gluonts.torch.modules.loss module#

class gluonts.torch.modules.loss.CRPS[source]#

Bases: gluonts.torch.modules.loss.DistributionLoss

class gluonts.torch.modules.loss.DistributionLoss[source]#

Bases: pydantic.main.BaseModel

A torch.nn.Module extensions that computes loss values by comparing a Distribution (prediction) to a Tensor (ground-truth).

class gluonts.torch.modules.loss.EnergyScore[source]#

Bases: gluonts.torch.modules.loss.DistributionLoss

class gluonts.torch.modules.loss.NegativeLogLikelihood(*, beta: float = 0.0)[source]#

Bases: gluonts.torch.modules.loss.DistributionLoss

Compute the negative log likelihood loss.

Parameters

beta (float in range (0, 1)) – beta parameter from the paper: “On the Pitfalls of Heteroscedastic Uncertainty Estimation with Probabilistic Neural Networks” by Seitzer et al. 2022 https://openreview.net/forum?id=aPOpXlnV1T

beta: float#