Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Opponent-modeling in computer chess

Author: Reinhard Scharnagl

Date: 23:38:32 07/14/05

Go up one level in this thread


On July 15, 2005 at 01:31:39, Tony Werten wrote:

>Opponent modelling assumes using 2 evaluation functions. Yours (the "correct"
>one ) and your opponents ( the "wrong" one).
>
>The point is maximizing your score (the best position) while also maximizing
>your opponents one (the most likely continuation for your opponent).
>Alpha beta (negamax etc) will fail to do this since it tries to minimize one,
>wich you need to "correct" your maximizing score.
>
>When using only 1 evaluation function you will not be able to notice wether the
>score comes from the weaknesses of your opponents evaluation.

Hi Tony,

you almost got the point despite of naming the evaluation pair 'correct'
and 'wrong'. The first would be able to overrule the second in some way. Instead
it would be more realistic to have two concurrent functions without priorizing
one of both. This assumption is related to the (actual) unability to point to
the 'best' move in every position. Otherwise the 'best' chess game ever would
have been played already and chess would have been solved.

The minimax approach will be replaced by an attempt a) to maximize at own
decision points and b) to optimal predict in opponents decision points. Within
that evaluations have to be recalculated and could not be lifted by simply
negating.

Reinhard.



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.