eb110220d9a39a294479433cefc274e42506737e,main.py,,main,#,75

Before Change

            returns[step] = returns[step + 1] * \
                args.gamma * masks[step] + rewards[step]

        value_loss = (values[:-1] - Variable(returns[:-1])).pow().mean()

        advantages = returns[:-1] - values[:-1].data
        action_loss = -(Variable(advantages) * action_log_probs).mean()

After Change

            episode_rewards *= masks[step].cpu()

        returns[-1] = actor_critic(Variable(states[-1], volatile=True))[].data

        for step in reversed(range(args.num_steps)):
            returns[step] = returns[step + 1] * \

Instances