ShashChess

[Paloma]

Hi Vincent,

1r1q1r1k/5pp1/3p1b2/pp1BpPB1/1n2P1QP/1PN5/1PPK4/R7 w - - bm Ne2; id "corr LIPEAD40 (PER) , 10";
is a unknown good pos. (move 24)

[pgn][Event "LIPEAD40/f (PER)"]
[Site "corr ICCF"]
[Date "2016.09.30"]
[Round "?"]
[White "Los, Andrii"]
[Black "Kubicki, Tadeusz"]
[Result "1-0"]
[ECO "B90"]
[WhiteElo "2368"]
[BlackElo "2433"]
[PlyCount "97"]
[EventDate "2016.??.??"]
[SourceTitle "ICCF2018 re-edit"]
[Source "Tim Harding"]
[SourceDate "2020.02.05"]
[SourceVersion "2"]
[SourceVersionDate "2020.02.05"]
[SourceQuality "1"]

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be3 e5 7. Nb3 Be6 8. h3
Be7 9. f4 Nc6 10. f5 Bxb3 11. axb3 Nb4 12. g3 Qc8 13. Rh2 O-O 14. Bc4 b5 15.
Bd5 Rb8 16. g4 Qd8 17. Rg2 h6 18. g5 hxg5 19. h4 g4 20. Kd2 Kh8 21. Rxg4 Nxg4
22. Qxg4 Bf6 23. Bg5 a5 24. Ne2 Kh7 25. Ng3 Nxd5 26. Qh5+ Kg8 27. exd5 Be7 28.
Bh6 gxh6 29. Qxh6 Qb6 30. Ne2 f6 31. Qg6+ Kh8 32. c3 Qf2 33. Rg1 Qxg1 34. Qxg1
Rb7 35. Qg6 Bd8 36. Qh6+ Kg8 37. Ng3 Rh7 38. Qe3 Kh8 39. Ne4 Be7 40. Qh3 Rg8
41. h5 Rg1 42. Kd3 Rd1+ 43. Kc2 Rg1 44. Qd3 Rg2+ 45. Kb1 b4 46. cxb4 axb4 47.
Qb5 Kg7 48. Qxb4 Rg4 49. Qc4 1-0[/pgn]

24...Kh7 is maybe better than 24...a4

[peter]

about half of these positions are well known for some years. And they are too easy for current engines.

Hi Vincent,
I know they are well known, I just wrote, they aren't in well known other suites, afaik.

Even more than that, as for the 222 it was only important to me, they were not in HTC neither in Eret.
As for their difficulty, Crystal solves 15 of the 20 within 15" at 30 threads, as for the 108 left in your HTC- suite, that's 85/108 with same hardware- TC, so that's about the same rate, isn't it?
I just wanted to replace the 17 doublets that are as well in HTC 108 as in Eret, the 111 of which are solved to an even higher rate nowadays.

You just have the choice to take positons far from practical game- playing like in cyclic- zugzwang- and other composed studies, such there are plenty still too "difficult" for engines at reasonable time because anti- engine- constructed, such positions must not be too many within a test suite with which you want to compare engines of different tactical strength, and there are as many positions as you want to have, if you don't take the definition of "single one game changing best move" too narrow- bounded, but to be too generous about that, that's not good for solved- unsolved- only positions neither.

100 very hard positions are of less statistical relevance, if most engines solve only a few of them in given TC, then 200 that are solved to about 50 to 80 percent, even more, if you evaluate solution- time of each engine and position also, like EloStatTS does.(BTW using that fine program makes discrimination of near to each other results like you have with different versions, branches and nets of same engines too, it makes error bar much lower with as well as any suite, I like to evaluate e.g. HTC or Eret with it much too. Arasan- suite, to have an example of a good but even less difficult suite too, judged at single core STC and evaluated with EloStatTS gives statistical fine results of sensitivity and selectivity too.)
If you finde the time, take a look at the links I gave concerning Frank Schubert's EloStatTs, pity you have to have .cbh- suites for using it, but it's easier to import .epd to .cbh then to convert from .pgn to .epd.

The 222 of HTC108+Eret minus the 17 doublets+the 20 added are to me a good practicable suite for the engines of my interest (not to stick to SF-branches only) and the SMP- hardware- time I like to invest for the results I need.

I like the way you do your positional testing with single core and 30 minutes/position too, yet to me results with a larger suite with short SMP- TC and a broader range of positions of very (but not too) hard (neither) and moderate difficulty mixed, are statistically more relevant for a broader range of engines, branches, setting, nets and hardware- TC.

