Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: ProbCut: An Effective Selective Extension of the Alpha-Beta Algorith

Author: Tord Romstad

Date: 08:06:34 07/22/04

Go up one level in this thread


On July 22, 2004 at 09:57:37, martin fierz wrote:

>probcut will work on the majority of positions where one side has bludered
>material. to get it working for the cases where one side sacced material for a
>deadly attack is going to be the problem!

I don't use probcut, but I sometimes reduce the search depth in positions
when it seems clear that one side is winning (particulary in the endgame,
where the null move tends to be less effective than in earlier phases of
the game).  The problem you describe above is not impossible to cope with,
at least if you evaluate internal nodes.

The idea is to look at the material *and* the positional evaluation of
the position, and perhaps to subdivide the positional evaluation into
its various components.  If one side is clearly ahead in material, and
at the same time has an advantage in most positional factors, it is
usually safe to reduce the search depth or (in extreme cases) prune
the entire subtree.  In simple endgames, this can often be done almost
without any risk.  It is more difficult in the middle game, but it is
still possible to make it work.

A related trick which I consider interesting is what I call "compensation
extensions".  These are used when one side is clearly ahead in material,
but the other side has a huge positional advantage (usually in the form
of a dangerous attack or advanced passed pawns).  In such positions, the
static eval tends to be highly unreliable.  It makes sense to extend the
search in an attempt to find out what happens when the dust settles.

Tord



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.