ShashChess

[criko]

New releases
https://github.com/amchess/ShashChess/releases/tag/25
https://github.com/amchess/BrainLearn/releases/tag/19

Please be carefull!! ShashChess is buggy

[Jouni]

HTC 110 positions:

ShashChess 20.2 91/110 time 0.33.35
ShashChess 25 81/110 time 0.40.15

[Eduard]

Which is why I don't like Shashchess. The Engine is defeated too easily on the server time and time again. The reason for this is the minimalistic evaluation and the resulting wrong moves (and plans).

Here is a position my engine recently won on the server against Shashchess 25. At move 35 my engine is at +1.50 while Shashchess 25 is only showing +0.50. The following analyzes were made on Ryzen 3900X. When playing on the server I played with 7 cores, ShashChess 25 about as strong.


[fen]r3k2r/ppp1qpp1/2n4p/2b1pb2/2B1N3/2PP1N2/PP3PPP/R2QK2R w KQkq - 0 12[/fen]


So far both engines are in the book (last bookmove). The position is playable for both sides!


[fen]h4k3/1pp3p1/2n2p1p/2P1p3/p1b5/2P2P1P/P5P1/2NR2K1 w - - 0 1[/fen]


Here on move 35 my engine has gained an advantage of +1.50, while Shashchess 25 mean it's all good!

Analysis by ShashChess 25:

35.Rd2 a3 36.Nd3 Nb8 37.Nb4 c6 38.Nc2 Nd7 39.Nxa3 Be6 40.Rc2 Nxc5 41.c4 Kd7 42.Nb1 Kd6 43.Nd2 f5 44.Kf1 Na4 45.Ke2 Kc5 46.Ke3 g5 47.Rc1 b6 48.g4 f4+ 49.Ke4 Kd6 50.Nb3 Ke7 51.Rc2 Kf6 52.Rd2 b5 53.cxb5 cxb5 54.Rd6 b4 55.Ra6 Nc3+ 56.Kd3 Kf7 57.Ra7+ Kf6
+/= (0.53 ++) Depth: 44/24 00:00:27 481MN, tb=42015

Analysis by Solista Attack v2-avx2:

35.Rd2 a3 36.Nd3 Nb8 37.Nb4 c6 38.Nc2 Be6 39.Nxa3 Nd7 40.c4 Nxc5 41.Rc2 Bf5 42.Rc1 Nd3 43.Rb1 Nc5 44.Rd1 Be6 45.Rc1 Kd7 46.Nb1 Nd3 47.Rc3 Nf4 48.Kf2 Kd6 49.h4 g5 50.hxg5 hxg5 51.g3 Nh3+ 52.Ke3 f5 53.Nd2 f4+ 54.Ke2 fxg3 55.Ne4+ Kc7 56.Nxg3 Nf4+ 57.Kd2
+/- (1.51 ++) Depth: 38/55 00:00:17 336MN, tb=6219


[fen]8/3k4/1R6/1p2p1p1/3n1pP1/5P2/P1bN2K1/8 b - - 0 1[/fen]


Analysis by ShashChess 25:

77...Kc7 78.Ra6 Bd3 79.Ra7+ Kd6 80.Ra8 Bc2 81.Rd8+ Ke6 82.Rg8 Kf6 83.Kf2 Nc6 84.Rc8 Nd4 85.Rf8+ Ke7 86.Rg8 Kf6 87.Kg2 Nc6 88.Ra8 Ke6 89.Ra6 Kd6 90.Rb6 Bd3 91.Kf2 Kc7 92.Ra6 Bc2 93.Ra8 Nd4 94.Re8 Kd6 95.Rg8 Ne6 96.Nb3 Bd3 97.a3 Bc4 98.Nd2 Bd5 99.Rh8 Ke7 100.Rh7+ Kd6 101.Rh5 Bc6 102.Rh3 Nd4 103.Rh6+ Kd7 104.Rg6 Ne6 105.Rf6 Nd4 106.Rh6 Bd5 107.Rh5 Ne6 108.Nb1 Bc6 109.Nc3 Kd6 110.Rh6 Ke7 111.Rh7+ Kd6 112.Rh5 Kc5
+/= (0.56) Depth: 52/71 00:00:44 862MN, tb=1420786

Analysis by Solista Attack v2-avx2:

77...Bd3 78.Ra6
+- (1.77 --) Depth: 44/60 00:00:36 723MN, tb=1858991

Shashschess played weaker and weaker, now it's getting really critical! And Shashchess couldn't hold the endgame, the eval with only 0.56 as a disadvantage turned out to be wrong. I'm sorry but I've seen many similar games by Shashchess. The author should reconsider the Shashchess evaluation scale.

