Author: Sune Larsson
Date: 12:55:43 12/29/05
Go up one level in this thread
On December 29, 2005 at 15:46:58, Sune Larsson wrote:
>On December 29, 2005 at 15:14:16, Joseph Ciarrochi wrote:
>
>>Thanks for all your work on this. This test suite certainly produces interesting
>>positions. So far i think the suite is good for illustating key endgame themes
>>for us humans, and, it can diagnos some flaws in engine evaluation (as you
>>illustrated in earlier post with rybka).
>>
>>I think the simple test of the sensitivity of the suite will be to pit the
>>current rybka against the future rybka that has endgame knowledge . We should
>>see the future rybka 1.2 do quite well against its old self
>>
>>If people where willing to help out with using the testsuite, we could get the
>>numbers up to do sensitivity analysis (similar to test analysis in psychology).
>>This would involve looking at how success on some items tend to cluster with
>>success on other items (e.g. maybe 3 or four items are good at picking up
>>misevaluation in pawn structure). We could also determine whether some items
>>are insensitive to endgame engine differences (i.e., success on those items will
>>not correlate with success on other endgame items, or with engine strength based
>>on cegt)
>>
>>anyway, let's keep it going.
>>
>>best
>>joseph
>
>
> Yup, I share your opinion. There is a lot of interesting information to get
> from these games. Contrary to ordinary eng-eng matches *the games* will be
> studied - not only the numbers. Furthermore they just might influence the
> coming February-Rybka...
>
> I think some changes will still happen with the 20 positions. First 9 are
> kept since they are Nunn...but thanks to your games we can spot positions
> where one side "always" wins. In danger is the position I posted - where
> Rybka lost in a few moves. The best way is 1.Be6+! - giving up a pawn - but
> getting a highway for the king. None so far plays this move - but since black
> plays bad and white has an advantage anyway - the games so far has ended with
> white wins. I think that position is rather doubtful.
>
> /S
Just noticed that Rybka only got a draw as white in this position....
>
>
>
>
>>
>>
>>best
>>Joseph
>>
>>On December 29, 2005 at 13:27:23, Sune Larsson wrote:
>>
>>>Position 17 was removed and a new one is added - to make the total of 20 pos.
>>>The first 9 are from Nunn - the following 11 are my own adds.
>>>
>>>8/1ppr1kp1/p1p4p/8/8/5P2/PPP1RKPP/8 w - - 0 1
>>>8/1pk3pp/p7/3p1p2/3K4/6P1/PP2PP1P/8 b - - 0 1
>>>8/4k3/r4pp1/pR6/P4KP1/5P2/8/8 w - - 0 1
>>>8/p1p1k3/1p1p2p1/3Pq2p/2P1P3/1P3QPK/P7/8 w - - 0 1
>>>8/p3ppk1/1p4pp/3nN3/6PP/1P3P2/P3PK2/8 w - - 0 1
>>>7k/pp4pp/2n5/8/8/P7/1P4PP/2K1B3 b - - 0 1
>>>8/8/p4Bp1/1pPb2P1/1P2kp2/P7/5K2/8 w - - 0 1
>>>8/4k3/p1B4p/2K5/1P4bP/8/8/8 b - - 0 1
>>>3R4/1p6/2b5/2P1k2p/p3p2P/P6r/1P2KB2/8 b - - 0 1
>>>
>>>
>>>8/5b2/1k2p1p1/2NpP1P1/1K1P4/8/8/8 w - - 0 1
>>>8/5p1k/r5pp/P7/3R3P/6P1/5PK1/8 w - - 0 1
>>>4b3/5k2/4p1p1/3pP2p/2pP1P1P/2P5/6N1/2K5 w - - 0 1
>>>8/1p1b2k1/p1p1p1p1/2P1P2p/1P3P1P/P2B4/5K2/8 w - - 0 1
>>>6k1/2p3np/1p1p2p1/3P4/1PPK1R2/6PB/7P/4r3 w - - 0 1
>>>1n6/4k2p/p3ppp1/1pPp4/3P1PP1/3NP3/P3K2P/8 w - - 0 1
>>>3r4/7p/Rp4k1/5p2/4p3/2P5/PP3P1P/5K2 b - - 0 1
>>>6k1/p2b1ppp/8/8/3N4/1P5P/5PP1/6K1 b - - 0 1
>>>2r5/1p1k1p2/p2p2p1/P1PPnp1p/1P1K1N1P/5P2/4R1P1/8 w - - 0 1
>>>8/6b1/4k2p/p1p1p1p1/2P1K1P1/PP3N1P/8/8 w - - 0 1
>>>2r3k1/4pp1p/3p2p1/p2R4/1r6/1P4P1/P3PP1P/R5K1 b - - 0 1
>>>
>>>The new one is an old celebrated ending Stoltz-Kashdan, 1928. It shows the
>>>bishop's superiority over the knight, in an ending with pawns on both wings.
>>>It's considered won for black. Fairly basic - but the test is to find out
>>>whether the engines can handle these basic endings well or not.
>>>
>>>
>>>[D]6k1/p2b1ppp/8/8/3N4/1P5P/5PP1/6K1 b - - 0 1
>>>
>>>
>>>/S
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.