Monday, February 21, 2011

2010 March Madness Contest Winning Algorithm: Code

Posted by Danny Tarlow
Last year's March Madness algorithm predictions contest ended with me winning the full bracket, and Scott Turner (aka "The Pain Machine") and I tied for the sweet 16 bracket:
http://blog.smellthedata.com/2010/04/duke-is-not-only-winner-from-last-night.html

I've gotten a few emails asking for the code I used. I haven't looked at it since last year, but the Python code is available here:
http://www.cs.toronto.edu/~dtarlow/march_madness_public.tgz

You can follow links for the description of the probabilistic matrix factorization model, and a post using a similar model applied to a recommendation system.

Unfortunately, the documentation is a bit lacking beyond that. Hopefully the posts are enough to explain things at a high level. If you have more detailed questions (or find bugs), feel free to ask (or point them out) in the comments.

If somebody adapts this code to work with this year's data format, please share! (and let me know)

This code may also be of interest.

No comments: