Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Probability of nullmove fails

Author: Tord Romstad

Date: 13:54:31 01/17/04

Go up one level in this thread


On January 17, 2004 at 16:28:15, Klaus Friedel wrote:

>Now the question:
>What are characteristic attributes of posiitions in which the null search is
>more likely to fail with error 2 (besides zugzwang positions).

Dieter's example is a good one.  In my experience, there are three main problems
with
null move pruning.  The first one is, of course, zugzwang positions.  The second
is the
horizon effect, as illustrated by Dieter's example.  The third is that it tends
to prune away
maneuvers where a piece is temporarily moved to a bad square on its way to a
very
good square.

In order to solve the first two problems, I use the evaluation function to
decide whether
or not to use null move pruning, and also to choose the null move reduction
factor.  When
the number of safe legal moves (that is, moves which do not lose material) is
low, or there
is very little material left, the risk of zugzwang is big, and it is best to
avoid the null move.
When the king safety for the side to move is bad, when the opponent has a one or
more
dangerous-looking advanced pawns, or if the side to move has a pinned or trapped
piece,
I reduce the reduction factor by 1.

I don't yet do anything to cope with the third problem.  Sergei Markoff
(SmarThink) makes
use of one of Botvinnik's ideas, called "trajectories".  I am not sure precisely
what this means,
but as far as I have understood he does some kind of horizontal search within
the evaluation
function to identify paths a piece can use to safely reach some advantageous
square, and
avoids null move pruning at nodes directly after a move which is part of such a
path.  The
idea certainly seems interesting, and I'll probably give it a try sooner or
later.

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.