🚀 go-pugleaf

"Mike S." wrote:
> 
> "Charles Blair" <c-blair@uiuc.edu> schrieb:
> 
> >    Has anyone done any comparison of quality of program analysis
> > of endgame studies (not including database searchers in this) or
> > crucial moments from famous games?

Problem is programmer with endgame table bases may not have
focused on end game features in their code as much, so the
results may distort.

No idea how big a factor this is, but GNU chess has a KP v K
ending routine, if it used tablebases no one would have bothered
implementing this, but it can quickly evaluate KPvK which is a
big plus in simple endgame analysis.

> There are many test suites with such positions (most are not
> endgame-specific though; some are splitted in thematic parts):
> 
> http://www.andreas-schwartmann.de/chess.html#Test

Some of these tests also focus on ideas. One I recall has a
Bishop sac to achieve K, wrong Bishop and doubled rook pawn,
against K. Assuming it doesn't have time to search out the draw
a program either knows this is drawn or it doesn't, many
programs only know the single pawn variant. Again skill on this
test may not relate to significantly improved playing strength.

> I read constantly that test positions are
> proven to be wrong, IOW. the intended solutions turn out to be doubtful
> sometimes.

For as long as you check out the tests machines repeatedly get
wrong at increasing time controls, this isn't usually a problem,
although there are still copies of "Win At Chess" with errors in
lurking on the Internet.

Errors in the test suites affect all programs roughly evenly.

-- 
Are you using the Internet to best effect ? www.eighth-layer.com
Tel: +44(0)1395 232769      ICQ: 116952768
Moderated discussion of teleworking at news:uk.business.telework