Free Software for Bayesian Statistical Inference

Regression and classification:

• Software for Flexible Bayesian Modeling and Markov Chain Sampling, by Radford Neal. Includes neural networks, Gaussian processes, and other models.
• Tpros: Gaussian Processes Papers and Software, by Mark Gibbs. Alternate link.
• gp-mc-1: Gaussian processes for regression, fully Bayesian approach, by Carl Rasmussen. Note: The link for the paper describing the algorithms used in the paper is wrong; this is the correct link.
• gp-map-1: Gaussian processes for regression, MAP approach, by Carl Rasmussen. Note: The link for the paper describing the algorithms used in the paper is wrong; this is the correct link.
• An Octave-based demo of Gaussian Processes (Gzipped tar file).
• IND — Wray Buntine's Bayesian decision tree software, based on his Ph.D. dissertation.
• Bayesian Logistic Regression Software for sparse models. This software can pick out an appropriate set of features from a set of tens of thousands of predictors; it was developed with text categorization in mind, with the features being presence or absence of a word.
• Bayesian Essay Test Scoring System - BETSY. A text classification tool.

General statistical analysis:

• BUGS / WinBUGS (Bayesian Inference Using Gibbs Sampling). Probably the most popular and flexible software for Bayesian statistics around. Has a powerful model description language, and uses Markov Chain Monte Carlo to do a full Bayesian analysis.
• BayesX. Software for semiparametric regression using MCMC, inference for STAR (structured additive predictor) models, model selection for Gaussian and non-Gaussian DAGs, etc.
• R. An open-source implementation of the S language for data analysis. Mostly oriented towards frequentist statistics, but there are some packages for Bayesian statistics.
• bugs.R: Functions for running WinBUGS from R.
• Bayesian Output Analysis Program. Convergence diagnostics for Markov Chain Monte Carlo.
• R-CODA. Convergence diagnostics for Markov Chain Monte Carlo.
• Software for Bayesian meta-analysis: hblm in S-plus. Hierarchical Bayes Linear Models. Used to combine the results from several independent studies. S-plus code and documentation.
• BACC: Bayesian Analysis, Computation, and Communication. A collection of statistical routines implemented for R, S-plus, and Matlab.
• Software and Datasets (Adrian Raftery). A collection of S functions for various statical analyses, many of them Bayesian or useful as part of a full Bayesian analysis.
• Bayesian Model Selection Software (Adrian Raftery). S functions for computing posterior probabilities of models.
• Bayesian Model Averaging Home Page. Software in S for model averageing, which accounts for uncerty in model selection when making predictions.

Bayesian networks:

• Kevin Murphy's list of Software Packages for Graphical Models / Bayesian Networks.
• Intel's Open Source Probabilistic Networks Library (PNL). Both learning of and inference with Bayesian networks.
• Bayesian Network Tools in Java Both inference from network, and learning of network.
• JavaBayes: Bayesian Networks in Java. Software for inference with Bayesian networks; as of June 18 2004 it does not include facilities for learning structure or parameters of the networks.
• J. Cheng's Bayesian Belief Network Software. KDDCup 2001 Data Mining Competition winner. BN PowerConstructor, BN PowerPredictor, DataPreprocessor. Learns Bayesian networks, both structures and parameters.
• WinMine Toolkit Home Page. Data mining software. Learns both structure and parameters of Bayesian networks.
• MSBNx: Microsoft Bayesian Network Editor and ToolKit. Can be used to both create and evaluate a Bayesian network. Alternate link.
• BNG: Bayesian Network Generator. Knowledge-based construction of Bayesian networks.
• jBNC: Bayesian Network Classifier Toolbox. Java toolkit for training, testing, and applying Bayesian network classifiers.
• Kevin Murphy's Bayes Net ToolBox for Matlab. Includes a variety of algorithms for both inference (evaluation of net), parameter learning, and structure learning. Has support for some kinds of continuous conditional distributions, as well as utility nodes (influence diagrams).
• GeNIe 2.0 & SMILE. SMILE is a library of C++ classes for manipulating and evaluating Bayesian networks and influence diagrams. GeNIe is a graphical end-user application built on top of SMILE.
• DEAL: Learning Bayesian Networks in R. Can handle discrete and/or continuous variables (continous nodes must be conditionally Gaussian). Can learn parameters of network, and has some ability to learn structure (improve on an initial guess).
• Coco. Analysis of associations between discrete variables (log-linear interactions). Runs under XLISP-Stat.
• GRAPPA. A suite of functions in R for probability propagation in discrete graphical models.
• TETRAD: Causal Models and Statistical Data. Program for creating, simulating data from, estimating, testing, predicting with, and searching for causal/statistical models (Bayesian networks or graphical Gaussian models).
• Stochastic Computation For Gaussian Graphical Models. Software to evaluate models.
• GDAGSim. C library for conditional simulation of Gaussian DAG models.
• GMRFLib. C library for fast and exact simulation of Gaussian Markov Random Fields (GMRF) on graphs. The library performs,unconditional simulation of a GMRF, various types of conditional simulation from a GMRF, evaluation of the corresponding log-density, and generation of blockupdates in MCMC-algorithms.
• gR: gRaphical Models in R. An umbrella for various software packages for graphical models that are either written in R or can be called from R.

Time Series / dynamic models:

• Nonstationary Time Series Analysis and Decomposition using Time-Varying Parameter Autoregressions. Software for fitting, analysis and exploration of time series using classes of time-varying autoregressions — or TVAR models.
• Autoregressive component models, model uncertainty and unit roots. Software for fitting autoregressive models to time series.
• Bayesian Filtering Library for inference with models such as Kalman filters, hidden Markov models, particle filters, etc.
• Bayes++ Bayesian Filter Classes. Library of C++ classes for Bayesian filtering (e.g., Kalman filter, extended Kalman filter, etc.)
• SsfPack: C routines for state-space approach to time series analysis.
• ReBEL : Recursive Bayesian Estimation Library. Matlab toolkit of functions and scripts, designed to facilitate sequential Bayesian inference (estimation) in general state space models (Kalman filter, extended Kalman filter, sigma-point Kalman filter, particle filters, etc.)

Miscellaneous:

• First Bayes. Teaching package for elementary Bayesian statistics. Runs on all versions of Windows.
• Autoclass: automatic class discovery. Discovers clusters/classes in data that may include both real and discrete attributes. Alternate link.
• Autoclass C. C language implementation of Autoclass.
• Model-based Clustering Software (MCLUST / EMCLUST). Written in FORTRAN and interfaced to S-plus and R.
• GSL: Gnu Scientific Library. Contains many functions that are useful for writing statistical software.
• MCSim. Software that takes a model specification and creates a C program to do Markov Chain Monte Carlo evaluation of that model. The advantage is speed of simulation.
• Hydra MCMC Library. A platform-neutral library for performing Markov chain Monte Carlo, written in Java. Alternate links: Sourceforge, download page.
• BAYESPACK (scroll down to BAYESPACK entry in list after clicking on link). A collection of Fortran software for numerical evaluation of integrals that arise in Bayesian statistical analysis. Paper describing the system.
• BayeSys. Software for Markov Chain Monte Carlo and computation on evidence (a.k.a. model likelihood or marginal likelihood).
• OpenNLP Maxent. Java package for training and using maximum-entropy models.
• [B/D]: The Bayes Linear Programming Language.