Anyhow, each and every single test position as well as any collection of test positions as well as any eng-eng-match started from opening- test- postions has results of its own meaning and comparabiltiy to other results only, there isn't anything like "overall playing strength" as well as there isn't any "overall tactical strength" neither, at least nothing like that is really measurable, neither in Elo nor in any other scale.

Thanks for your fine work in finding, analysing and collecting test- postions over the years and that anybody can test with the positions you found and collected and with others in as many different suites and ways as there are different testers, engines and hardware- TC- settings.
So everybody has enough choice to pick just the positions and the ways of testing for a very special interest of one's own

[Paloma]

from the +20 added:

Nebenvariante 41.Ke4! (9)

Nebenvariante from what game/study?

[peter]

from the +20 added:

Nebenvariante 41.Ke4! (9)

Nebenvariante from what game/study?

forum3/viewtopic.php?p=918948#p918948

[peter]

from the +20 added:

Nebenvariante 41.Ke4! (9)

Nebenvariante from what game/study?

forum3/viewtopic.php?p=918948#p918948

Edit- time over, didn't come to the database soon enough, but here's the game now, deleted all unnecessary lines and comments except the side- line of interest:

[pgn][Event "Nebenvariante 41.Ke4!"]
[Site "Breslau GER"]
[Date "1889.07.24"]
[Round "1"]
[White "Lasker, Emanuel"]
[Black "Lipke, Paul "]
[Result "1-0"]
[ECO "C26"]
[Annotator "me"]
[PlyCount "93"]
[EventDate "1889.??.??"]

1. e4 e5 2. Nc3 Nf6 3. g3 Bc5 4. Bg2 O-O 5. Nge2 d6 6. O-O Nc6 7. d3 Be6 8. Nd5
Bxd5 9. exd5 Ne7 10. Bg5 Nd7 11. d4 exd4 12. Nxd4 h6 13. Be3 Ne5 14. Qe2 Qd7
15. h3 Rae8 16. Rae1 a6 17. f4 N5g6 18. Qf2 Bxd4 19. Bxd4 Nf5 20. Bc3 Rxe1 21.
Rxe1 Re8 22. Bf3 Rxe1+ 23. Qxe1 Nfe7 24. Kg2 Qa4 25. Qd2 Qxa2 26. Qd4 f6 27.
Qe4 Qb1 28. h4 Kf8 29. Bh5 f5 30. Qd4 Qxc2+ 31. Kh3 Qe4 32. Qxg7+ Ke8 33. Bxg6+
Kd7 34. Bxf5+ Qxf5+ 35. Qg4 Qxg4+ 36. Kxg4 Nxd5 37. Bd2 Nf6+ 38. Kf5 Ke7 39. g4
d5 $2 (39... Kf7 40. g5 Nd5 41. Ke4 $1 c6 42. f5 b6 43. g6+ Kg7 44. b3 b5 45.
b4) 40. g5 hxg5 41. fxg5 Nd7 42. g6 Kf8 43. h5 d4 44. h6 Kg8 45. h7+ Kh8 46.
Ke6 Nf8+ 47. Kf7 1-0
[/pgn]

[Paloma]

Thanks Peter

[amchess]

New releases
ShashChess 21:
https://github.com/amchess/ShashChess/releases/tag/21

ShashChess tests results:
-testsuite
https://github.com/amchess/ShashChess/w ... te-results
-match
https://github.com/amchess/ShashChess/wiki/Matches

BrainLearn 16:
https://github.com/amchess/BrainLearn/releases/tag/16

[Jouni]

Nice! But even with GoldDigger option set no progress in test positions after my first tests.

[peter]

New releases
ShashChess 21:
https://github.com/amchess/ShashChess/releases/tag/21

ShashChess tests results:
-testsuite
https://github.com/amchess/ShashChess/w ... te-results
-match
https://github.com/amchess/ShashChess/wiki/Matches

BrainLearn 16:
https://github.com/amchess/BrainLearn/releases/tag/16

Thanks for the update!

One thing I'm still not sure about is MCTS- threads- option.
If it's set to 1 (default), how does the engine choose this single one thread?
I always thought MCTS- search to compare all possible moves to each other anyhow and in MultiPV- Mode all threads were treated equal as for MCTS, aren't they?

So how does the engine choose a single thread (which according to readme never is the same as primary line of A-B-search) and how does it treat it as for comparing it to A-B-primary- line?
Or doesn't MCTS and A-B-search interact at all?

Or do "threads" in this respect mean something hardware- related like threads of CPU- cores do?

Thanks in advance for a try to explain such probably rather complicated questions to a rather raw recruit as for coding at all regards

[amchess]

Hi, Peter.
The main thread (id=0) is always A/B.
The MCTSThreads option specifies the number of threads (except the first main one) for MMCTS.
Andrea