ec849adaf4ceb42ed52ca142c839f627c34b9434,slm_lab/agent/algorithm/reinforce.py,Reinforce,calc_advantage,#Reinforce#Any#,158

Before Change

                rewards.insert(0, big_r)
            rewards = torch.Tensor(rewards)
            logger.debug3(f"Rewards: {rewards}")
            rewards = (rewards - rewards.mean()) / (rewards.std() + np.finfo(np.float32).eps)
            logger.debug3(f"Normalized rewards: {rewards}")
            advantage.append(rewards)
        advantage = torch.cat(advantage)

After Change

            T = len(epi_rewards)
            returns = np.empty(T, "float32")
            for t in reversed(range(T)):
                big_r = epi_rewards[t] + self.gamma * big_r
                returns[t] = big_r
            logger.debug3(f"Rewards: {returns}")
            returns = (returns - returns.mean()) / (returns.std() + 1e-08)

Instances