backprop enc single task

ml-agents/mlagents/trainers/policy/transfer_policy.py

         """
         combined_input = tf.concat([encoded_state, encoded_action], axis=1)
         hidden = combined_input
-        if not self.transfer:
-            hidden = tf.stop_gradient(hidden)
+        #if not self.transfer:
+        #    hidden = tf.stop_gradient(hidden)
 
         for i in range(forward_layers):
             hidden = tf.layers.dense(
         combined_input = tf.concat([encoded_state, encoded_action], axis=1)
         hidden = combined_input
 
-        if not self.transfer:
-            hidden = tf.stop_gradient(hidden)
+        #if not self.transfer:
+        #    hidden = tf.stop_gradient(hidden)
 
         for i in range(forward_layers):
             hidden = tf.layers.dense(