id 16
Sample id 18
Sample id 20
Sample id 22
Sample id 24
Sample id 26
timestamp: <class 'numpy.ndarray'>
persistent ts [cftime.DatetimeGregorian(2017-02-01 02:00:00)]
persistent index in test set: 2
persistent_X.shape (20, 128, 160, 3)
name _Stochastic_id_0_Time_20170202-020000
timestamp: <class 'numpy.ndarray'>
persistent ts [cftime.DatetimeGregorian(2017-02-01 03:00:00)]
persistent index in test set: 3
persistent_X.shape (20, 128, 160, 3)
name _Stochastic_id_0_Time_20170202-030000
Save persistent all
Save generate all
Sample id 28
timestamp: <class 'numpy.ndarray'>
persistent ts [cftime.DatetimeGregorian(2017-02-01 04:00:00)]
persistent index in test set: 4
persistent_X.shape (20, 128, 160, 3)
name _Stochastic_id_0_Time_20170202-040000
timestamp: <class 'numpy.ndarray'>
persistent ts [cftime.DatetimeGregorian(2017-02-01 05:00:00)]
persistent index in test set: 5
persistent_X.shape (20, 128, 160, 3)
name _Stochastic_id_0_Time_20170202-050000
Save persistent all
Save generate all
Sample id 30
timestamp: <class 'numpy.ndarray'>
persistent ts [cftime.DatetimeGregorian(2017-02-01 06:00:00)]
persistent index in test set: 6
persistent_X.shape (20, 128, 160, 3)
name _Stochastic_id_0_Time_20170202-060000
timestamp: <class 'numpy.ndarray'>
persistent ts [cftime.DatetimeGregorian(2017-02-01 07:00:00)]
persistent index in test set: 7
persistent_X.shape (20, 128, 160, 3)
name _Stochastic_id_0_Time_20170202-070000
Save persistent all
Save generate all
Sample id 32
timestamp: <class 'numpy.ndarray'>
persistent ts [cftime.DatetimeGregorian(2017-02-01 08:00:00)]
persistent index in test set: 8
persistent_X.shape (20, 128, 160, 3)
name _Stochastic_id_0_Time_20170202-080000
timestamp: <class 'numpy.ndarray'>
persistent ts [cftime.DatetimeGregorian(2017-02-01 09:00:00)]
persistent index in test set: 9
persistent_X.shape (20, 128, 160, 3)
name _Stochastic_id_0_Time_20170202-090000
Save persistent all
Save generate all
Sample id 34
Zam347_scripts/generate_era5.sh
+ 3
− 3
View file @ 02055510
......@@ -2,9 +2,9 @@
python -u ../scripts/generate_transfer_learning_finetune.py \
--input_dir /home/${USER}/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords \
--dataset_hparams sequence_length=20 --checkpoint /home/${USER}/models/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/vae \
--mode test --results_dir /home/${USER}/results/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500 \
--input_dir /home/${USER}/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords \
--dataset_hparams sequence_length=20 --checkpoint /home/${USER}/models/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/vae \
--mode test --results_dir /home/${USER}/results/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500 \
--batch_size 2 --dataset era5 > generate_era5-out.out
#srun python scripts/train.py --input_dir data/era5 --dataset era5 --model savp --model_hparams_dict hparams/kth/ours_savp/model_hparams.json --output_dir logs/era5/ours_savp
Zam347_scripts/train_vae.sh
+ 1
− 1
View file @ 02055510
......@@ -2,5 +2,5 @@
python ../scripts/train_dummy.py --input_dir /home/${USER}/preprocessedData/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/tfrecords --dataset era5 --model vae --model_hparams_dict ../hparams/kth/ours_savp/model_hparams.json --output_dir /home/${USER}/models/era5-Y2015toY2017M01to12-64x64-74d00N71d00E-T_MSL_gph500/vae
python ../scripts/train_dummy.py --input_dir /home/${USER}/preprocessedData/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/tfrecords --dataset era5 --model vae --model_hparams_dict ../hparams/era5/vae/model_hparams.json --output_dir /home/${USER}/models/era5-Y2015toY2017M01to12-128x160-74d00N71d00E-T_MSL_gph500/vae
#srun python scripts/train.py --input_dir data/era5 --dataset era5 --model savp --model_hparams_dict hparams/kth/ours_savp/model_hparams.json --output_dir logs/era5/ours_savp
env_setup/create_env_zam347.sh
+ 1
− 1
View file @ 02055510
......@@ -22,7 +22,7 @@ pip3 install mpi4py
pip3 install netCDF4
pip3 install numpy
pip3 install h5py
pip3 install tensorflow-gpu==1.14.0
pip3 install tensorflow-gpu==1.13.1
#Copy the hickle package from bing's account
#cp -r /p/project/deepacf/deeprain/bing/hickle ${WORKING_DIR}
......
