Computer Chess Club Archives




Subject: Re: bayeselo: new Elo-rating tool, applied to CCT7

Author: Rémi Coulom

Date: 01:11:55 02/15/05

Go up one level in this thread

On February 15, 2005 at 00:37:48, Charles Roberson wrote:

>  Remi,
>    Nice work but, I find it hard to believe that Zappa is > 1500 pts stonger
>  than GnuChess (SuperBot). If we arbitrarily rate zappa at 3000 that puts
>GNUChess at 14xx. Or did I read the table incorrectly?
>  Charles

Yes, you read the table correctly. In a Swiss tournament with many more
participants than rounds, bayeselo tends to produce ratings that are on a much
wider range than ELOStat. That's because bayeselo is based on a very different
statistical approach. So, the elo ratings it produces should not be considered
equivalent to "usual elo".

The real way to test the value of a rating system is by measuring is predictive
power. I have not done it already, because this program is very new, but I will
run cross-validation tests soon.

Thanks for your feedback, anyway. I had not noticed this problem before. I will
have to do some testing to find out whether it is really a problem. Maybe I can
find a way to tweak my program so that the ratings it produces are closer to
"usual elo".


This page took 0.18 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.