Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: General Objection Against CEGT Stats

Author: Heinz van Kempen

Date: 06:08:14 12/07/05

Go up one level in this thread


<<If you take Kasparov. You let him play 500 opponents of 1500-2200 Elo.
Normally
he gets 100%! From his strength. But due to chance and other factors that are
more or less irrelevant he gets "only" 95%.

Also: my relativation was NOT meant this way that a reasonable number of amateur
programs would come close to SHREDDER or FRITZ. What I meant was that a few
could make a reasonable match. But my argument is totally ignored that if you
take these other 495 programs who are absolutely out of any reach, that they are
breaking the testing result importance through their irrelevance.

I thought that it was clear that we discussed chess strength and NOT the
stability of the engines over a longer testing from the mere technical view.
I dont know how to make it clearer. If normally you expect 100% results, 22-0,
then it is no abberation if you get 19-3 due to hardware failure or such mere
artefakts how we call it in stats.>>

Hi Rolf,

as always you are putting it to the extreme :-). In CEGT we care for giving
games only against opponents which ratings do not differ by more than +-250 ELO
points. You might find very few games not fitting to this standards, but this
are exceptions not having any effect with so many games overall.

Anyway a valuable comment from your side giving us some stuff to discuss in our
private forum. Thanks.

Best Regards
Heinz



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.