db9d883aecb6cdfba6c6bbc76b83d85397fef28d,maml_rl/utils/torch_utils.py,,weighted_normalize,#Any#Any#Any#Any#,29

Before Change

    return distribution

def weighted_normalize(tensor, dim=None, weights=None, epsilon=1e-8):
    if weights is None:
        weights = torch.ones_like(tensor)
    mean = weighted_mean(tensor, dim=dim, weights=weights)
    centered = tensor * weights - mean
    std = torch.sqrt(weighted_mean(centered ** 2, dim=dim, weights=weights))
    return centered / (std + epsilon)

After Change

    mean = weighted_mean(tensor, dim=dim, weights=weights)
    out = tensor * weights - mean
    std = torch.sqrt(weighted_mean(out ** 2, dim=dim, weights=weights))
    out.div_(std + epsilon)
    return out

def detach_distribution(pi):
    if isinstance(pi, Categorical):

Instances