CraigKelly / grample

Sampling for Probabilistic Graphical Models

title	subtitle
grample README	Sampling for Probabilistic Graphical Models

Introduction

This software package is designed to read Markov networks and perform marginal estimation using Gibbs sampling. The main motivation for creating Yet Another MCMC software package was research: this is the experimental implementation of Adaptive Rao-Blackwellisation as reported in Adaptive Rao-Blackwellisation in Gibbs Sampling for Probabilistic Graphical Models

Using and Citing

This code is licensed under the MIT license: see LICENSE for details. If you use this code in a published work, please cite the paper here:

@inproceedings{AdaptiveRBGibbs,
  title={Adaptive Rao-Blackwellisation in Gibbs Sampling for Probabilistic Graphical Models},
  author={Kelly, Craig and Sarkhel, Somdeb and Venugopal, Deepak},
  booktitle={The 22nd International Conference on Artificial Intelligence and Statistics},
  editor = 	 {Chaudhuri, Kamalika and Sugiyama, Masashi},
  pages={2907--2915},
  year={2019},
  series = {Proceedings of Machine Learning Research},
}

Installing and Running

There's no real installion. Use go get -u github.com/CraigKelly/grample to get the latest code. From inside the grample directory, run make to build. Then you can run ./grample -h to get command line help. You can see some examples in ./script/experiment

Using As a Library

If you want to grample as a library, that's fairly easy. There aren't any directions right now, but see ./cmd/root.go for examples. That's our main command line implementation, so you can get a good idea of how to use the sampler package.

Dependencies

As of this writing, this code has only been tested with Go 1.17.

This repo now uses the standard go mod commands for dependencies. Note that this was ported from using dep for dependency management, and we have removed the vendor dir. The short story is that we don't have many dependencies, but we are using github.com/spf13/cobra to manage the command line and github.com/stretchr/testify for unit test assertions.

Hacking

Use the Makefile, which delegates to scripts located in ./scripts.