BayesLinear

class torch_uncertainty.layers.bayesian.BayesLinear(in_features, out_features, prior_sigma_1=0.1, prior_sigma_2=0.4, prior_pi=1, mu_init=0.0, sigma_init=-7.0, frozen=False, bias=True, device=None, dtype=None)

Bayesian Linear Layer with Mixture of Normals prior and Normal posterior.

Parameters:

• in_features (int) – Number of input features

• out_features (int) – Number of output features

• prior_sigma_1 (float, optional) – Standard deviation of the first prior distribution. Defaults to 0.1.

• prior_sigma_2 (float, optional) – Standard deviation of the second prior distribution. Defaults to 0.1.

• prior_pi (float, optional) – Mixture control variable. Defaults to 0.1.

• mu_init (float, optional) – Initial mean of the posterior distribution. Defaults to 0.0.

• sigma_init (float, optional) – Initial standard deviation of the posterior distribution. Defaults to -7.0.

• frozen (bool, optional) – Whether to freeze the posterior distribution. Defaults to False.

• bias (bool, optional) – Whether to use a bias term. Defaults to True.

• device (optional) – Device to use. Defaults to None.

• dtype (optional) – Data type to use. Defaults to None.

Paper Reference:

Blundell, Charles, et al. “Weight uncertainty in neural networks” ICML 2015.

freeze()

Freeze the layer by setting the frozen attribute to True.

sample()

Sample the bayesian layer’s posterior.

unfreeze()

Unfreeze the layer by setting the frozen attribute to False.