torch_blue.vi.NonBayesianDivergence

class torch_blue.vi.NonBayesianDivergence

Bases: KullbackLeiblerModule

Placeholder Kullback-Leibler divergence for non-Bayesian models.

This module can be used to disable the prior matching term of the AnalyticalKullbackLeiblerLoss. Together with a NonBayesianPredictiveDistribution it yields a non-Bayesian loss.

static forward(*args: torch.Tensor) → torch.Tensor

Return placeholder zero.

At least one argument of type Tensor is expected, since the device of the last one is used to define the device of the output Tensor.

This is not affected by _globals._USE_NORM_CONSTANTS.