Author: Miguel A. Ballicora
Date: 14:59:20 05/06/02
Go up one level in this thread
On May 06, 2002 at 16:12:40, Dieter Buerssner wrote:
>On May 06, 2002 at 13:59:40, Miguel A. Ballicora wrote:
>
>>In other words, a test like this might introduce some noise and should not be
>>present in a very high quality test suite, IMHO.
>
>I think, this is an interesting point. And I agree - almost. However, I would
>also find a testsuite with imperfections, perhaps even extremely, interesting
>(and maybe of "high quality"). The positions should be not too difficult then.
>Why not give the engine the chance to be lucky sometimes? Or test its
>anticipation. It would need many positions (so the noise will average out).
>Perhaps, this would be much more helpful to test positional (or even material,
>when unequal material) scores. I think, there will be many positions, where
>(depending on the search time) many engines can find the correct move - without
>really understanding (=seeing the tactics) it. If they can win from it later -
>why not? Of course, such a test suite result would be more difficult to
>interprete. And perhaps totally unsuitable for comparing engines. Also, as with
>other testsuites, a danger may be, that high positional scores/high value of
>pawn can help. It should also have ("am") positions, where the typical sacs look
>interesting, but are wrong. I think, all those imperfections, could make it
>still very valuable (at least for some engine authors). So, "right move, wrong
>reason": if you have enough of those, and if you can (and will) try it out, may
>still be very helpful. But it would need rather many positions.
Yes, I guess that it always depend on what you want to measure. The problem
could be practical, the more positions needed to get a reliable answer the more
time for running the test, unless it is run with no much time per position.
For tactical tests where there is a concrete combination it could be possible to
select positions where the answer is pretty clear.
Regards,
Miguel
>
>Regards,
>Dieter
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.