torchbayesian.bnn#

Short description.

Containers#

BayesianModule

This class is a Torch 'nn.Module' container that reparametrizes in-place the 'nn.Parameter' parameters of any torch model or module with a variational posterior, and allows computation of the KL divergence.

Layers#

Description.

BayesianModule

This class is a Torch 'nn.Module' container that reparametrizes in-place the 'nn.Parameter' parameters of any torch model or module with a variational posterior, and allows computation of the KL divergence.

Bayesian Dropout Layers#

Description.

`BayesianAlphaDropout`	This class is an implementation of 'torch.nn.AlphaDropout' that remains active in eval mode.
`BayesianDropout`	This class is an implementation of 'torch.nn.Dropout' that remains active in eval mode.
`BayesianDropout1d`	This class is an implementation of 'torch.nn.Dropout1d' that remains active in eval mode.
`BayesianDropout2d`	This class is an implementation of 'torch.nn.Dropout2d' that remains active in eval mode.
`BayesianDropout3d`	This class is an implementation of 'torch.nn.Dropout3d' that remains active in eval mode.
`BayesianFeatureAlphaDropout`	This class is an implementation of 'torch.nn.FeatureAlphaDropout' that remains active in eval mode.

Variational Posteriors#

Description.

`VariationalPosterior`	This class serves as a base class for all variational posteriors used for Bayes by Backprop (BBB) variational inference (VI).
`GaussianPosterior`	This class is a diagonal Gaussian variational posterior.
`PretrainedGaussianPosterior`	This class is a diagonal Gaussian variational posterior whose mean parameter is initialized from an existing tensor.

Priors#

Description.

`Prior`	This class serves as a base class for all priors used for Bayes by backprop (BBB) variational inference (VI).
`GaussianPrior`	This class is a diagonal Gaussian prior distribution used for Bayes by Backprop (BBB).