Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Programs with Positional Understanding

Author: Dann Corbit

Date: 12:51:16 11/14/01

Go up one level in this thread


On November 14, 2001 at 15:22:08, Nolan Denson wrote:

>There has been talk around this forum about programs with positional
>understanding.  Do they really understand how to make a positional sack with no
>material advantage.  Well i decided to try and tweak Dr.Hyatt's Crafty to play
>just that way.  It has played some very interesting games ... the evals are
>totally different than any other program I have seen to date.  I have notice
>that its better at longer time controls so this game I am about to share with
>you is 5 min with 5 sec increments.  There were times when the other program was
>up by as much as 1.50 according to its eval .. but never from my side looking at
>crafty did it think it was down. I am not sure if Yace played all the correct
>moves but the game was very interesting the way crafty showed great
>understanding of the position.  The rating of TombRaider doesn't reflect its
>true strenght .. because of all the tweaking that I am doing.

That sort of thing happens all the time.  I have seen even stranger things (e.g.
two different programs playing against each other and *both* programs think they
are up a full piece.)

Indeed, for some positions, crafty has the best positional understanding around.
The same is true for Yace.
The same is true for each of the commercial programs.
All successful programs have some positional understanding.  A material only
eval simply does not work.

I don't think you can conclude that Crafty has superior positional understanding
from the data you have displayed.  It may or may not be true.  All the top
programs do some things very well, and other things not quite as well.

When a computer program can navigate through the Evans Gambit again and again
without an opening book or learning file successfully, then I will imagine that
they have very good positional understanding.  A 100% score at LCT II would also
go a long way in that direction.



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.