Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Recapture Extension

Author: Ulrich Tuerke

Date: 08:30:34 01/09/02

Go up one level in this thread


On January 09, 2002 at 10:15:53, David Rasmussen wrote:

>Some time ago we talked about the recapture extension. Bob's experiments with
>Crafty seemed to show that recapture extension was a bad idea at least for his
>program.

I don't know if the notion "the recapture extension" is really so well defined.
I think that there are several ways to support recapture extensions; I don't
know what Crafty does.

A long time ago, I have experimented a lot with these extensions.
Currently (IIRC), I extend when there is a capture and in the subsequent half
move there is another capture to the same square gaining the same material
(ignoring the difference between bishop and knight). I do this extension at most
once in a line.

This isn't too expensive and helps a bit in tactics.

I have also tried to make this extension independent on the square (i.e. just
re-gaining the same material one ply later), but I wasn't completely happy with
it.

Another idea is to define a recapture as 2 subsequent captures to the same
square bringing the score back into the aspiration window. I'm afraid that this
may be expensive at the start of an iteration where alpha-beta usually uses a
larger window. It could also produce some search anomalies because I'm not sure
if in all cases the condition for extension will be reproduced in the
verification search.

Anyway, this extension generates some path dependence, i.e. may occasionally be
missed by transposition table entries.

Just some simple ideas ...

Uli

>
>How many people have the recapture extension because they positively believe
>that it is better for their program, and how many has removed it from their
>program because the positively believe that it is bad for their program?
>
>I am sure a lot of people have it or don't have it, but for no specific reason,
>that is, they have not tested the effect of this extension thoroughly.
>
>/David



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.