sRBM-sampler

A simple and fast method for sampling from sparse high-dimensional pairwise Markov networks.

About

sRBM-sampler is an R implementation of the Gibbs sampler introduced in Pensar et al (2020) for sampling from sparse high-dimensional pairwise Markov networks over binary variables. The idea behind the sampler is to transform the Markov network into a corresponding sparse Restricted Boltzmann Machine (sRBM) with edge-specific Gaussian auxiliary varibles, from which samples can be generated efficiently using a block-Gibbs approach. For more details about the sampler, see

Pensar, J., Xu, Y., Puranen, S., Pesonen, M., Kabashima, Y., Corander, J. High-dimensional structure learning of binary pairwise Markov networks: A comparative numerical study, Computational Statistics & Data Analysis, 2020, 141:62-76.

Please cite the above paper when using this method (modified or as is).

Usage

See the included script read_me.R for step-by-step instructions on how to use the code.

MATLAB version

The original (and somewhat faster) MATLAB version of the sampler can be found at https://bitbucket.org/jopensar/srbm-sampler/.