update

algorithms/a2c.py

@@ -55,7 +55,7 @@ class PolicyNetwork(nn.Module):
         loss = 0
         for log_prob, value, Gt in zip(log_probs, state_values, returns):
             advantage = Gt - value.item()
-            policy_loss = (-log_prob * advantage)
+            policy_loss = -log_prob * advantage
 
             Gt = torch.unsqueeze(Gt, 0)
             value_loss = F.smooth_l1_loss(value, Gt)
@@ -65,8 +65,7 @@ class PolicyNetwork(nn.Module):
         self.optimizer.zero_grad()
         loss.backward()
         self.optimizer.step()
-        return loss
-        # self.scheduler.step()
+        return loss.item()
 
     def get_action(self, state):
         action_probs, state_value = self.predict(state)