diff --git a/agent/PPO_agent.py b/agent/PPO_agent.py
index 41adf1e..70c00b1 100644
--- a/agent/PPO_agent.py
+++ b/agent/PPO_agent.py
@@ -87,7 +87,7 @@ class PPOAgent(BaseAgent):
                 obj = ratio * sampled_advantages
                 obj_clipped = ratio.clamp(1.0 - self.config.ppo_ratio_clip,
                                           1.0 + self.config.ppo_ratio_clip) * sampled_advantages
-                policy_loss = -torch.min(obj, obj_clipped).mean(0) + config.entropy_weight * entropy_loss.mean(0)
+                policy_loss = -torch.min(obj, obj_clipped).mean(0) + config.entropy_weight * entropy_loss
 
                 value_loss = 0.5 * (sampled_returns - values).pow(2).mean()
 
diff --git a/network/base_network.py b/network/base_network.py
index 60ba6dd..01211ad 100644
--- a/network/base_network.py
+++ b/network/base_network.py
@@ -199,7 +199,7 @@ class DiscreteActorCriticWrapper:
         if action is None:
             action = dist.sample()
         log_prob = dist.log_prob(action).unsqueeze(1)
-        return action, log_prob, entropy_loss, value
+        return action, log_prob, entropy_loss.mean(0), value
 
     def variable(self, x, dtype=torch.FloatTensor):
         return self.network.variable(x, dtype)