FACILITIES PROVIDED BY THIS SOFTWARE

This software is meant to support research and education regarding:

   * Flexible Bayesian models for regression and classification 
     based on neural networks and Gaussian processes, and for
     probability density estimation using mixtures.  Neural net
     training using early stopping is also supported.

   * Markov chain Monte Carlo methods, and their applications to
     Bayesian modeling, including implementations of Metropolis,
     hybrid Monte Carlo, slice sampling, and tempering methods.

These facilities might be useful for actual problems, but you should
note that many features that might be needed for real problems have
not been implemented, that the programs have not been tested to the
extent that would be desirable for important applications, and that
permission to use the software for free is granted only for purposes
of research and education.

The complete source code (in C) is provided, allowing researchers to
modify the program to test new ideas.  It is not necessary to known C
to use the programs (assuming you manage to install them correctly).

This software is designed for use on a Unix system, using commands
issued to the Unix command interpreter (shell).  No particular window
system or other GUI is required, but a plotting program will be very
useful.  I use the xgraph plot program, written by David Harrison,
which allows plots to be produced by just piping data from one of the
commands; it can be obtained from my web page.


Markov chain Monte Carlo facilities.

All the Bayesian models are implemented using Markov chains to sample
from the posterior distribution.  For the elaborate models based on
neural networks, Gaussian processes, and mixtures, this is done by
combining general-purpose Markov chain sampling procedures with
special modules written in C.  Other models could be implemented in
the same way, but this is a fairly major project.

To allow people to play around with the various Markov chain methods
more easily, a facility is provided for defining distributions (on
R^n) by giving a simple formula for the probability density.  Many
Markov chain sampling methods, such as the Metropolis algorithm,
hybrid Monte Carlo, slice sampling, and simulated tempering, may then
be used to sample from this distribution.  Bayesian posterior
distributions can be defined by giving a formula for the prior density
and for the likelihood based on each of the cases (which are assumed
to be independent).

A long review paper of mine on "Probabilistic Inference Using Markov
Chain Monte Carlo Methods" can be obtained from my web page.  This
review discusses methods based on Hamiltonian dynamics, including the
"hybrid Monte Carlo" method.  These methods are also discussed in my
book on "Bayesian Learning for Neural Networks".  My web page also has
papers on slice sampling ("Markov chain Monte Carlo methods based on
`slicing' the density function") and on Annealed Importance Sampling,
both of which are implemented in this software.


Neural network and Gaussian process models.

The neural network models are described in my thesis, "Bayesian
Learning for Neural Networks", which has now been published by
Springer-Verlag (ISBN 0-387-94724-8).  The neural network models
implemented are essentially as described in the Appendix of this book.
The Gaussian process models are in many ways analogous to the network
models.  The Gaussian process models implemented in this software, and
computational methods that used, are described in my technical report
entitled "Monte Carlo implementation of Gaussian process models for
Bayesian regression and classification", available in compressed
Postscript at URL ftp://ftp.cs.utoronto.ca/pub/radford/mc-gp.ps.Z.
The Gaussian process models for regression are similar to those
evaluated by Carl Rasmussen in his thesis, "Evaluation of Gaussian
Processes and other Methods for Non-Linear Regression", available from
his web page, at the URL http://www.cs.utoronto.ca/~carl/; he also
talks about neural network models.  To understand how to use the
software implementing these models, it is essential for you to have
read at least one of these references.

The neural network software supports Bayesian learning for regression
problems, classification problems, and survival analysis (experimental), 
using models based on networks with any number of hidden layers, with
a wide variety of prior distributions for network parameters and
hyperparameters.  The Gaussian process software supports regression
and classification models that are similar to neural network models
with an infinite number of hidden units, using Gaussian priors.

The advantages of Bayesian learning for both types of model include
the automatic determination of "regularization" hyperparameters,
without the need for a validation set, the avoidance of overfitting
when using large networks, and the quantification of uncertainty in
predictions.  The software implements the Automatic Relevance
Determination (ARD) approach to handling inputs that may turn out to
be irrelevant (developed with David MacKay).  

