Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: piece mobility?

Author: Tord Romstad

Date: 14:15:40 01/08/04

Go up one level in this thread


On January 08, 2004 at 17:00:37, Steve Maughan wrote:

>Tord,
>
>>I don't think it is.  Unfortunately I don't have many other data points,
>>but if I recall correctly Rebel searches more than 1 million nodes per
>>second in a typical middle game, and of course it does much more than
>>just computing attack tables at every node.
>>
>>My guess is that my generating speed is well below average.
>
>Remember that Rebel does not compute the attack tables at all of the leaf nodes
>since it does a form of lazy eval.  Leaf nodes form the bulk of nodes and since
>Ed reports a x2.6 speed-up when he applies the lazy eval I would estimate that
>only ~40% of Rebel's nodes are fully evaluated.  Of course this is only an
>estimate and Ed would need to confirm.

This sounds really strange.  If I understand Ed's pages correctly, he doesn't do
what the rest of us calls lazy eval at all.  What he describes as lazy eval is
what is
more commonly known as futility pruning in the qsearch (i.e. not searching
captures
which are not likely to bring the score anywhere near alpha), which is done by
almost all chess programs I know, and which is an entirely different idea than
lazy eval.  It also doesn't affect the NPS count at all, because the moves in
question
are never executed, and the corresponding nodes are never visited.

I suppose that Ed means that he needs (on average) 2.6 times more nodes to
complete a given search depth when he disables the qsearch futility pruning.
If this interpretation is wrong, and Rebel's nps count really increases by a
factor
of 2.6 when the "lazy eval" is applied, I must have misunderstood the
description
on Ed's pages completely.  It would be nice if Ed could clarify this.

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.