GaussianPrior

class torchbayesian.bnn.GaussianPrior(shape: Size | list[int] | tuple[int, ...] | None = None, mu: float | Tensor | None = None, sigma: float | Tensor | None = None, *, dtype: dtype | None = None, device: device | str | int | None = None)

Bases: Prior

This class is a diagonal Gaussian prior distribution used for Bayes by Backprop (BBB).

Parameters

shape_size

The supposed shape of the parameter for which to initialize a Prior.

muOptional[float | Tensor]

The mean of the Gaussian prior distribution. Either a float that will be assigned to each element of the mean matrix or a Tensor whose shape, dtype and device match the mean matrix. Optional. Defaults to 0.

sigmaOptional[float | Tensor]

The standard deviation of the Gaussian prior distribution. Either a float that will be assigned to each element of the std matrix or a Tensor whose shape, dtype and device match the std matrix. Optional. Defaults to 1.

dtype: Optional[_dtype]

The supposed dtype of the parameter for which to initialize a Prior. Optional. Defaults to torch default dtype.

device: Device

The supposed device of the parameter for which to initialize a Prior. Optional. Defaults to torch’s default device.

Attributes

muTensor

The mean of the diagonal Gaussian prior distribution.

sigmaTensor

The standard deviation of the Gaussian prior distribution

property distribution: Distribution

Returns a ‘torch.distributions.Normal’ distribution.

Returns

distributionDistribution

A torch.Distribution.

extra_repr() → str

Returns the extra representation of the prior.

Returns

extra_reprstr

The str extra representation of the prior.