Software for Bayesian inference with Directed Mixed Graph Models

The following file provides a R package and Java classes for our inference software, dmgBayes. The dmgBayes software does MCMC-based Bayesian inference in directed mixed graph models. This is an on-going project and we expect to expand it for other families of probabilistic models. The current version deals with Gaussian models only (a.k.a structural equation models):

• Version 1.0, July 24th 2008: source code and compiled Java classes. Includes a R file that has to be manually configured (full documentation included). This distribution (300K) does not include the Tetrad java library. You will need the latest JAR file containing the Tetrad library from the Tetrad project webpage. Go to the download directory and get the tetrad-4.3.8-6.jar file, or a more recent version.

A future self-contained R package is being planned. It is not hard to set up the current R code, however: it is just one file, as explained in the included documentation. However, you will need to a Java virtual machine running in your systems, plus the R package rJava (available from CRAN).

The current main reference for dmgBayes is our 2008 report. The following are also relevant references to better understand mixed graph models:

• K. Bollen (1989). Structural Equations with Latent Variables. Wiley.
• T. Richardson and P. Spirtes (2002). Ancestral graph Markov models. Annals of Statistics. Technical report here.
• T. Richardson (2003). Markov properties for acyclic directed mixed graphs. Scandinavian Journal of Statistics. Technical report here.
• M. Drton and T. Richardson (2004) Iterative conditional fitting for Gaussian ancestral graph models. Proceedings of UAI'04.

Last modification: July 25th 2008