axtreme.sampling.independent_sampler

Base class for sampling methods that ignore covariance between different input points.

Todo: TODO - For MultiTask models (e.g GPs with multiple targets that are not indpendant), this covariance should be considered. - Currently ALL covariance is ignored.

Functions

diagonalize_distribution(posterior)

Diagonalize the distribution of the posterior.

Classes

IndependentMCSampler(sample_shape[, seed])

Abstract base class for MCSamplers that independantly apply the same set of base samples at each x point.

axtreme.sampling.independent_sampler.diagonalize_distribution(posterior: GPyTorchPosterior) → GPyTorchPosterior

Diagonalize the distribution of the posterior.

The points in the posterior need to be treated as independent from each other. Therefore we want the covariance matrix to be diagonal.

Parameters:

posterior – The posterior to diagonalize.

Returns:

The diagonalized posterior.