1 coral_loss

rri_metrics.coral.coral_loss(
    logits,
    labels,
    *,
    num_classes,
    importance_weights=None,
    reduction='mean',
)

Compute CORAL loss (sum of binary cross-entropies over thresholds).

1.1 Parameters

Name	Type	Description	Default
logits	Tensor	Tensor["... K-1", float32] threshold logits.	required
labels	Tensor	Tensor["...", int64] ordinal labels in [0, K-1].	required
num_classes	int	Number of ordinal classes K.	required
importance_weights	Tensor | None	Optional per-threshold weights Tensor["K-1"].	None
reduction	Literal['mean', 'sum', 'none']	Reduction mode.	'mean'

1.2 Returns

Name	Type	Description
	Tensor	Loss tensor. If reduction=“none”: Tensor["..."].