Drafted skeleton of SEQ2SEQ_VAEP (Variational AutoEncoding Predictor) in models.py

source/models.py

@@ -222,6 +222,7 @@ class SEQ_2_SEQ_VAE:
         Model_B3 = BatchNormalization()
         Model_B4 = BatchNormalization()
         Model_B5 = BatchNormalization()
+        Model_B6 = BatchNormalization()
         Model_D0 = DenseTranspose(
             Model_E5, activation="relu", output_dim=self.ENCODING,
         )
@@ -283,6 +284,7 @@ class SEQ_2_SEQ_VAE:
         generator = Model_D4(generator)
         generator = Model_B5(generator)
         generator = Model_D5(generator)
+        generator = Model_B6(generator)
         x_decoded_mean = TimeDistributed(Dense(self.input_shape[2]))(generator)
 
         # end-to-end autoencoder
@@ -302,6 +304,7 @@ class SEQ_2_SEQ_VAE:
         _generator = Model_D4(_generator)
         _generator = Model_B5(_generator)
         _generator = Model_D5(_generator)
+        _generator = Model_B6(_generator)
         _x_decoded_mean = TimeDistributed(Dense(self.input_shape[2]))(_generator)
         generator = Model(g, _x_decoded_mean, name="SEQ_2_SEQ_VGenerator")
 
@@ -395,6 +398,7 @@ class SEQ_2_SEQ_VAEP:
         Model_B3 = BatchNormalization()
         Model_B4 = BatchNormalization()
         Model_B5 = BatchNormalization()
+        Model_B6 = BatchNormalization()
         Model_D0 = DenseTranspose(
             Model_E5, activation="relu", output_dim=self.ENCODING,
         )
@@ -456,6 +460,7 @@ class SEQ_2_SEQ_VAEP:
         generator = Model_D4(generator)
         generator = Model_B5(generator)
         generator = Model_D5(generator)
+        generator = Model_B6(generator)
         x_decoded_mean = TimeDistributed(Dense(self.input_shape[2]))(generator)
 
         # Define and instanciate predictor
@@ -484,6 +489,7 @@ class SEQ_2_SEQ_VAEP:
                 kernel_constraint=UnitNorm(axis=1),
             )
         )(predictor)
+        predictor = BatchNormalization()(predictor)
         x_predicted_mean = TimeDistributed(Dense(self.input_shape[2]))(predictor)
 
         # end-to-end autoencoder
@@ -505,6 +511,7 @@ class SEQ_2_SEQ_VAEP:
         _generator = Model_D4(_generator)
         _generator = Model_B5(_generator)
         _generator = Model_D5(_generator)
+        _generator = Model_B6(_generator)
         _x_decoded_mean = TimeDistributed(Dense(self.input_shape[2]))(_generator)
         generator = Model(g, _x_decoded_mean, name="SEQ_2_SEQ_VGenerator")