LaplaceRepresenter

class probly.method.laplace.torch.LaplaceRepresenter(predictor: BaseLaplace, num_samples: int, pred_type: str = 'glm', sampling_strategy: SamplingStrategy = 'sequential', sample_factory: SampleFactory[CategoricalDistribution, S]=<flextype.singledispatch.Flexdispatch object>, sample_axis: int = -1)

Bases: Sampler[In, CategoricalDistribution, S], Generic[In, S]

Representer over BaseLaplace.predictive_samples (classification only; see laplace-torch docs).

Parameters:

• predictor – A fitted BaseLaplace instance.

• num_samples – Number of posterior samples to draw.

• pred_type – Forwarded to BaseLaplace.predictive_samples; typically "glm" or "nn".

• sampling_strategy – How repeated predictions are computed.

• sample_factory – Factory used to build the returned Sample.

• sample_axis – Axis along which samples are stacked in the output tensor.

Initialize the Laplace representer.

__call__(*args: In.args, **kwargs: In.kwargs) → R

Alias for the represent method.

num_samples

pred_type: str

predict(*args: In.args, **kwargs: In.kwargs) → R

Predict the representation for a given input.

predictor

represent(*args: In.args, **kwargs: In.kwargs) → S

Bulk-sample from the posterior and return a TorchCategoricalDistributionSample.

sample_axis

sample_factory

sampling_strategy

Examples using probly.method.laplace.torch.LaplaceRepresenter

Laplace Approximation on Two Moons

Laplace Approximation on Two Moons

Laplace on MNIST

Laplace on MNIST