remove bidirectional RNN since it violates causality. Add resilience for when batch_normalization is not defined in the conf

Julian Kates-Harbeck · Julian Kates-Harbeck · commit 735e1c26a2e9 · 2018-03-06T19:28:06.000-05:00
diff --git a/plasma/models/builder.py b/plasma/models/builder.py
@@ -72,12 +72,13 @@ def get_0D_1D_indices(self):
     def build_model(self,predict,custom_batch_size=None):
         conf = self.conf
         model_conf = conf['model']
-        use_bidirectional = model_conf['use_bidirectional']
         rnn_size = model_conf['rnn_size']
         rnn_type = model_conf['rnn_type']
         regularization = model_conf['regularization']
         dense_regularization = model_conf['dense_regularization']
-        use_batch_norm = model_conf['use_batch_norm']
+        use_batch_norm = False
+        if 'use_batch_norm' in model_conf:
+            use_batch_norm = model_conf['use_batch_norm']
 
         dropout_prob = model_conf['dropout_prob']
         length = model_conf['length']
@@ -187,12 +188,6 @@ def slicer_output_shape(input_shape,indices):
         x_input = Input(batch_shape = batch_input_shape)
         x_in = TimeDistributed(pre_rnn_model) (x_input)
 
-        if use_bidirectional:
-            for _ in range(model_conf['rnn_layers']):
-                x_in = Bidirectional(rnn_model(rnn_size, return_sequences=return_sequences,
-                 stateful=stateful,kernel_regularizer=l2(regularization),recurrent_regularizer=l2(regularization),
-                 bias_regularizer=l2(regularization),dropout=dropout_prob,recurrent_dropout=dropout_prob)) (x_in)
-                x_in = Dropout(dropout_prob) (x_in)
         else:
             for _ in range(model_conf['rnn_layers']):
                 x_in = rnn_model(rnn_size, return_sequences=return_sequences,#batch_input_shape=batch_input_shape,
