Elliptical Slice Sampling for Linearly Truncated Multivariate Normal Distributions

This repository implements an improved elliptical slice sampling for truncated multivariate normal distributions described in the paper "A Fast, Robust Elliptical Slice Sampling Implementation for Linearly Truncated Multivariate Normal Distributions".

Updates

• [12/2024] This paper is presented at the BDU workshop at NeurIPS 2024.
• [07/2024] This implementation is also available in BoTorch since v0.11.3 at here.

TODOs

• Add Gibbs sampling.
• Add separation of variables for estimating multivariate normal probabilities.

Run the Code

To run the code, you will need to install the package by running the following command.

pip install -e .

Refer to ./notebooks/ for an example on using elliptical slice sampling for Bayesian probit regression.