Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Principal variation search question.

Author: Daniel Mehrmannn

Date: 05:50:25 05/10/05

Go up one level in this thread


On May 10, 2005 at 07:47:53, Richard Pijl wrote:

>
>>Most of the time they do the same job as 80% of the time
>>the best move that causes a cut of is the first.
>>It seems logical for me to wait until you found a PV move before
>>applying null windows rather than starting it after first move is searched.
>>daniel
>
>I don't see this logic, but, at the same time, must confess that I haven't
>tested the differences thoroughly yet.
>
>Normally a scout is used to be able to quickly discard moves.
>There is little use in doing that for the first move you try, as it is likely to
>yield a result within the a-b window. After the first move gave a nice value,
>scouts are used to test if there is a better move available. In most cases there
>probably isn't.
>In cases that the first move didn't give a nice pv value and failed low, we
>basically have the same situation. Chances are that the second and third move
>will fail low as well. So why not quickly find out with a scout whether there is
>a move that gives a value within the a-b window instead of searching all moves
>with an a-b window?
>Richard.
>Richard.

Well, i tryed this idea months ago. The result was not very good. My debug shows
mich that the first move, expect PV or hashMove, is not easy the best move.

So i had many researches that cost a lot of time.

My idea was allso if we have searched all captures and killer moves and still no
move above alpha, we use yet our bestmove/result  as new upperbound and try a
pvs, if it failed we research with the original window again.

This method works but paysoff not so much as i thought.

daniel





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.