This interface allows JAGS ("Just Another Gibbs Sampler") to be used in combination with Matlab. JAGS is a program for Bayesian hierarchical models using Markov chain Monte Carlo (MCMC) inference methods. JAGS is similar to the OpenBUGS and WinBUGS programs but can operate on a number of platforms including Windows and Max OS X.
This Matlab interface is based on matbugs written by Kevin Murphy and Maryam Mahdaviani for WinBUGS and OpenBUGS. This JAGS interface uses very similar syntax.
One of the main advantages of this interface is that the MCMC chains can run in parallel if the Matlab Parallel Computing Toolbox is installed. This can lead to near-linear speedups in simulation time. For example, running 8 chains in parallel on a 8-core machine can lead to a speedup factor of almost 8.
This version of MATJAGS was tested on JAGS version 3.3.0, with Matlab version R2013A. It was designed for a windows platform but the latest code should also run on a Mac.
Script Description Required software exampleJAGS1 running multiple MCMC chains serially for rate inference problem exampleJAGS2 running multiple MCMC chains in parallel for rate inference problem Parallel Computing Toolbox exampleJAGS3 comparing JAGS to WinBUGS for the rate inference problem Matbugs (included) and WinBUGS exampleJAGS4 calculate the speedup factor when running the MCMC chains in parallel Parallel Computing Toolbox exampleJAGS5 comparing JAGS to WinBUGS for the kappa inference problem Matbugs (included) and WinBUGS
10/31/2011 Matjags was updated to work with Mac platforms (thanks to Mike Kalish for helpful suggestions)