For problems and networks of moderate size (eg, 200 training cases, 10
inputs, 20 hidden units), fully training a neural network model (to
the point where one can be reasonably sure that the correct Bayesian
answer has been found) typically takes several hours to a day on our
SGI machine.  However, quite good results, competitive with other
methods, are often obtained after training for under an hour.  The
time required to train the Gaussian process models depends a lot on
the number of training cases.  For 100 cases, these models may take
only a few minutes to train (again, to the point where one can be
reasonably sure that convergence to the correct answer has occurred).
For 1000 cases, however, training might well take a day.

The software also implements neural network training using early
stopping, as described in my paper on "Assessing relevance
determination methods using DELVE" (to appear in Generalization in
Neural Networks and Machine Learning, C. M. Bishop (editor),
Springer-Verlag).  A similar early stopping method is also described
in Carl Rasmussen's thesis (see above).


Bayesian mixture models.

The software Bayesian mixture models for multivariate real or binary
data, with both finite and countably infinite numbers of components.
The countably infinite mixture models are equivalent to Dirichlet
process mixture models.  The sampling methods that I have implemented
for these models are described in my technical report on "Markov chain
sampling methods for Dirichlet process mixture models", which can be
obtained from ftp://ftp.cs.utoronto.ca/pub/radford/mixmc.ps.Z.  See
also my technical report on "Bayesian mixture modeling by Monte Carlo
simulation", at ftp://ftp.cs.utoronto.ca/pub/radford/bmm.ps.Z.  

The software implements the basic sampling operations for these
models, but is rather preliminary in some respects, lacking many
facilities that would be useful in practical applications.


Software components.

The software consists of a number of programs and modules.  Each major
component has its own directory, as follows:
  
    util    Modules and programs of general utility.

    mc      Modules and programs that support sampling using Markov 
            chain Monte Carlo methods, using modules from util.

    dist    Programs for doing Markov chain sampling on a distribution
            given by a simple formula, or by giving a Bayesian prior
            and likelihood, using the modules from util and mc.

    net     Modules and programs that implement Bayesian inference
            for models based on multilayer perceptron neural networks, 
            using the modules from util and mc.  Also implements simple
            gradient descent training, possibly with early stopping.

    gp      Modules and programs that implement Bayesian inference
            for models based on Gaussian processes, using the modules
            from util and mc.

    mix     Modules and programs that implement Bayesian inference
            for finite and infinite mixture models, using modules
            from util and mc.

In addition, the 'bvg' directory contains modules and programs for
sampling from a bivariate Gaussian distribution, as a simple
demonstration of how the Markov chain Monte Carlo facilities can be
used from a special module written in C.  Other than by providing this
example, and the detailed documentation on various commands, I have
not attempted to document how you might go about using the Markov
chain Monte Carlo modules for another application written in C.

The following directories contain examples of how these programs can
be used, many of which are discussed in the documentation:

    ex-netgp  Examples of Bayesian regression and classification 
              models based on neural networks and Gaussian processes.

    ex-mix    Examples of Bayesian mixture models.  Includes command
              files for the test in my paper on "Markov chain sampling 
              methods for Dirichlet process mixture models"

    ex-dist   Examples of Markov chain sampling on distributions
              specified by simple formulas.

    ex-bayes  Examples of Markov chain sampling for Bayesian models
              specified using formulas for the prior and likelihood.

    ex-gdes   Examples of neural network learning using gradient 
              descent and early stopping.

    ex-ais    Contains command and data files used for the tests in
              my paper on "Annealed importance sampling".

The 'bin' directory contains links to all the programs.  The 'doc'
directory contains all the documentation.


Portability of the software.

The software is written in ANSI C, and is meant to be run in a UNIX
environment.  Specifically, it was developed on an SGI machine running
IRIX Release 5.3.  It also seems to run OK on a SPARC machine running
SunOS 5, using the 'gcc' C compiler, and on DEC Alpha machines,
provided that the -ieee and -std options are given to the C compiler.
As far as I know, the software does not depend on any peculiarities of
these environments (except perhaps for the use of the drand48
pseudo-random number generator, and the lgamma function), but you may
nevertheless have problems getting it to work in substantially
different environments, and I can offer little or no assistance in
this regard.  There is no dependence on any particular graphics
package or graphical user interface.  (The 'xxx-plt' programs are
designed to allow their output to be piped directly into the 'xgraph'
plotting program, but other plotting programs can be used instead, or
the numbers can be examined directly.)