Author: Rainer Neuhäusler
Date: 07:56:58 05/01/01
Go up one level in this thread
Well done, Kurt! I agree, that it might be an inefficient method to let play different Chessmaster personalities against each other. In this way someone will never find the right standard settings, maybe. Seen statistical it is a matter of a restricted variance. Despite of different personalities, King remains King and you are testing the algorithms against themselves. It could be you find with personality matches the best settings for an “Indoor-Champion” but against the “Outdoor-Engines” he is an absolute loser :-( But about this the scientists may discuss (if there is which one here?). There is a simple empirical way to check it out and everyone can do it like you have done, Kurt. Mr. Merlino should not test in a few hundred games Chessmaster against another personalities but against the top engines on tournament level. This is named “Validity” in the theory of testing. I have some other critical remarks about the Utzinger results. They concern the book learning and the position learning and the use of tablebases. Engines profit from these abilities according to the number of games which are played (the increase probably might not be linear). Therefore it could be, that 10 games are not enough for a significant effect. In other words, in the games number 11 to 20 (if ever played) such progs could score probably better which take advantage of these learning modes and the tablebases. On the other hand the games 1 to 10 are ev. more represantative for the pure strength of the engine. It's not easy the testing-job ! In the end I hope we will find many new things realized in the Chessmaster 9000 in the sense of a serious Tournament-Engine. Regards Rainer
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.