Author: CLiebert
Date: 02:22:49 06/09/99
Go up one level in this thread
On June 08, 1999 at 11:54:13, James Robertson wrote: >On June 08, 1999 at 07:32:30, CLiebert wrote: > >Unfortunately, as I found out the hard way testing my program, 10, 20, even 30 >games is NOT enough to get a result with any kind of confidence.... :( I was >horrified to find that greater than 300 is needed for a reliable result.... > >James > On principle your are right. To my experience after 100 games against 4 or 5 opponents your have first nice trend. Not more. 200 games would be much better and everthing over 300 seems to be more stabil and reliable, I agree. But in our case the results against crafty 16.6 are always poor - the horror-opponent of lgg :-)) - so I was astonished about the first games. For now: it still looks good, lgg beta just beats Nimzo with the same result (and loses against Hiarcs4). Testings are going on... Another remark: the rating depends very much from the choice of opponents you take for a certain engine. It gets much better if you choose weaker onces. May be the problem with the ratings-results of cm6000 in sweden?! Christian
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.