Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Very interesting

Author: Kolss

Date: 00:43:52 04/05/05

Go up one level in this thread


On April 04, 2005 at 17:31:20, Ray Banks wrote:

>Especially the bit about his testing methodology - the fact that he says he
>needs hundreds if not thousands of games to be sure that any changes he makes
>are an improvement or not. We all know that a large number of games is
>necessary, but possibly over a thousand was a surprise...

Hi,

Thousand games is *minimum* if you want to distinguish between versions which
are close in strength. And that while rigorously controling testing conditions.
Up there (for engines with Pharaon's strength), when you fight for 10 or 20 Elo
improvements, a few hundred games simply are not enough...

Best regards - Munjong.



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

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