Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Huge quiescence tnode counts

Author: Robert Hyatt

Date: 07:44:03 06/26/99

Go up one level in this thread


On June 25, 1999 at 22:02:45, James Robertson wrote:

>My program typically spends 2/3 (or more) of it's nodes in the quiescence
>search. Since this is the only place I call my evaluation function, it means
>that it is spending more than 2/3 of the total time in the quiescene search. I
>even have SEE and delta pruning (a la EXchess).... I am really looking for and
>hoping for ways to reduce the percentage of my search spent quiesceing. I looked
>at EXchess, and from the node counts it displays, it looks as if it only spends
>about 10% quiescing! How does Dan do this? Are there any useful pruning
>techniques? How much time does Crafty spend in the quiescence search?
>
>Thanks for all help!
>James

2 questions:

1.  how are you counting quiescent nodes?  ie for _every_ node you search
at the last full ply, you will have to call quiesce, yet this is not optional
and will make it impossible to get your count down to 10%.  Generally we call
those q-nodes 'leaf' nodes and then anything _after_ them we call q-nodes...

2.  are you tossing out loser captures?  are you tossing out captures that don't
bring the material close enough to alpha to make it worth searching?  IE down a
queen, then taking a pawn isn't worth the trouble...



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.