928f1d0b3ba401ea7614f19f3595e1a44c171205,maml_rl/utils/reinforcement_learning.py,,reinforce_loss,#Any#Any#Any#,29

Before Change

    log_probs = pi.log_prob(episodes.actions.view((-1, *episodes.action_shape)))
    log_probs = log_probs.view_as(episodes.actions)
    if log_probs.dim() > 2:
        log_probs = log_probs.sum(dim=2)

    losses = -weighted_mean(log_probs * episodes.advantages,
                            lengths=episodes.lengths)

    return losses.mean()

After Change

                params=params)

    log_probs = pi.log_prob(episodes.actions.view((-1, *episodes.action_shape)))
    log_probs = log_probs.view(len(episodes), episodes.batch_size)

    losses = -weighted_mean(log_probs * episodes.advantages,
                            lengths=episodes.lengths)

Instances