This directory contains data and command files for the examples used
in the talk on ``Survival analysis using a Bayesian neural network''
that I gave at the 2001 Joint Statistical Meetings in Atlanta.  The
slides from this talk are available via the following web page:

    http://www.cs.toronto.edu/~radford/misc-docs.html

The data set used is from a clinical trial of a drug to treat Primary
Biliary Cirrhosis (PBC), which was used as an example by Fleming and
Harrington in their book on Counting Processes and Survival Analysis.
This data was obtained from Statlib (http://lib.stat.cmu.edu/).  The
file statlib-src contains the data and documentation as obtained from
their.

The entire data set in slightly different format is in the file
"data".  The R script process.r is used to process this data and write
just the cases pertaining to the clinical trial and just the variables
used to the file "data.use".  The files "data.trn" and "data.tst" are
a random split of this data into "training" and "test" sets.  

The shell files ending with .cmds fit various neural network models to
either all the data in data.use, or just the data in data.trn,
depending on whether "use" or "trn" is given as the argument.  The log
files produced have the corresponding name ending in .log.use or
.log.trn.

The process.r script defines various R functions for displaying the
results of these runs.  The scripts plot1.r, plot2.r, etc. were used
to produce various graphs in the talk.
