target encoder

config/ppo_transfer/CrawlerStatic.yaml

       encoder_layers: 2
       action_layers: 2
       policy_layers: 2
-      forward_layers: 1
+      forward_layers: 0
       value_layers: 3
       feature_size: 128
       action_feature_size: 128

config/ppo_transfer/OldCrawlerStatic.yaml

       encoder_layers: 2
       action_layers: 2
       policy_layers: 2
-      forward_layers: 1
+      forward_layers: 0
-      reuse_encoder: true
+      reuse_encoder: false
       in_epoch_alter: false
       use_op_buffer: false
       use_var_predict: true

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

                 self.current_action,
                 self.h_size,
                 self.action_feature_size,
-                action_layers
+                action_layers,
-            # if not reuse_encoder:
-            #    self.targ_encoder = tf.stop_gradient(self.targ_encoder)
-            #    self._create_hard_copy()
+            if not reuse_encoder:
+                self.targ_encoder = tf.stop_gradient(self.targ_encoder)
+                self._create_hard_copy()
 
             # if self.inverse_model:
             #     with tf.variable_scope("inverse"):
         load_policy=False,
         load_value=False,
         load_encoder=False,
-        load_action=False
+        load_action=False,
     ):
         load_nets = []
         if load_model:
                 activation=tf.tanh,  # ModelUtils.swish,
                 kernel_initializer=tf.initializers.variance_scaling(1.0),
             )
+        if not reuse_encoder:
+            latent_targ = tf.stop_gradient(latent_targ)
-        # return tf.stop_gradient(latent_targ)
 
     def _create_encoder(
         self,
         return latent
 
     def _create_action_encoder(
-        self,
-        action: tf.Tensor,
-        h_size: int,
-        action_feature_size: int,
-        num_layers: int,
+        self, action: tf.Tensor, h_size: int, action_feature_size: int, num_layers: int
-        
+
-            action, 
-            h_size, 
+            action,
+            h_size,
-            num_layers, 
+            num_layers,
-            reuse=False
+            reuse=False,
         )
 
         with tf.variable_scope("action_enc"):
                 name="latent",
-                activation=tf.tanh,  
+                activation=tf.tanh,
                 kernel_initializer=tf.initializers.variance_scaling(1.0),
             )
         return latent