hparams/era5/vae/model_hparams.json
+ 1
− 1
View file @ 02055510
{
"batch_size": 8,
"lr": 0.0002,
"nz": 32,
"nz": 16,
"max_steps":20
}
......
video_prediction/models/vanilla_vae_model.py
+ 6
− 40
View file @ 02055510
......@@ -21,20 +21,11 @@ class VanillaVAEVideoPredictionModel(BaseVideoPredictionModel):
super(VanillaVAEVideoPredictionModel, self).__init__(mode, hparams_dict, hparams, **kwargs)
self.mode = mode
self.learning_rate = self.hparams.lr
self.nz = self.hparams.nz
self.aggregate_nccl=aggregate_nccl
self.gen_images_enc = None
self.g_losses = None
self.d_losses = None
self.g_loss = None
self.d_loss = None
self.g_vars = None
self.d_vars = None
self.train_op = None
self.summary_op = None
self.image_summary_op = None
self.eval_summary_op = None
self.accum_eval_summary_op = None
self.accum_eval_metrics_reset_op = None
self.recon_loss = None
self.latent_loss = None
self.total_loss = None
......@@ -65,6 +56,7 @@ class VanillaVAEVideoPredictionModel(BaseVideoPredictionModel):
specified during instantiation.
"""
default_hparams = super(VanillaVAEVideoPredictionModel, self).get_default_hparams_dict()
print ("default hparams",default_hparams)
hparams = dict(
batch_size=16,
lr=0.001,
......@@ -73,35 +65,9 @@ class VanillaVAEVideoPredictionModel(BaseVideoPredictionModel):
lr_boundaries=(0,),
max_steps=350000,
nz=10,
beta1=0.9,
beta2=0.999,
context_frames=-1,
sequence_length=-1,
clip_length=10, #Bing: TODO What is the clip_length, original is 10,
l1_weight=0.0,
l2_weight=1.0,
vgg_cdist_weight=0.0,
feature_l2_weight=0.0,
ae_l2_weight=0.0,
state_weight=0.0,
tv_weight=0.0,
image_sn_gan_weight=0.0,
image_sn_vae_gan_weight=0.0,
images_sn_gan_weight=0.0,
images_sn_vae_gan_weight=0.0,
video_sn_gan_weight=0.0,
video_sn_vae_gan_weight=0.0,
gan_feature_l2_weight=0.0,
gan_feature_cdist_weight=0.0,
vae_gan_feature_l2_weight=0.0,
vae_gan_feature_cdist_weight=0.0,
gan_loss_type='LSGAN',
joint_gan_optimization=False,
kl_weight=0.0,
kl_anneal='linear',
kl_anneal_k=-1.0,
kl_anneal_steps=(50000, 100000),
z_l1_weight=0.0,
)
return dict(itertools.chain(default_hparams.items(), hparams.items()))
......@@ -167,7 +133,7 @@ class VanillaVAEVideoPredictionModel(BaseVideoPredictionModel):
@staticmethod
def vae_arc3(x, l_name=0):
def vae_arc3(x,l_name=0,nz=16):
seq_name = "sq_" + str(l_name) + "_"
print("DBBUG: INPUT", x)
conv1 = ld.conv_layer(x, 3, 2, 8, seq_name + "encode_1")
......@@ -184,8 +150,8 @@ class VanillaVAEVideoPredictionModel(BaseVideoPredictionModel):
conv3_shape = conv3.get_shape().as_list()
print("conv4_shape",conv3_shape)
# Todo: to conv3 to
z_mu = ld.fc_layer(conv4, hiddens = 16, idx = seq_name + "enc_fc4_m")
z_log_sigma_sq = ld.fc_layer(conv4, hiddens = 16, idx = seq_name + "enc_fc4_m"'enc_fc4_sigma')
z_mu = ld.fc_layer(conv4, hiddens = nz, idx = seq_name + "enc_fc4_m")
z_log_sigma_sq = ld.fc_layer(conv4, hiddens = nz, idx = seq_name + "enc_fc4_m"'enc_fc4_sigma')
eps = tf.random_normal(shape = tf.shape(z_log_sigma_sq), mean = 0, stddev = 1, dtype = tf.float32)
z = z_mu + tf.sqrt(tf.exp(z_log_sigma_sq)) * eps
print("latend variables z ", z)
......@@ -210,7 +176,7 @@ class VanillaVAEVideoPredictionModel(BaseVideoPredictionModel):
z_log_sigma_sq_all = []
z_mu_all = []
for i in range(20):
q, z_mu, z_log_sigma_sq, z = VanillaVAEVideoPredictionModel.vae_arc3(self.x[:, i, :, :, :], l_name = i)
q, z_mu, z_log_sigma_sq, z = VanillaVAEVideoPredictionModel.vae_arc3(self.x[:, i, :, :, :], l_name=i, nz=self.nz)
X.append(q)
z_log_sigma_sq_all.append(z_log_sigma_sq)
z_mu_all.append(z_mu)
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment