remove normalize advantages

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

                 gamma=self.optimizer.reward_signals[name].gamma,
                 lambd=self.hyperparameters.lambd,
             )
-            local_advantage = np.array(local_value_estimates) - np.array(m_value_estimates)
+            local_advantage = np.array(local_value_estimates) - np.array(
+                m_value_estimates
+            )
-            #agent_buffer_trajectory[f"{name}_returns"].set(local_return)
+            # agent_buffer_trajectory[f"{name}_returns"].set(local_return)
             agent_buffer_trajectory[f"{name}_returns_q"].set(returns_q)
             agent_buffer_trajectory[f"{name}_returns_b"].set(returns_b)
             agent_buffer_trajectory[f"{name}_advantage"].set(local_advantage)
             int(self.hyperparameters.batch_size / self.policy.sequence_length), 1
         )
         # Normalize advantages
-        advantages = np.array(self.update_buffer["advantages"].get_batch())
-        self.update_buffer["advantages"].set(
-            list((advantages - advantages.mean()) / (advantages.std() + 1e-10))
-        )
+        # advantages = np.array(self.update_buffer["advantages"].get_batch())
+        # self.update_buffer["advantages"].set(
+        #    list((advantages - advantages.mean()) / (advantages.std() + 1e-10))
+        # )
         num_epoch = self.hyperparameters.num_epoch
         batch_update_stats = defaultdict(list)
         for _ in range(num_epoch):
         discounted_r[t] = running_add
     return discounted_r
 
+
-        returns[t] = gamma * lambd * returns[t+1] +  (1 - lambd) * (r[t] + gamma * next_value_estimates[t]) 
+        returns[t] = gamma * lambd * returns[t + 1] + (1 - lambd) * (
+            r[t] + gamma * next_value_estimates[t]
+        )
+
 
 def get_gae(rewards, value_estimates, value_next=0.0, gamma=0.99, lambd=0.95):
     """
     """
     rewards = np.array(rewards)
     returns_q = lambd_return(rewards, next_value_estimates, gamma=gamma, lambd=lambd)
-    returns_b = lambd_return(rewards, next_marginalized_value_estimates, gamma=gamma, lambd=lambd)
+    returns_b = lambd_return(
+        rewards, next_marginalized_value_estimates, gamma=gamma, lambd=lambd
+    )
     return returns_q, returns_b