Computer Chess Club Archives


Search

Terms

Messages

Subject: Duplicate games

Author: Bruce Moreland

Date: 11:27:05 03/16/98

Go up one level in this thread



On March 16, 1998 at 13:48:00, Ed Schröder wrote:

>Get rid of double games. What do they contribute to comp-comp games?
>Nothing in my opinion.

Assume we have program A, which will play strong chess and learn from
its mistakes.  If program A figures out that it scores 100% against you
with 1. ... c5, 50% with 1. ... e5, and 0% with 1. ... e6, it will play
1. ... c5 every time or almost every time, and will rarely or never play
1. ... e6.

Assume program B just plays strong chess.  It will play 1. ... c5, 1.
... e5, and 1. ... e6 randomly, regardless of result against you.

Program A is obvioiusly a more challenging opponent.  The programmer of
program A has spent time adding a learning feature that is effective at
playing against the weaknesses of specific opponents, be they computer
or human, and this is a perfectly honest thing to do, a legitimate
increase in strength.

Obviously program A deserves to be higher on any rating list.

We saw Kasparov do this against Anand in the last PCA world
championship, he played the Sicilian Dragon a couple of times, and
nobody every said this was unfair.  It was obviously up to Anand to
realize that he should find an improvement or vary.

This is distinct from killer book discussions.  I don't believe that
program A should play 1000 games against program B, learning the whole
time, then face a fresh install of program B, who has never seen program
A.

It's also distinct from discussions about SSDF testing methods.  One
game each on 20 machines is obviously much different than 20 games on
one machine.

bruce



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.