Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Zotron LIVES!

Author: Rémi Coulom

Date: 08:38:23 12/11/00

Go up one level in this thread


On December 11, 2000 at 09:54:35, Pat King wrote:

>Zotron 2000 is finally up and running, and losing regularly on fics under the
>account "Zotron". Thanks to everyone for your comments over the months.
>
>Note to fellow newbie programmers: It was recently suggested to me to add a
>random factor to the eval to make Zotron more interesting while I improved the
>eval. If you use MTD (or, I assume, PVS) this is a BAD idea. The unrepeatable
>eval triggers many useless researches. I achieved the same effect with a random
>sort of the move list.
>
>Pat

The right way to add randomness to the play of a chess engine is not to add a
random value at each call of the evaluate() function. I guess this is what you
did and you noticed well that it is a very bad idea.

To add randomness, the randomness must be made consistent. I did it in my
program by adding a random value to piece/square tables at the root of the
search. This way, the random evaluation does not vary during the search. It is
kept constant during a game. The general idea is that if you call the evaluate()
function during a search, it should return the same evaluation if it is in the
same position.

Remi



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.