Meanwhile on ChessBase, there is another set of predictions using computer simulations (http://en.chessbase.com/post/computer-simulates-and-predicts-candidates-winner). Our methods differ a bit, though the predictions are fairly similar. They estimated the draw rates using a database of recent games between the players. This is a smaller sample than in my model, and there is the difficulty of deciding how recent the games have to be in order to be relevant for this tournament. However, it might be able to approximate the effect of different styles. For example, even when Giri and Topalov have similar ratings and face similar opponents, they may still have different draw rates. Is it better to estimate this effect with a small sample (knowing that small samples can be unreliable) than it is to ignore it, as my model does? At the moment, it's hard to say. Once the university upgrades my computer, I might be able to get an answer. Their model also uses deep learning, which is something I don't know very much about.
I have not tried to estimate the results of the tiebreaks. For many tournaments, the tiebreak rules seem to be complicated or arbitrary. If it comes down to rapid and blitz games, the databases don't have a lot of reliable data on that yet.
No comments:
Post a Comment