Re: Question

[Simon Waters]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Martin Mačok wrote:
> On Tue, Apr 15, 2003 at 01:43:50PM -0400, Sternbach, William [IT] wrote:
>
> I think that the most easier way to improve GNU Chess is to do low
> level optimizations and/or implement better positional evaluation.

Low level optimisation may be easier as you don't have to reinvent stuff
but it doesn't offer a huge potential for gain I suspect comparing the
nodes per second with programs like Crafty, sure we can do a lot better
but it is a lot of coding for little gain.

The evaluation function could do with an overhaul.

Comet from a shared code base is substantially stronger, and that is due
mainly to improvements to the search. This would be refinement of Null
Move, and extensions, and pruning, rather than a substantive change to
the method used from what I am told.

It's long passed master level at Blitz on even very old PC's, I presume
blitz on modern PC's would be at or around Grandmaster level. We don't
get enough games at normal time controls against human chess masters to
make a reasonable comparison.

Pragmatically I suspect there is at least one bug in GNU Chess 5.06 wrt
 pondering, as the rating gain from pondering was too small, to
non-existent, despite it clearly remembering lines between moves correctly.

On this basis debugging pondering further, and minor enhancements to the
search algorithmn will give the most bang per line of code in the short
term.

I've still to beat the pondering version in a fair game (book and
reasonable hashsize) on a very old PC, so for me strength is not such a
priority any more. This is the first version I didn't beat at least once
in prerelease testing.

Some of the recent endgame eval function changes allow GNU Chess to
handle special cases of endgames, these typically make it marginally
weaker, but more useful when analysing endings.
-----BEGIN PGP SIGNATURE-----
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQE+nJQ6GFXfHI9FVgYRApilAJ98oSuzq/KL05OsfFLZmb0lr6PLLACgzzDa
0DoL4xar2Ur2pmsebiXp7qs=
=wdFw
-----END PGP SIGNATURE-----