Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Question about prescan and evaluating in Crafty

23 views
Skip to first unread message

Willem Broekema

unread,
Jan 7, 1998, 3:00:00 AM1/7/98
to

I have a question about the use of a 'prescan' (when evaluating
positions and searching for the best line) in Crafty .

I have read that programs like Genius and Fritz and Nimzo first analyse
the current position, then make some modifications in the evaluation
function (for example, depending on the pawn structure the optimal
fields for the pieces are changed). Moves that seem to be good
accordingly to that new evaluation get a bonus.
After that, they start searching moves.

That way, a lot of work is done before any move is searched, so the
evaluation function can be much faster.

As far as I have looked at the source of Crafty, it doesn't seem to have
such a prescan method.

I would like to know if you have considered such a prescan funtion.


Willem Broekema
(w.o.r.b...@rcondw.rug.nl)

Robert Hyatt

unread,
Jan 7, 1998, 3:00:00 AM1/7/98
to

Willem Broekema <w.o.r.b...@rcondw.rug.nl> wrote:
: I have a question about the use of a 'prescan' (when evaluating

I considered it, but have not used it. The problem: you set things
up for the board, as it is now, but with today's deep searches, you
are evaluating endpoints that are *far* removed from the initial
position. IE imagine you find black's king exposed and sitting on g8,
with no pawns around it, so you set up piece/square tables to attract
all of white's pieces toward g8. And suppose there's nothing to prevent
black from running the king over to b8, and then starting some tactical
action of its own. Meanwhile, all your pieces are congregating around
empty space, thinking they are piling up on a king which is now long-
gone.

Most prefer full endpoint evaluation so you aren't caught in this
trap. but it does cost, since anything done at the root is only done
once, while things done at the tips are dont about 100K times per
second...

I go for the 'accurate" approach..


: Willem Broekema
: (w.o.r.b...@rcondw.rug.nl)

--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170

0 new messages