autocast.losses.ensemble

class EnsembleLoss(reduction='mean')

Bases: Module

Base class for ensemble losses.

Parameters:

reduction (str)

forward(preds, targets)

Compute the loss.

Parameters:

• preds (Float[Tensor, 'batch time spatial *spatial channel ensemble']) – Predictions of shape (B, …, M)

• targets (Float[Tensor, 'batch time spatial *spatial channel']) – Targets of shape (B, …)

Returns:

Scalar loss (or tensor if reduction is ‘none’)

Return type:

Tensor

class CRPSLoss(reduction='mean')

Bases: EnsembleLoss

Continuous Ranked Probability Score (CRPS) Loss.

Parameters:

reduction (str)

class FairCRPSLoss(reduction='mean')

Bases: EnsembleLoss

Fair Continuous Ranked Probability Score (fCRPS) Loss.

Parameters:

reduction (str)

class AlphaFairCRPSLoss(alpha=0.95, reduction='mean')

Bases: EnsembleLoss

Alpha-Fair Continuous Ranked Probability Score (afCRPS) Loss.

Parameters:

• alpha (float)

• reduction (str)

class EnsembleMAELoss(reduction='mean')

Bases: EnsembleLoss

Mean absolute error computed from the ensemble mean forecast.

Parameters:

reduction (str)