torchbayesian.bnn.utils

Short description.

Functionals

Description.

enable_mc_dropout

Enables only the dropout layers of a module, even in eval mode.

Bayesian Conversion Backend

Short description.

register_reparametrization	Replaces a tensor (parameter or buffer) in a module by registering a stochastic reparametrization module in its place.
Reparametrization	This class wraps the variational posterior module and handles safety checks for the replacement of the parameter or buffer by the variational posterior and its forward call.

Short description.

get_posterior	Creates an instance of a 'VariationalPosterior' subclass to replace a given tensor.
get_prior	Creates an instance of a 'Prior' subclass.

Posterior and Prior Customization

Description.

PosteriorFactory	This class is a base class for object factories.
PriorFactory	This class is a base class for object factories.