Here is various stuff for the BIRS workshop on Statistical Inference Problems in High Energy Physics and Astronomy, by Radford Neal.
Here is a write-up of some preliminary notes and experiments: Postscript or PDF.
Here are some additional notes on systematic errors: Postscript or PDF.
Here are my divisions of Byron Roe's data into training and test sets: training and test.
Both of the above files have one line of 51 numbers per event, with that line containing 50 event descriptors followed by the class, coded as 0 for background and 1 for signal. There are no headers.
The predictive probabilities for test cases I found using a Bayesian neural network model are here.
Some plots exploring what's going on:
Scatterplot of two variables (class marked)
Comparison of two classifiers
Check for residual information
Plot showing linear/additive/non-additive effect
My estimates of the relevance of each of the 50 PID variables are here. Bigger numbers mean more relevance. (Don't take these too seriously, however, since they're just the simplest version of interpreting the network hyperparameters, which could be a bit misleading.) Update: Actually, you should take the previous values even less seriously than I thought. A better set of relevances is here, taking account of some strange aspects of the data that I hadn't noticed before. There may still be some strangenesses.