starling.data.distributions.DiagonalGaussianDistribution

class DiagonalGaussianDistribution

Bases: object

Methods

__init__	Diagonal Gaussian distribution that can be used for sampling from given certain parameters.
kl
mode	Return the mode of the distribution.
nll
sample	Sample from the parameterized distribution.

__init__(parameters: Tensor, deterministic: bool = False)

Diagonal Gaussian distribution that can be used for sampling from given certain parameters.

Parameters:

• parameters (torch.Tensor) – Parameters of the distribution.

• deterministic (bool, optional) – Whether to sample deterministically, by default False

sample() → Tensor

Sample from the parameterized distribution.

Returns:

The sampled value.

Return type:

torch.Tensor

kl(other=None)

nll(sample, dims=[1, 2, 3])

mode() → Tensor

Return the mode of the distribution.

Returns:

Mode of the distribution.

Return type:

torch.Tensor