This repo contains the code related to the paper "Pseudo-extended Markov chain Monte Carlo" accepted to NeurIPS 2019.
This folder contains files, each corresponding to the following examples from the paper:
- Mixture of univariate Gaussians
- Mixture of 20 bivariate Gaussians
- Boltzmann machine relaxation
- Sparse logistic regression with horseshoe priors
For each of the folders there is a .py which runs the code. There also a .stan file which contains the STAN code for the model.
For the Boltzmann machine comparisons, please check out the supporting code for the excellent continuously tempered HMC paper. The relaxation parameters were generated using the code produced by @matt-graham
Additionally, code for the RAM, EE and PT algorithms used in Section 4.1 follows from the Tak et al. (2018). A Repelling–Attracting Metropolis Algorithm for Multimodality. Journal of Computational and Graphical Statistics. 27(3), 479-490. Supporting code found here.