Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: interior recognition question?

Author: Robert Hyatt

Date: 11:11:56 01/26/04

Go up one level in this thread


On January 26, 2004 at 12:06:53, Daniel Shawul wrote:

>Hi all
>
>I changed my endgame eval by interior node recognizers
>but I can't implement the simplest KBNKrecog(which the author gives example for)
>correctly. I did everything according to Ernst Heinz's paper but it is not
>able to find a KBNK mate now. When the bonus for KBNK is large
>it will find the mate,like Slate's bound method.But since that has
>its own problems,i didn't want to do that.
>
>No recognition is done in quiescence
>
>eval when bishop is dark is as follows
>
>         eval=Material[side]
>  	      +300
>	      -13*distance[kingsq[white]][kingsq[black]]
>	      -25*dark_corner_distance[kingsq[opp]];
>


That looks a bit strange to me.  IE the +300.  Shouldn't that be for the side
with the bishop/knight, not always positive?  IE with the above, assuming white
is winning, then a score would be +600 (assuming the material score is +600 for
having a bishop and knight and the opponent having nothing) +300 less a penalty
for your king being too far away, less another penalty for the opponent king
being in the wrong corner.  IE +900 - a couple of dings.

If black is winning, you get -600 + 300 - a couple of dings thare are actually
bonuses for black if the white king is in the wrong corner and if the black king
is too far away.

Your eval seems to mix "side to move scoring" with absolute +=good for white
scoring, which is going to produce odd results.




>         where distances range from 0 to 7.
>
>The eval returnded ranges from 7 - 9 which is closer to the static eval.
>The recognizer works when eval is a draw(exact score).example_ my KBPKrecog
>works well.
>
>What is causing the problem? what are the things i need to take care of
>when writing recognizers? Any suggestions are welcome
>
>regards
>daniel



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.