Working continuous updates

ml-agents/mlagents/trainers/ppo/models.py

             num_layers = 1
         if brain.vector_action_space_type == "continuous":
             self.create_cc_actor(h_size, num_layers, vis_encode_type)
-            self.entropy = tf.ones_like(tf.reshape(self.entropy, [-1])) * self.entropy
         else:
             self.create_dc_actor_critic(h_size, num_layers, vis_encode_type)
 
 
         self.all_log_probs = tf.identity(all_probs, name="action_probs")
 
-        self.entropy = 0.5 * tf.reduce_mean(
+        single_dim_entropy = 0.5 * tf.reduce_mean(
+        # Make entropy the right shape
+        self.entropy = tf.ones_like(tf.reshape(mu[:, 0], [-1])) * single_dim_entropy
 
         # We keep these tensors the same name, but use new nodes to keep code parallelism with discrete control.
         self.log_probs = tf.reduce_sum(

ml-agents/mlagents/trainers/ppo/optimizer.py

         :param out_dict: Output dictionary mapping names to nodes.
         :return: Dictionary mapping names to input data.
         """
-        print(feed_dict)
         network_out = self.sess.run(list(out_dict.values()), feed_dict=feed_dict)
         run_out = dict(zip(list(out_dict.keys()), network_out))
         return run_out