Introduction

MCMCMS (Markov chain Monte Carlo over Model Structures) is a collection of programs which implement the ideas introduced in [CussensCussens2000], and further analysed in [Angelopoulos CussensAngelopoulos Cussens2001]. The programs were initially developed under the EPSRC research grant Induction of Stochastic Logic Programs, November 2000-October 2001. Currently the development is supported by EPSRC's MATHfit programme, under the grant Stochastic Logic Programs for MCMC, September 2003-August 2005. The software can be downloaded from http://www.cs.york.ac.uk/~nicos/sware/slps/mcmcms/ while most of the papers can be downloaded from http://www.cs.york.ac.uk/~jc.

To date we have run experiments over Bayesian networks (BNs), Restricted Acyclic Partially Directed Graphs (RPDAGs), pedigrees and Classification and Regression Trees (C&RTs). The methodology is quite generic and can be applied to any model space, provided an SLP with the desired prior is constructed, and a method for computing the likelihood of constructed models is incorporated.

The remainder of this guide is structured as follows: Section 2 is a tutorial tour of the software. Section 4 describes the basic directory structure and the main files. Section 5 shows how to run MCMC experiments over SLPs. Section 6 gives information on the model spaces for which MCMCMS has been used so far. Section 7 details how the programs can work over new spaces. Finally Section 8 briefly reviews some features of the system which are still to be documented fully.