snapshot...

modules/deeplearning/cloudheight.py

@@ -17,6 +17,7 @@ CACHE_GFS = True
 DISK_CACHE = True
 
 PROC_BATCH_SIZE = 40
+PROC_BATCH_BUFFER_SIZE = 50000
 NumLabels = 1
 BATCH_SIZE = 384
 NUM_EPOCHS = 200
@@ -422,7 +423,7 @@ class CloudHeightNN:
         dataset = tf.data.Dataset.from_tensor_slices(time_keys)
         dataset = dataset.batch(PROC_BATCH_SIZE)
         dataset = dataset.map(self.data_function, num_parallel_calls=8)
-        dataset = dataset.shuffle(PROC_BATCH_SIZE*BATCH_SIZE)
+        dataset = dataset.shuffle(PROC_BATCH_BUFFER_SIZE)
         dataset = dataset.prefetch(buffer_size=1)
         self.train_dataset = dataset