Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Question on Search Windows

Author: Peter Fendrich

Date: 11:46:42 02/06/99

Go up one level in this thread


On February 06, 1999 at 11:02:59, William Bryant wrote:

>After a depth of 4, I tried using the alpha and beta returned from the
>previous iteration as the initial alpha and beta for the next depth search
>
>This generated a speed increase (I assume for generating early cutoffs).
>
>Is this wise or are there pitfalls I haven't seen yet.
>
>Would it be better to widen the search window by a pawn or so by subtracting
>100 from alpha and adding 100 to beta before the next iteration?
>
>Is there a better (read safter and/or more efficient) way of doing this?
>
>Thank you in advance.
>
>William
>wbryant@ix.netcom.com

In principle I think it's a good idea but it can be done with variations.

I do it like this, in the beginning of each new iteration:
 alfa = eval - alfaMargin;
 beta = eval + betaMargin;

alfaMargin and betaMargin, depending on various factors, are set to about 40
where pawn is 100.
If the search returns a value outside alpha/beta, I have to re-search with a
wider interval. Even here I take a chance with a window narower than the maximal
window with risk for one more re-search.
If it's the first move in the iteration, in some situations, I don't re-search
that move if it's below alpha. Insead I hope for the next move to match the
alpha/beta window. The risk, of cource is more re-search...

I think it all depends of your program behaviour so the only way to know is to
test...
//Peter



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.