Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Learning failure in Gnadalf 4.32f??

Author: Mogens Larsen

Date: 17:54:59 11/19/00

Go up one level in this thread


On November 19, 2000 at 11:54:56, Tor Forl wrote:

>Hi everybody
>
>I have gandalf 4.32f playing on a chesserver. I do not have a super-pc(266MMX,
>48mb ram).. But I encountered a strange thing. I have learning set to 1. Below
>are two games against the same person. In the first game gandalf came bad out of
>the opening(-1.00) and was under lot of pressure.. The human made a mistake and
>resigned.. But the funny thing was that gandalf used the excact line as its
>openingbook in the next black-game, even it came worse out the first game, and
>eventually resigned.. I though that setting learning on would make the engine
>better and not jump into a bad opening, and not base the openinglearning on the
>outcome of earlier games(1-0, 1/2-1/2, 0-1)???

It's a combination of booklearning only and "remembering" a winning line/game.
There are advantages and disadvantages with an approach like that. The human
player knew that the opening line was good enough and repeated it. A computer
opponent probably wouldn't depending on the program.

Mogens.



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.