PGN:

[Event "Rated game, 5 min"]
[Site "Engine Room"]
[Date "2022.10.16"]
[Round "?"]
[White "SolistaChess, Dynamics 151022"]
[Black "xxxxxx, Shashchess 25"]
[Result "1-0"]
[ECO "C65"]
[WhiteElo "2901"]
[BlackElo "2885"]
[Annotator "0.42;0.13"]
[PlyCount "239"]
[EventDate "2022.10.16"]
[SourceTitle "playchess.com"]
[TimeControl "300"]

{ShashChess 25 (16 threads): 36.1 plies; 9.149kN/s 11th Gen Intel(R) Core(TM) i9-11900H @ 2.50GHz 2496MHz, (8 cores, 16 threads), SolistaPro v1.ctg, 2048 MB} 1. e4 {B 0} e5 {0.13/38 15} 2. Nf3 {B 0} Nc6 {0.20/37 3} 3. Bb5 {B 0} Nf6 {0.12/36 3} 4. d3 {B 0 (0-0)} Bc5 {0.09/29 4} 5. c3 {B 0} d5 {0.09/35 19} 6. exd5 {B 0} Qxd5 {0.07/32 3} 7. Bc4 {B 0} Qd8 {0.06/35 5} 8. Bg5 {B 0 (Sbd2)} h6 {0.19/27 3} 9. Bxf6 {B 0} Qxf6 {0.08/29 4} 10. Nbd2 {B 0} Bf5 {0.14/31 17} 11. Ne4 {B 0 (De2)} Qe7 {0.07/30 4} 12. O-O {0.42/34 39 (b4)} O-O-O {0.13/36 9} 13. Qe2 {0.30/32 2} Bb6 {0.13/32 9} 14. b4 {0.49/33 10} Rhe8 {0.10/32 10} 15. Rfe1 {0.20/33 12} Bh7 {0.11/34 1 (Kb8)} 16. Bb5 {0.32/33 28} Qd7 {0.17/39 4} 17. Rad1 {0.46/29 3} a6 {0.20/32 0 (Kb8)} 18. Ba4 {0.50/30 5} Re6 {0.14/35 3 (Qe6)} 19. h3 {0.38/28 4} Qe8 {0.16/31 7 (Bg6)} 20. Qc2 {0.54/27 8 (a3)} f6 {0.17/30 14 (Qf8)} 21. Nfd2 {0.81/26 6} Ba7 {0.26/31 5} 22. Rb1 {0.88/28 0} Qd7 {0.31/27 2 (Bg6)} 23. Qb2 {0.83/28 2 (b5)} Qxd3 {0.29/27 4} 24. Rbd1 {1.12/30 6} Qd7 {0.16/32 1} 25. Nc5 {1.12/29 2} Bxc5 {0.21/34 4} 26. bxc5 {1.20/31 0} Qe7 {0.35/31 2} 27. Bb3 {0.99/31 1} Bd3 {0.37/31 3 (Bg8)} 28. Bxe6+ {1.20/29 4} Qxe6 {0.31/34 0} 29. Qb3 {1.24/32 3} Qxb3 {0.46/39 19 (Qg8)} 30. Nxb3 {1.31/35 4} a5 {0.40/37 2} 31. f3 {1.22/33 8 (Sc1)} a4 {0.38/34 3} 32. Nc1 {1.27/32 0} Bc4 {0.46/36 4} 33. Rxd8+ {1.39/32 0} Kxd8 {0.53/36 5} 34. Rd1+ {1.32/31 2} Ke8 {0.38/30 1 (Ke7)} 35. Rd2 {1.55/31 3} a3 {0.48/32 0} 36. Nd3 {1.54/33 3 (Sb3)} Nb8 {0.53/34 2} 37. Nb4 {1.26/31 2} c6 {0.51/36 0} 38. Nc2 {1.50/31 2} Be6 {0.42/39 4 (Na6)} 39. Nxa3 {1.35/33 4} Nd7 {0.44/35 3 (Na6)} 40. c4 {1.54/34 9} Nxc5 {0.49/38 0} 41. Rc2 {1.54/31 1} Ke7 {0.51/36 1} 42. Nb1 {1.54/32 1} f5 {0.62/31 3 (Kd6)} 43. Nd2 {1.43/38 5} Kd6 {0.50/34 0} 44. g3 {1.34/35 3 (Kf2)} g5 {0.47/33 4 (b6)} 45. g4 {1.45/35 4} f4 {0.49/31 0} 46. Rc3 {1.52/33 2 (Tc1)} Kc7 {0.48/34 2} 47. Kg2 {1.51/36 7 (Tc1)} Bf7 {0.57/35 3} 48. Rc1 {1.51/38 0} Bg6 {0.48/35 1} 49. h4 {1.56/35 2} Bd3 {0.58/37 3} 50. Kf2 {1.65/34 0 (Td1)} b6 {0.62/33 1} 51. Re1 {1.70/38 6 (hxg5)} Bc2 {0.51/37 2} 52. Rh1 {1.61/39 0} Kd7 {0.50/38 4 (b5)} 53. hxg5 {1.57/39 7} hxg5 {0.57/43 0} 54. Rh8 {1.52/39 1} Ne6 {0.63/37 3} 55. Rg8 {1.57/37 0 (Tb8)} b5 {0.47/36 3} 56. cxb5 {1.52/39 0} cxb5 {0.46/35 0} 57. Ke2 {1.60/41 2 (Ta8)} Kc7 {0.46/45 1 (Nd4+)} 58. Kf2 {1.61/41 2 (Ta8)} Kd7 {0.46/42 1 (Kd6)} 59. Rb8 {1.52/50 2 (Kg2)} Nd4 {0.46/38 1} 60. Rb7+ {1.44/38 1 (Ke1)} Kd6 {0.43/40 1} 61. Rb6+ {1.44/41 0 (Tg7)} Kd7 {0.43/40 1 (Kc7)} 62. Rb8 {1.44/42 2 (Ta6)} Ke7 {0.46/42 1 (Kd6)} 63. Rb7+ {1.44/50 3 (Tg8)} Kd6 {0.43/50 1} 64. Rb6+ {1.44/46 0} Ke7 {0.43/49 0 (Kd7)} 65. Kg2 {1.44/48 4} Kd7 {0.43/55 0} 66. Rb7+ {1.44/48 2 (Tb8)} Kc6 {0.43/49 1 (Kd6)} 67. Re7 {1.44/46 2 (Tg7)} Kd6 {0.43/52 1} 68. Rg7 {1.44/51 5} Ne6 {0.43/57 0} 69. Rg8 {1.44/48 2} Kd7 {0.43/56 0 (Bd3)} 70. Kf1 {1.44/46 2 (Tb8)} Bd3+ {0.43/48 1} 71. Ke1 {1.44/48 2} Bc2 {0.43/53 0 (Kd6)} 72. Rb8 {1.44/47 1 (Kf2)} Nd4 {0.43/47 1} 73. Rg8 {1.44/45 3 (Tb7+)} Ne6 {0.43/45 1} 74. Kf2 {1.44/47 0} Ke7 {0.43/48 0 (Kd6)} 75. Rb8 {1.44/51 3} Nd4 {0.43/54 0 (Bd3)} 76. Rb6 {1.44/48 1} Kd7 {0.43/51 4} 77. Kg2 {1.44/47 0 (Ta6)} Kc7 {0.42/41 1 (Bd3)} 78. Rf6 {2.21/41 6} Kd7 {0.42/57 0} 79. Ne4 {2.12/32 2 (Tf7+)} Bxe4 {0.39/25 1 (Bb1)} 80. fxe4 {2.86/28 1} Ke7 {0.46/34 0} 81. Rb6 {3.21/29 0} Kf7 {0.43/32 0} 82. Kf1 {3.49/27 1 (Kf2)} Ke7 {1.14/35 2 (Kg7)} 83. Ke1 {4.29/26 2} Nc2+ {1.10/38 0} 84. Kd2 {4.56/24 0} Nd4 {1.23/34 0} 85. Rh6 {4.56/29 1} Ne6 {1.38/32 1} 86. Kc2 {4.59/26 0} Kd7 {1.39/24 0 (Nd4+)} 87. Rh7+ {5.21/24 3} Kd6 {1.67/33 0} 88. Rh2 {5.87/26 4} Nc5 {1.87/34 0 (Kc6)} 89. Rh6+ {5.96/21 1} Ke7 {1.93/27 0} 90. Rh1 {6.07/21 1} Ne6 {2.07/22 0} 91. Rd1 {6.47/20 0} Nd4+ {2.13/22 0 (Kf7)} 92. Kc3 {6.81/21 1} Kd6 {2.24/25 0 (Ne2+)} 93. Kb4 {6.98/20 1} Kc6 {2.48/22 0} 94. Rd2 {7.48/19 0} Kd6 {2.58/21 0 (Kb6)} 95. Rh2 {7.78/19 0} Kc6 {2.68/22 0 (f3)} 96. Rh6+ {8.27/17 1} Kd7 {3.07/20 0 (Kc7)} 97. Kc5 {8.88/20 1} b4 {3.20/20 0 (Ke8)} 98. Kxb4 {9.41/18 1} Ne2 {3.43/20 0 (Nc2+)} 99. a4 {11.04/17 0 (Kc4)} f3 {4.02/19 1} 100. Rh2 {14.13/19 0} Nf4 {4.12/18 0} 101. Kb5 {15.15/18 0} Ne2 {4.16/19 0 (Ng2)} 102. Rf2 {25.11/20 1} Kc7 {5.40/21 0} 103. a5 {141.86/18 0} Kb7 {5.83/15 0 (Nd4+)} 104. a6+ {142.17/23 0 (Kc5)} Ka7 {5.14/11 0} 105. Kc5 {142.26/42 0} Kxa6 {7.86/16 0 (Nd4)} 106. Rxf3 {142.29/44 0} Kb7 {11.15/16 0 (Nf4)} 107. Rf5 {142.30/46 0 (Kd6)} Kc7 {4.02/12 0 (Nf4)} 108. Rxg5 {142.32/46 0 (Kd5)} Nc3 {2.95/1 0 (Nd4)} 109. Rxe5 {142.33/38 0 (Tg7+)} Kd7 {2.57/4 0 (Ne2)} 110. g5 {2/1 0} Na4+ {152.61/29 0} 111. Kd4 {1/0 0} Kd6 {#12/36 0} 112. g6 {1/1 0 (Tf5)} Nc3 {4.40/2 0} 113. Kxc3 {1/1 0 (Td5+)} Kxe5 {1/0 0} 114. g7 {1/1 0} Kf4 {1/0 0} 115. e5 {1/1 0} Kf5 {1/0 0} 116. e6 {1/0 0} Kxe6 {1/0 0} 117. g8=Q+ {7/1 0} Ke5 {7/0 0} 118. Kd3 {6/0 0} Kd6 {6/0 0} 119. Kd4 {5/1 0} Kd7 {5/0 0} 120. Kd5 {4/0 0 xxxxxx,ShashChess 25 gibt auf (Lag: Av=0.26s, max=1.2s)} 1-0

