DistributionNLLLoss#

class torch_uncertainty.losses.DistributionNLLLoss(reduction='mean')[source]#

Negative Log-Likelihood loss using given distributions as inputs.

Parameters:

reduction (Optional[Literal['mean', 'sum']]) – Specifies the reduction to apply to the output. Must be one of 'none', 'mean' or 'sum'. Defaults to "mean".

forward(dist, targets, padding_mask=None)[source]#

Compute the NLL of the targets given predicted distributions.

Parameters:

  • dist (Distribution) – The predicted distributions.

  • targets (Tensor) – The target values.

  • padding_mask (Tensor | None) – The padding mask. Sets the loss to 0 for padded values. Defaults to None.

Return type:

Tensor