Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Recapture Extensions

Author: Georg v. Zimmermann

Date: 02:17:33 06/02/01

Go up one level in this thread


On June 01, 2001 at 16:13:41, Scott Gasch wrote:

>Hi all,
>
>I have some questions about recapture extensions.
>
>First of all, what is the point?  I assume to extend in a position where a
>recapture move is the only reasonable move so as to see further beyond the
>forced exchange.  Somewhat like singular extensions or one legal reply
>extensions, I think a properly done recapture extension should not be too
>expensive.  I'd also expect that a well done recapture extension would improve
>the tactical strength of the engine.
>
>Well, I've been trying to get a good recapture extension working on my engine
>for a while now.  I've tried a bunch of different recapture definitions /
>conditions.  Most recently I did:
>
>  1. last move was a capture and this move captures the last moved piece
>  2. this is the first capture considered (weighed by SEE)
>  3. do not allow two recaptures in a row
>  4. this move restores the material balance to what it was at the root.
>
>...in which case I extended 1/2 ply.  Since I am expecting better tactical
>ability with the recapture turned on, I test it with the ECM suite.  Every time
>I try, my results are worse with the recapture turned on than with no recapture.
>
>I know that a bunch of other engines are doing recapture... so there must be a
>good idea there somewhere.  Can anyone address these questions or propose a
>solid recapture condition?
>
>Thanks,
>Scott

I do this :

  /* Capture extensions. */

  if ( CapturedPiece[ply-1] &&
     (ply > 1) &&
     (treemoves[ply-2].to() == treemoves[ply-1].to())  &&
     (!AIBoard.getattacks(AIBoard.getColorOffMove(),
     (treemoves[ply-1].to()) ) ) )

		{
			extensions += CAPTURE_EXTENSION;
                        /* [... some debug stuff ] */
		}

In other words :

If I capture to the same square and no other capture to that square was
possible, extend by half a ply.

This helps a lot in my program, for Crafty the last time I tried it solved one
WAC position more.

Regards,
Georg v. Zimmermann




This page took 0.01 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.