The Elo rating system was invented half a century ago by Hungarian-born physicist and chess master Arpad Elo. It is the most famous technique for rating chess players and is used throughout the chess world. It has been applied to many other contests as well, including other board games, sports, and video games. However, it has never really been demonstrated that the Elo approach to calculating chess ratings is superior. Elo's formula was derived theoretically, in an era without large amounts of historical data or significant computing power. With the benefit of powerful computers and large game databases, we can easily investigate approaches that might do better than Elo at predicting chess results.
There are several alternatives to the Elo approach. Professor Mark Glickman developed the Glicko and Glicko-2 systems, which extend the Elo system by introducing additional parameters to represent the reliability and volatility of player ratings. Ken Thompson uses a linearly weighted average of a player's last 100 results to calculate a weighted performance rating. Jeff Sonas (who put together this competition) developed Chessmetrics ratings to maximize predictive power. More details are available on the
We want to see if somebody out there can do even better. Competitors train their rating systems using a training dataset of over 65,000 recent results for 8,631 top players. Participants then use their method to predict the outcome of a further 7,809 games.
Along with the opportunity to help shape the future of chess ratings, the top ten entries win prizes (assuming they share their methodology).
1. Fritz DVD autographed by world champions Viswanathan Anand, Garry Kasparov, Anatoly Karpov and Viktor Korchnoi (see image)
2. ChessBase 10 Starter Package
3. Big Database 2010 ...
For details and to participate, see kaggle.com/chess
Update: after 1 day, Elo benchmark was already beaten - see