approximate eval function

[Joel Rivat]

In a previous post Bob Hyatt mentionned I did some testing that showed
that only a few positions need full eval...
I just want to point out that this idea is not new, and appeared at
least in the Slate/Atkin paper in Chess Skill in Man and Machine
(P.W. Frey, Editor) Springer-Verlag...
(The MINI MODULE, D. End-point evaluation...)

F-H Hsu also mentionned recently that only about 20% of the positions
need full eval, and I came to the same conclusion...

Joel Rivat

[Robert Hyatt]

In article <43ogj6$c...@cismsun.univ-lyon1.fr>,
Joel Rivat <ri...@caissa.univ-lyon1.fr> wrote:
-->In a previous post Bob Hyatt mentionned I did some testing that showed
-->that only a few positions need full eval...
-->I just want to point out that this idea is not new, and appeared at
-->least in the Slate/Atkin paper in Chess Skill in Man and Machine
-->(P.W. Frey, Editor) Springer-Verlag...
-->(The MINI MODULE, D. End-point evaluation...)
-->
-->F-H Hsu also mentionned recently that only about 20% of the positions
-->need full eval, and I came to the same conclusion...
-->
-->Joel Rivat

I wasn't intending to slight anyone with my comments. Joel and I chat
regularly, and while I had the "staged" eval on my to-do list, I had not
reached a point where I had started on it. His "numbers" got my interest
up, since crafty was spending over 30% of search time in eval. I was doing
the "cutout" in front of the call to eval, but not the cutout in the
middle of eval. I simply relied on his test data, and decided to play
with eval. The savings were remarkable (eval is under 10% now, with *no*
knowledge removed.)

Thanks, Joel...

--
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