Monday, December 13, 2010

Ratings Contest

A website called Kaggle have regular contests where they give people a data set, and have them to predict something about the data set. They have done it for a variety of topics (e.g., tourism prediction, grant application success) and, of course, they just finished one for chess ratings.

Spurred on by Jeff Sonas (the Chessmetrics guy), there were a lot of entries, and suffice it to say, someone came up with a method that is 6.6% better than standard Elo ratings.

You can read all about it here, and wait for the follow up contest in January.

No comments: