ca907342507c1139696f542de0a3351d7a382eee,reinforcement_learning/actor_critic.py,,finish_episode,#,65

Before Change

    rewards = (rewards - rewards.mean()) / (rewards.std() + np.finfo(np.float32).eps)
    for (log_prob, value), r in zip(saved_actions, rewards):
        reward = r - value.data[0, 0]
        policy_loss -= (log_prob * reward).sum()
        value_loss += F.smooth_l1_loss(value, Variable(torch.Tensor([r])))
    optimizer.zero_grad()
    (policy_loss + value_loss).backward()
    optimizer.step()
    del model.rewards[:]
    del model.saved_actions[:]

After Change

    rewards = (rewards - rewards.mean()) / (rewards.std() + np.finfo(np.float32).eps)
    for (log_prob, value), r in zip(saved_actions, rewards):
        reward = r - value.data[0, 0]
        policy_losses.append(-log_prob * reward)
        value_losses.append(F.smooth_l1_loss(value, Variable(torch.Tensor([r]))))
    optimizer.zero_grad()
    loss = torch.cat(policy_losses).sum() + torch.cat(value_losses).sum()
    loss.backward()

Instances