update some small changes in the data and model

video_prediction_tools/env_setup/wrapper_container.sh

@@ -7,6 +7,8 @@ EXE_DIR="$(basename "$ENV_SETUP_DIR")"
 VENV_DIR=$WORKING_DIR/virtual_envs/$1
 shift                     # replaces $1 by $2, so that $@ does not include the name of the virtual environment anymore
 
+echo "VENV_DIR in wrapper is ${VENV_DIR}"
+
 # sanity checks
 if [[ "${EXE_DIR}" = "HPC_scripts"   ]] || [[ "${EXE_DIR}" = "no_HPC_scripts" ]];
 then

video_prediction_tools/hparams/era5/savp/model_hparams_template.json

 {
-    "batch_size": 32,
+   "shuffle_on_val":true,
+    "shift":1,
+    "batch_size": 4,
+    "sequence_length":24,
     "lr": 0.0002,
     "beta1": 0.5,
     "beta2": 0.999,

video_prediction_tools/model_modules/video_prediction/datasets/dataset.py

-__author__ = "Bing Gong"
+__author__ = "Simon Grasse"
 __date__ = "2022-03-17"
-__email__ = "b.gong@fz-juelich.de"
+__email__ = "b.gong @fz-juelich.de"
 
 import json
 import os

video_prediction_tools/model_modules/video_prediction/models/base_model.py

@@ -101,6 +101,8 @@ class BaseVideoPredictionModel(object):
             context_frames=-1,
             sequence_length=-1,
             repeat=1,
+            shift = 1,
+            shuffle_on_val=True,
             opt_var="0"
         )
         return hparams
@@ -118,7 +120,11 @@ class BaseVideoPredictionModel(object):
         return parsed_hparams
 
     def build_graph(self, inputs):
-        self.inputs = inputs
+        print("inputs", inputs)        
+        input_reshape = tf.reshape(inputs, [self.hparams.batch_size,self.hparams.sequence_length,tf.shape(inputs['images'])[2],tf.shape(inputs['images'])[3]])
+        print(tf.shape(input_reshape))
+        self.inputs["images"] = input_reshape
+        
 
     def metrics_fn(self, inputs, outputs):
         metrics = OrderedDict()