Computer Chess Club Archives


Search

Terms

Messages

Subject: Play against Fallible Opponents

Author: GuyHaworth

Date: 09:15:11 12/03/02

Go up one level in this thread



I have described 'Reference Endgame Play', defining a spectrum of Reference
(Fallible) Endgame Players.

This was presented at the 7th Computer Olympiad Workshop, Maastricht, 2002. A
revised and generalised presentation of the concept is targetted to be in the
ICGA Journal, December 2002.

The basic idea is that an REP is less likely to play a move that is worse in a
value/depth sense. Some REPs are less likely to play that move than other REPs.



The idea generalises to non-endgame play.  One might say that the opponent is
less likely to play a move that rates worse according to your evaluation
function - but as your evaluation functions do not agree, the chances are indeed
non-zero.  Different 'opponents' will be less likely than others to play a
'sub-optimal' move.

With Bayesian inference, you can assign probabilities to the different
'opponents' ... and thereby come to an expected value for a position rather than
just its absolute minimax value.  This may cause you to choose other than the
minimax-optimal move on occasion.

g



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.