Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Multiple Linear Regression - Secret of the Commercial Chess Programmers

Author: Steve Maughan

Date: 05:22:49 04/30/04

Go up one level in this thread


Rasjid,

All strong Othello programs, mine included (Edge Reversi), use regression
models.  This is due to the fact that Othello is close to 100% positional.
Chess, on the other hand, is 99% tactics - this is the problem.

When I'm not playing chess I am usually creating commercial econometric models
so this is something I am know quite a bit about.  IMNSHO it will be difficult
to get the regression approach to overcome the tactical component of chess.
What is likely to happen is the models will underestimate the value of the
parameters due to tactical 'noise'.  Another problem is non-linearity -
multiple-LINEAR-regression cannot cope (easily) with non-linearity e.g. penalty
for squares around the King controlled by the opponent; if it's one the penalty
might be -0.1 but if it's eight it should be much higher than -0.8!

This is one approach I was going to experiement with but I certainly don't think
it is the secret sauce for m(any) commercial engines.

Regards,

Steve



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.