Computer Chess Club Archives




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

Author: Rémi Coulom

Date: 09:52:27 02/14/05

Go up one level in this thread

On February 14, 2005 at 10:40:30, Steve Maughan wrote:

>Interesting and cool - I really like the matrix view.  Could you explain at a
>reasonably high level the theory behind this approach?

The theory is explained on the page (,
and in the links that are at the bottom of the page.

If you found it too detailed or too complicated, I would be glad to explain
more. But you'll have to tell me first what you do not understand. Do you
understand Bayesian inference ? Do you know what is a likelihood distribution
and a conditional probability ? Do you know what is a Gaussian, a covariance
matrix, a Hessian matrix ? Do you understand how ELOStat works ?

In practice, the most important differences between my tool and ELOStat are:
 - The estimation of ratings is based on a more sound statistical method (well,
this may be open to debate, but I am deeply convinced of it)
 - bayeselo takes the advantage of playing white into consideration, whereas
ELOStat does not.
 - bayeselo is able to deal correctly with tricky situations, such as: A plays
many games against B, C plays many games against D, and B plays only one game
against C. An example is given there:


This page took 0.03 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.