Winboard Forum

Hi,
I'm now going to introduce endgame knowledge to Alaric.

1) Do you have any good endgame test suites?

2) There was a discussion here or in CCC about "nunn positions" for endgame. I can't find it but do you remember the suggested positions?

Thanks
/Peter

Hi Peter,
I've posted all I have on my site at:

http://www.ascotti.org/programming/chess/chess.htm

I think there is enough to keep you entertained for a while!

Peter Fendrich wrote:2) There was a discussion here or in CCC about "nunn positions" for endgame. I can't find it but do you remember the suggested positions?

I don't, but if I recall correctly the suite consisted of only 10 positions, which is not really enough.

An idea I have considered for some time is to create a big collection of Nunn-like positions organized by different themes, with 20 positions or so for each theme. A few possible themes could be opposite castlings, closed centre, open centre, queenless middlegames, bishop vs knight, pawn majorities on opposite wings, and rook endgames. It is important that the chosen positions are relatively balanced, and offer reasonable winning chances for both sides. Playing matches from starting positions organized by themes seems like an effective way to identify the strengths and weaknesses of an engine, and to test changes in some specific component of the evaluation function.

If the positions are well chosen, I think such a collection of positions would be much more useful than traditional test suites.

Tord

Alessandro Scotti wrote:Hi Peter,
I've posted all I have on my site at:
http://www.ascotti.org/programming/chess/chess.htm
I think there is enough to keep you entertained for a while!

It sure is. Thanks!

An idea I have considered for some time is to create a big collection of Nunn-like positions organized by different themes, with 20 positions or so for each theme. A few possible themes could be opposite castlings, closed centre, open centre, queenless middlegames, bishop vs knight, pawn majorities on opposite wings, and rook endgames. It is important that the chosen positions are relatively balanced, and offer reasonable winning chances for both sides. Playing matches from starting positions organized by themes seems like an effective way to identify the strengths and weaknesses of an engine, and to test changes in some specific component of the evaluation function.

If the positions are well chosen, I think such a collection of positions would be much more useful than traditional test suites.

I agree - when will you start?
It's quite a work to put all these themes together with quality. Maybe if we were x volonteers who could take one theme each and present say 25 position. The next step would be to discuss the positions and exclude the 5 worst.
/Peter

Hi, I have some positions I've colletected myself, and I'm using them to test many different versions of my program.
I've started by the simplest: pawn endings and queen endings (which usually transpose to each other). Here are my positions:

Pawns:

Code: Select all

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "4k3/6p1/1pp4p/2p5/2P4P/8/PP5P/5K2 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/3k1p2/1p4pp/2p5/8/5P2/PPPK3P/8 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "2k5/1p3p2/p3p3/6pp/8/1P3PP1/P1P4P/3K4 b - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/5p1p/4p1p1/ppk5/5P2/PP2PP1P/3K4/8 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "6k1/1p3p1p/6p1/3Pp3/2P5/8/4K1PP/8 w - - 0 1"]
*


[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "4k3/8/1p6/p2p3p/P1pP3P/2P1K3/8/8 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "6k1/pp6/4p1p1/3pP2p/P2PpP1P/2P5/1P4K1/8 b - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/8/1p3kp1/5p1p/P2K3P/6P1/5P2/8 b - - 0 1"]
*


[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/8/p5p1/1p1p1k1p/1P3P1P/2K1P3/P7/8 w - - 0 1"]
*


[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/1p3p1k/6p1/6p1/Pp2P3/8/1P3PKP/8 b - - 0 1"]
*


[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/2p5/6p1/2k2p2/7P/6P1/K4P2/8 b - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "3k4/p5pp/P5p1/1p4P1/1P2P3/4KP2/8/8 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/8/2pk4/4p2p/p1PpP2P/6P1/PP3K2/8 b - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/pp3pp1/2p3kp/8/2P5/4P3/P4PPP/3K4 b - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/p4p1p/2p2p2/1k6/2p3P1/P3PP2/1K5P/8 b - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/p4kpp/3pp3/8/4PP2/3PK3/P6P/8 b - - 0 1"]
*


[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/6k1/4p2p/2Pp1pp1/8/4P1PK/5P1P/8 b - - 0 1"]
*


Queens:

Code: Select all

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "6k1/p4q1p/1pp3pQ/2p1p2P/P3P3/3P4/1PP3P1/6K1 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/p5kp/1P3p2/7Q/8/6P1/5P2/4q1K1 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "6k1/1p3p1p/3p2p1/3P4/2P5/8/3QK1PP/1q6 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/5p1k/6p1/pQ5p/3pqP1P/PP4P1/6K1/8 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "6k1/2p2p2/p1Q2p1p/8/P7/1P3P1P/2Pq2P1/5K2 w - - 0 1"]
*


[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "2k5/pp6/4p2p/1p2Pp2/Pq1P2pP/6P1/5PK1/3Q4 w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "1Q6/2K2p2/6p1/6k1/8/4P3/5P2/7q b - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/5pk1/2p3p1/Q1p1p3/2q1P3/5P1P/5PK1/8 b - - 0 1"]
*


[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "6k1/7p/1p2p3/pQq1P3/P3p3/7P/1P4P1/7K w - - 0 1"]
*


[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "6k1/p6p/5pp1/8/1P6/P2q2P1/5PKP/Q7 b - - 0 1"]
*


[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "8/5p1k/p2Q4/1p2p3/4P2P/2q2P2/P6P/7K w - - 0 1"]
*

[White "Test W"]
[Black "Test B"]
[Result "*"]
[FEN "6k1/p6p/2ppq3/6P1/1P5K/6P1/P2Q3P/8 w - - 0 1"]
*


I'm using 10 5 time control, and some well known programs as sparring. As Averno has quite a decent eval for these endings, I'm using strong programs like Shredder, Tao, Gandalf and some others. Next step is rook endings. I've some positions but need to collect some more.
Oh, I forgot: all the positions are taken from real games, and most of them from computer chess. Some are even, but some are not. I consider being able to win won games and defend lost positions as an important part of a program's strength.

Endgame testsuites can also be found at George Lyapko's site: http://www.geocities.com/lyapko/tests.htm
and at Peter McKenzie's site: http://homepages.caverock.net.nz/~peter/chess.htm

Did you find the Nunn endgame positions? I believe, that they were discussed at the German CSS forum. Perhaps you can ask there: http://f23.parsimony.net/forum50826/index.htm

Regards,
Dieter

Dieter B?r?ner wrote:Did you find the Nunn endgame positions? I believe, that they were discussed at the German CSS forum. Perhaps you can ask there: http://f23.parsimony.net/forum50826/index.htm

Hi Dieter, I searched through big databases with Scid to find these positions. It took me quite a while because I had to verify they were interesting positions (not dead draws or immediate wins).

Thanks Jos? for your positions.
The Nunn test can be found at this address:
http://www.geocities.com/CapeCanaveral/Launchpad/2640/pgn/tests/
(NET.ZIP)

Also, you can generate positions by removing some pieces and pawns from the starting position. That may not be representative of real endgames, but it still adds some variety to the nunn positions.

R?mi