runjags (version 1.2.1-0)

runjags: Interface utilities for Just Another Gibbs Sampler (JAGS) using parallel and distributed computing

Description

This package provides high-level interface utilities for Just Another Gibbs Sampler (JAGS). The primary functions facilitate running parallel JAGS models for multiple chains, automated control of convergence and sample length diagnostics, evaluation of the performance of a model against simulated data, and compatibility with the WinBUGS syntax of model files with data and initial values lists. Model fitting can be done locally (via the rjags package or separate JAGS instances) or via distributed computing clusters such as those provided by snow (a Simple Network Of Workstations), Xgrid, or other user specified functions. A JAGS extension module implementing additional distributions for JAGS models (including the Pareto family of distributions) is also provided. Requires Just Another Gibbs Sampler (JAGS) for most functions, see: http://mcmc-jags.sourceforge.net

Details

JAGS is a program which allows analysis of Bayesian models using Markov chain Monte Carlo (MCMC) simulation, and was developed by Martyn Plummer to be an alternative to BUGS that ran on UNIX systems as well as Windows systems. The R package rjags is a native R interface to the JAGS library, and allows a greater level of control for compiled models, which may be more useful for model development. This package was intended to provide additional functions to help automate the process of running models, including interpretation of WinBUGS type text files including data and initial values, automated convergence diagnostics, automated collation and plotting of results, and convinience wrappers for running models (either individually or for multiple data sets) over distributed computing cluster such as those provided by snow and Apple's (now discontinued) Xgrid.

The package also includes a JAGS extension module providing additional distributions in the Pareto family - for more details see the vignette for this installed package. A standalone version of this JAGS module (as well as a version of the runjags package without this module included) is available from the runjags sourceforge page at: https://sourceforge.net/projects/runjags/

References

M. J. Denwood, "runjags: An R Package Providing Interface Utilities, Distributed Computing Methods and Additional Distributions For MCMC Models in JAGS," Journal of Statistical Software, [Under review].

See Also

run.jags and extend.jags for basic model runs

autorun.jags and autoextend.jags for automated running of models to convergence and automated calculation of necessary sample sizes

runjags-class for S3 methods relating to runjags objects, incluing conversion to/from jags objects (for compatibility with the rjags package)

runjags.options for ways to set default options for runjags functions

read.winbugs for translation of WinBUGS text files into JAGS compatible model, data and initial values files

combine.mcmc and dump.format for MCMC related tools

timestring, new_unique and ask for more general tools

xgrid.run.jags and xgrid.submit.jags for use of Xgrid clusters to run JAGS models remotely

xgrid.run and xgrid.submit for use of Xgrid clusters for remote execution of user specified R code

jags.model in the rjags package for fine control over the JAGS libraries