no encoder for single task

config/ppo_transfer/3DBall.yaml

       learning_rate_schedule: linear
       model_schedule: linear
       load_model: false
-      encoder_layers: 2
+      encoder_layers: 1
       action_layers: 1
       policy_layers: 1
       forward_layers: 1
       with_prior: false
       predict_return: true
       use_bisim: false
-      separate_value_train: true
+      separate_value_train: false
       in_batch_alter: true
     network_settings:
       normalize: true

config/ppo_transfer/3DBallHard.yaml

       learning_rate_schedule: linear
       model_schedule: linear
       load_model: false
-      encoder_layers: 2
+      encoder_layers: 1
       action_layers: 1
       policy_layers: 1
       forward_layers: 1
       with_prior: false
       predict_return: true
       use_bisim: false
-      separate_value_train: true
+      separate_value_train: false
       in_batch_alter: true
     network_settings:
       normalize: true

config/ppo_transfer/3DBallHardTransfer.yaml

       num_epoch: 3
       learning_rate_schedule: linear
       model_schedule: linear
-      encoder_layers: 2
+      encoder_layers: 1
       action_layers: 1
       policy_layers: 1
       forward_layers: 1
       with_prior: false
       predict_return: true
       use_bisim: false
-      separate_value_train: true
+      separate_value_train: false
       train_model: false
       load_model: true
       train_action: false

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

         """
         combined_input = tf.concat([encoded_state, encoded_action], axis=1)
         hidden = combined_input
-        # if 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)
-        hidden = combined_input
-        # if self.transfer:
-        #    hidden = tf.stop_gradient(hidden)
         for i in range(forward_layers):
             hidden = tf.layers.dense(
                 hidden,