Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Attack Tables

Author: Vincent Diepeveen

Date: 09:39:45 03/01/01

Go up one level in this thread


On March 01, 2001 at 10:32:42, Jon Dart wrote:

I think it has to do with your future plans in computerchess
programming. If you plan to improve your evaluation a bit, especially
with expensive things like scanning board in all kind of ways
like piece activity, mobility, center, scanning files and lines
and all kind of similar things like scans around the king, then
you sure should consider it.

If your goal is to outsearch others with a smaller set of knowledge
(or a similar set of knowledge) then assembly and getting 1.3M nodes
a second might be an option. Not using incremental attacktables
of course then as that would slow down your 1.3M nodes a second
considerably (though it would reduce tree size you search).

Using attacktables incremental sure means you are using attacks
in evaluation. If you don't use attacks in evaluation i would
consider generating attacks usually also a waste of time, either
incremental or nonincremental.

For me doing things incremental is faster as i don't need to
check whether my king is in check. I can do all kind of conditions
for extensions without recalculating attacks in between, i can
do whatever i want to, not in the last place to forget the
evaluation!

>I used to do this but don't anymore. It was a net speed gain taking it out and
>calculating attacks when needed.
>
>In my experience, you will call MakeMove/UndoMove in a lot of cases in which you
>wind up not using the attack info, or not using it enough to justify the
>incremental cost of updates on each call.
>
>But your mileage may vary.
>
>--Jon



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.