Computer Chess Club Archives


Search

Terms

Messages

Subject: diminishing returns in computer checkers

Author: martin fierz

Date: 21:35:11 02/11/02


aloha!

there was just a discussion on "diminishing returns" in computer chess here, so
i finally wrote up an experiment i did a year or two ago with my checkers
program on my webpage: http://www.fierz.ch/strategy3.htm
in contrast to an experiment by junghanns and schaeffer (i think?), where they
used chinook to play series of 40 games against itself, i played 282 games per
match, which leads to more reliable statistics - i think the error bar for the
chinook values reported on the page above are about 5%, making the conclusions
of that experiment questionable.
it's much easier to find this kind of effect in computer checkers for two
reasons:
1) the program can search much deeper
2) the "natural length scale" of a checkers game is much shorter than for a
chess game. with this i mean that typically after a certain number of moves, the
program will hit the endgame database, and from there on it will not matter how
deep it searches. this leaves the deeper searching program with a decreasingly
short number of plies where it can use it's superiority.

my message is: this effect will show up in computer chess too - but it will take
a looong time.

cheers
  martin



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.