Software for Flexible Bayesian Modeling and Markov Chain Sampling

This software supports Bayesian regression and classification models based on neural networks and Gaussian processes, and Bayesian density estimation and clustering using mixture models and Dirichlet diffusion trees. It also supports a variety of Markov chain sampling methods, which may be applied to distributions specified by simple formulas, including simple Bayesian models defined by formulas for the prior and likelihood.

Before trying to use the software, you may need to read various references that describe the models and the Markov chain methods used.

The software is written in ANSI C for Unix and Linux systems.

Some of the programs in this package are designed for use with the xgraph plot package for X-Windows, though the data can instead be plotted by whatever plot program you have available.

Current release of 2004-11-10:

Index to documentation in hypertext form
Compressed tar archive of everything (1.4 Megabytes)
Uncompressed tar archive of everything (4.0 Megabytes)

The tar archives include the complete hypertext documentation, so that you can set it up to be read from your local machine. The documentation is also readable the old-fashioned way. For directions on how to untar a tar archive, see the installation instructions within the above hypertext documentation.

NOTE: In this release (as well as the previous one), the maximum number of target values is set to 10000, which can be useful, but which leads to rather large log files for mixture models (other models are not affected). You may wish to reduce the value defined for Max_targets in mix/data.h to avoid this.

Old release of 2003-06-29:

Index to documentation in hypertext form
Compressed tar archive of everything (1.4 Megabytes)
Uncompressed tar archive of everything (3.9 Megabytes)

Old release of 2001-08-31:

Index to documentation in hypertext form
Compressed tar archive of everything (1.2 Megabytes)
Uncompressed tar archive of everything (3.4 Megabytes)

Warning: This version contains a bug in net-pred that affects predictions for multi-way classification models. It can be patched by changing line 392 of net/net-model.c, replacing "type=1" with "type==1". There's also a bug affecting the "n" option for gp-pred for multi-way classification models, which you can patch by replacing "+=" with "=" on line 440 of gp-pred.c.

Old release of 2000-08-21:

Index to documentation in hypertext form
Compressed tar archive of everything (1.2 Megabytes)
Uncompressed tar archive of everything (3.2 Megabytes)

Old release of 1999-12-06:

Index to documentation in hypertext form
Compressed tar archive of everything (1.1 Megabytes)
Uncompressed tar archive of everything (3.1 Megabytes)

Old release of 1999-03-13:

Index to documentation in hypertext form
Compressed tar archive of everything (1.1 Megabytes)
Uncompressed tar archive of everything (3.0 Megabytes)

Old release of 1998-09-01:

Index to documentation in hypertext form
Compressed tar archive of everything (1.0 Megabytes)
Uncompressed tar archive of everything (2.8 Megabytes)

Old release of 1998-08-02:

Index to documentation in hypertext form
Compressed tar archive of everything (1.0 Megabytes)
Uncompressed tar archive of everything (2.8 Megabytes)

This was the first release with the facilities for sampling from distributions specified by formulas for the "energy" or the prior and likelihood.

Old release of 1997-07-22:

Index to documentation in hypertext form
Compressed tar archive of everything (817K)
Uncompressed tar archive of everything (2150K)

This is the release that was used for the tests in my paper on Assessing relevance determination methods using DELVE.

Old release of 1997-01-18:

Index to documentation in hypertext form
Compressed tar archive of everything (600K)
Uncompressed tar archive of everything (1600K)

Old release of 1996-08-26:

Index to documentation in hypertext form
Compressed tar archive of everything (500K)
Uncompressed tar archive of everything (1200K)

This is the release that was used by Carl Rasmussen for the neural network models evaluated in his thesis. Gaussian processes were not implemented in this release.

Old release of 1995-08-09:

Documentation (50K)
Compressed tar archive (400K, includes documentation)
Uncompressed tar archive (800K, includes documentation)

Back to Radford Neal's home page