mixture_uce_loss

probly.train.evidential.torch.mixture_uce_loss(alpha: Tensor, mixture_weights: Tensor, y: Tensor, reduction: str = 'sum') → Tensor

Compute the LOP-GPN mixture uncertainty cross-entropy loss.

Parameters:

• alpha – Feature-level Dirichlet concentration parameters with shape (N, C).

• mixture_weights – Dense mixture weights with shape (B, N).

• y – Ground-truth labels for the mixed nodes with shape (B,).

• reduction – Reduction to apply, either "mean", "sum", or "none".

Returns:

Mixture uncertainty cross-entropy loss.

Raises:

ValueError – If reduction is unsupported.