[amchess]

We executed an interesting test for mcts by Shashin
by setting up the mctsThreads with a modified ShashChess version
22 threads in total
from 12 to 20 mcts threads there is the best performance in a match
so the difference is in hard positions solving
the result:
mctsThreads/Threads Solved hard positions
12/22 129
13/22 134
14/22 135
15/22 130
16/22 142
17/22 134
18/22 132
19/22 135
20/22 139

so, it seems the best is

Code: Select all

int mctsThreads = round((float)(Options["Threads"]*16/22));

Standard solves 129 positions
So, it's noticeable
and even in the match it's stronger!
I hope to release the related minor version as soon as possible.
Andrea

[peter]

We executed an interesting test for mcts by Shashin
by setting up the mctsThreads with a modified ShashChess version
22 threads in total
from 12 to 20 mcts threads there is the best performance in a match

How many of the given threads are used now (version 25) if MCTS- option is checked?
Amount of threads can't be set anymore unlike it was at 24,
regards

[amchess]

We executed an interesting test for mcts by Shashin
by setting up the mctsThreads with a modified ShashChess version
22 threads in total
from 12 to 20 mcts threads there is the best performance in a match

How many of the given threads are used now (version 25) if MCTS- option is checked?
Amount of threads can't be set anymore unlike it was at 24,
regards

https://gitlab.com/amchess/ShashChess/- ... /README.md
MCTS by Shashin
Boolean, Default: False If activated, thanks to Shashin theory, the engine will use the MonteCarlo Tree Search for Capablanca quiescent type positions and also for caos ones, in the manner specified by the following parameters. The idea is to exploit Lc0 best results in those positions types, because Lc0 uses mcts in the search. The main thread will always be A/B. The mcts ones, based on our tests, will be the nearest integer value to the following one: Threads * 16 /22

[Eduard]

I have a basic question. Why are Shashchess (and Brainlearn) evaluations so low?

[amchess]

I have a basic question. Why are Shashchess (and Brainlearn) evaluations so low?

Hello.
ShashChess, like Brainlearn, does not change the score provided by the evaluation function compared to Stockfish. What does change is the value sent to a gui to solve the alignment problem with the latter.
In fact, if you try to throw in infinite analysis Stockfish on the initial position, a gui provides a value of 0.40.
That value, according to the informant's symbols, means slight preference for White.
This, clearly, is misleading: for the initial position, a correct value is around 0.15, i.e., even position (while rewarding the trait advantage).
Stockfish community is just interested in getting a bullet monster first and foremost and doesn't care if the engine sends the gui a misleading value.
For Brainlearn, this is merely cosmetic.
For ShashChess, however, this impacts Shashin's theory.

Best regards,
Andrea

[Eduard]

Hello Andrea, I agree that Stockfish is a Bullet Monster. I do not like this. But what should the Stockfish community do? You need a large number of games, and only Bullet can do that.

Before few years I used to like shashchess. But as NNUE continues to make great steps, I see shashchess getting weaker (I'm not talking about testing positions).

Unfortunately, I don't share the opinion that Shashchess plays better than Stockfish in longer games. Before few years I had about 7.000 kns, but now it's 20.000 kns. This results in a transmission of 5 minutes game with 7000 kns at least 13 minutes.

On some losing games from the server (I get databases and study winning games) that Shashchess played, saw that Shashchess plays strange in critical positions. Often not even -1 is shown in a clear losing position. Sometimes Stickfish is already at +3. Shashchess seemed to me like he was blind.

[amchess]

Hello Andrea, I agree that Stockfish is a Bullet Monster. I do not like this. But what should the Stockfish community do? You need a large number of games, and only Bullet can do that.

Before few years I used to like shashchess. But as NNUE continues to make great steps, I see shashchess getting weaker (I'm not talking about testing positions).

Unfortunately, I don't share the opinion that Shashchess plays better than Stockfish in longer games. Before few years I had about 7.000 kns, but now it's 20.000 kns. This results in a transmission of 5 minutes game with 7000 kns at least 13 minutes.

On some losing games from the server (I get databases and study winning games) that Shashchess played, saw that Shashchess plays strange in critical positions. Often not even -1 is shown in a clear losing position. Sometimes Stickfish is already at +3. Shashchess seemed to me like he was blind.

https://github.com/amchess/ShashChess/wiki/Matches
There is another testing strategy based on positional characteristics, I illustrated in the wiki.
NNUE killed someway chess so mostly games are draw.
For the rest, ShashChess has a more realistic evaluation.
Andrea