change VAE Keras interface and update VAE TF arguments

Author: Mohsen Naghipourfar

scgen/models/_vae.py

@@ -32,19 +32,15 @@ class VAEArith:
     """
 
     def __init__(self, x_dimension, z_dimension=100, **kwargs):
-        tf.reset_default_graph()
         self.x_dim = x_dimension
         self.z_dim = z_dimension
         self.learning_rate = kwargs.get("learning_rate", 0.001)
         self.dropout_rate = kwargs.get("dropout_rate", 0.2)
         self.model_to_use = kwargs.get("model_path", "./models/scgen")
         self.alpha = kwargs.get("alpha", 0.00005)
         self.is_training = tf.placeholder(tf.bool, name='training_flag')
-        self.global_step = tf.Variable(0, name='global_step', trainable=False, dtype=tf.int32)
         self.x = tf.placeholder(tf.float32, shape=[None, self.x_dim], name="data")
         self.z = tf.placeholder(tf.float32, shape=[None, self.z_dim], name="latent")
-        self.time_step = tf.placeholder(tf.int32)
-        self.size = tf.placeholder(tf.int32)
         self.init_w = tf.contrib.layers.xavier_initializer()
         self._create_network()
         self._loss_function()
@@ -119,7 +115,8 @@ def _sample_z(self):
             # Returns
                 The computed Tensor of samples with shape [size, z_dim].
         """
-        eps = tf.random_normal(shape=[self.size, self.z_dim])
+        batch_size = tf.shape(self.mu)[0]
+        eps = tf.random_normal(shape=[batch_size, self.z_dim])
         return self.mu + tf.exp(self.log_var / 2) * eps
 
     def _create_network(self):
@@ -174,7 +171,7 @@ def to_latent(self, data):
                 latent: numpy nd-array
                     Returns array containing latent space encoding of 'data'
         """
-        latent = self.sess.run(self.z_mean, feed_dict={self.x: data, self.size: data.shape[0], self.is_training: False})
+        latent = self.sess.run(self.z_mean, feed_dict={self.x: data, self.is_training: False})
         return latent
 
     def _avg_vector(self, data):
@@ -429,8 +426,6 @@ def train(self, train_data, use_validation=False, valid_data=None, n_epochs=25,
         """
         if initial_run:
             log.info("----Training----")
-            assign_step_zero = tf.assign(self.global_step, 0)
-            _init_step = self.sess.run(assign_step_zero)
         if not initial_run:
             self.saver.restore(self.sess, self.model_to_use)
         if use_validation and valid_data is None:
@@ -442,9 +437,6 @@ def train(self, train_data, use_validation=False, valid_data=None, n_epochs=25,
         min_delta = threshold
         patience_cnt = 0
         for it in range(n_epochs):
-            increment_global_step_op = tf.assign(self.global_step, self.global_step + 1)
-            _step = self.sess.run(increment_global_step_op)
-            current_step = self.sess.run(self.global_step)
             train_loss = 0.0
             for lower in range(0, train_data.shape[0], batch_size):
                 upper = min(lower + batch_size, train_data.shape[0])
@@ -454,8 +446,7 @@ def train(self, train_data, use_validation=False, valid_data=None, n_epochs=25,
                     x_mb = train_data[lower:upper, :].X
                 if upper - lower > 1:
                     _, current_loss_train = self.sess.run([self.solver, self.vae_loss],
-                                                          feed_dict={self.x: x_mb, self.time_step: current_step,
-                                                                     self.size: len(x_mb), self.is_training: True})
+                                                          feed_dict={self.x: x_mb, self.is_training: True})
                     train_loss += current_loss_train
             if use_validation:
                 valid_loss = 0
@@ -466,8 +457,7 @@ def train(self, train_data, use_validation=False, valid_data=None, n_epochs=25,
                     else:
                         x_mb = valid_data[lower:upper, :].X
                     current_loss_valid = self.sess.run(self.vae_loss,
-                                                       feed_dict={self.x: x_mb, self.time_step: current_step,
-                                                                  self.size: len(x_mb), self.is_training: False})
+                                                       feed_dict={self.x: x_mb, self.is_training: False})
                     valid_loss += current_loss_valid
                 loss_hist.append(valid_loss / valid_data.shape[0])
                 if it > 0 and loss_hist[it - 1] - loss_hist[it] > min_delta:

scgen/models/_vae_keras.py

@@ -46,7 +46,6 @@ class VAEArithKeras:
     """
 
     def __init__(self, x_dimension, z_dimension=100, **kwargs):
-        tf.reset_default_graph()
         self.x_dim = x_dimension
         self.z_dim = z_dimension
         self.learning_rate = kwargs.get("learning_rate", 0.001)