Scythe MCMC Documentation

0.1

Library	Scythe MCMC - A Scythe Markov Chain Monte Carlo C++ Framework
Author	Tristan Zajonc (tristanz@gmail.com)
Source	http://ksghome.harvard.edu/~zajonct/
Version	0.1

INTRODUCTION

Scythe MCMC is a C++ header library that eases the development of Markov Chain Monte Carlo (MCMC). It is based on the Scythe Statistical Library. Scythe MCMC provides an execution framework, including command line and option file parsing, that reduces the amount of boilerplate code needed to write custom MCMC routines. It also provides common MCMC step types, including Gibbs, Metropolis-Hastings, and Slice sampling. Users can experiment with which sampling steps provide the best results and implement their own sampling steps as desired.

Scythe MCMC was motivated with the need to combine simpler samplers, such as univariate slice sampling, with more specialized samplers for particular problems. Future versions of Scythe MCMC will include common samplers for nonparametric Bayesian models, including a Collapsed Gibbs sampler for the Dirichlet Process, and Slice samplers for the Dirichlet Process and Indian Buffet Process.

There are many choices for writing MCMC samplers. Scythe MCMC is similar to, and heavily inspired by, MCMC++, but is based on the Scythe, which provides convenient matrix types, random number generators, and probability distributions. Scythe MCMC is different in focus from MCMCPack, which is also based on Scythe and interfaces with R. Users wishing to distribute a specific model to R users should consider contributing to MCMCPack directly. Unlike MCMCPack, Scythe MCMC makes no attempt to interface with R and assumes samplers are executed from the command line (although this is not required). It also provides common algorithms (Gibbs, Metropolis-Hastings, Slice Sampling) that ease development of samplers for many models.

Even though implementing models in ScytheMCMC requires more work than implementing comparable models in Bugs or JAGS, Scythe MCMC generally leads to faster samplers that can exploit the particular structure of a given problem. It also allows development of models that cannot be sampled effectively using either Bugs or Jags, such nonparametric Bayesian models without truncation approximations.

FEATURES

• Based on the Scythe Statistical Library.
• Eliminates commandline and option parsing boilerplate code using SimpleOpt and SimpleIni.
• Built in sampling steps:
  • Gibbs,
  • Metropolis-Hastings,
  • Slice Sampling.
  • Walker (2008) sampler for DP.
• Nonparameteric samplers including:
  • Collapsed gibbs samplers for Dirichlet Process,
  • Slice sampler for Dirichlet Process and Indian Buffet Process.
• MIT License.

GETTING STARTED

MIT LICENSE

The license text below is the boilerplate "MIT License" used from: http://www.opensource.org/licenses/mit-license.php

Copyright (c) 2009, Tristan Zajonc

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.