Re: GNU chess help

[Simon Waters]

Tim Bedding wrote:
> 
> I was wondering if there was anything I could do
> to help with the development of GNU chess.
> 
> I am thinking perhaps of improving algorithms
> to make a stronger program.

Contributions are always welcome. We currently work from the CVS
head version on Savannah towards a 5.05 release, although I've
kind of put 5.05 on hold till the PGN and related changes are
sorted.

Currently Lukas is working on substantive parts of input, and
related changes, so probably best to wait a few weeks before
braving those areas of the code.

David is sorting out the PGN handling code.


One idea lurking is to look at eval.c/eval.h with a view to
replacing the fixed weightings (eval.h) C macros with run time
configurable values. The idea is to offer a style of play, say
aggressive, or even patzer (you'd be amazed how many requests we
get to make it play worse!).

More importantly once the weightings are put into a more
flexible framework (one that doesn't require recompiling to
change them), it is possible to look at ways to automate the
tuning of such values.

Quite a lot has been written on automated tuning of such
weightings, but it is an area that should leave sufficient scope
for individual creativity.

Certainly even by manual inspection of the order in which moves
from root position are examined I could spot that the "pinned
Knight" penalty and the "Knight on the rim" penalty were out of
kilter, as moves like Na3, Nh3 were examined before moves like
Nc3, Nf3, implying the program rated such positions more highly.
So I think a system for automatically adjusting some of these
values could add significantly to playing strength.

The current eval code is pretty straight forward, start with
Evaluate in eval.c, the only gotcha is the lazy evaluation, and
the odd special case routine like KBNvK.

Anyway just an idea, if you have you own, or better ideas, I
suggest you post to bug-gnu-chess, before coding, so we can warn
you of any ongoing work in that area.