New engine: Caissa

[Gregory Owett]

Hi, Unfortunately, Caissa loses 50% of its games by crashing !

What kind of crash? Does it crash on a specific UCI command or position? Did you load Syzygy endgame tables into the engine?
I have found an issue with evaluating certain drawish positions potentially leading to search explosion.
[/quote]

Most of the time, in a winning position it stops playing, and loses at time.

[Gregory Owett]

( Syzygy is loaded )

[Rubinus]

Syzygy is defined.
*.uci file

Code: Select all

[ENGINE]
Name=Caissa 0.2
Author=Michal Witanowski
Filename=D:\uci\caissa\Caissa_0.2.exe

[OPTIONS]
Hash=1024
MultiPV=1
Threads=6
Ponder=false
EvalFile=nn-04cf2b4ed1da.nnue
SyzygyPath=D:\syzygy
UCI_AnalyseMode=false
UseSAN=false

[fen]r3k2r/pp3p1p/2ppn3/4b3/P1B1P1bP/2N1B3/1PPR2P1/2K4R b kq - 0 16[/fen]
The game stopped in this position. Some error window, something like "Caissa caused by exception". I don't get it. White's last move (Caisse) was 16.Bf1-c4 and then the opponent was on the move, overrunning the time. But the error message was about Caissa, so the game for 12 games overnight stopped after a few moves. I didn't take a screen, maybe I'll try again tonight.

It occurs to me that when both were playing at full power, 6 cores, that after playing a move Caissa did not release the thread and thus the opponent did not load - I am not a programmer, I do not know how it works during the game, I only understand the things described in the book "Chess on PC" - generator, move ordering, alpha-beta, etc. As long as I played matches on 1 core, programs didn't have to pass system resources to each other because I have 6 cores.

[Rubinus]

[Rubinus]

Caissa 0.2 - Fizbo 2 2:3=7, both cpu=1, GUI DR3/F11-CB, book C30-39-LC0.ctg (My experimental book on King Gambit - a few dozen correspondence analyses of LC0 and later Stockfish, some games inserted from the database, where I now go through the played games and use Stockfish to correct obvious mistakes...)-depth 8.
games

It's fine with one core. I'm gonna try to put two in both at night. We'll see if there is a bug if Caissa runs on multiple cores, or if it has to pass system resources in turns. From the picture above, I'd guess more trouble with multiple cores - the move is played there, but somehow noticeably fast, at depth 4.

[Witek]

OK, I managed to reproduce the issue. It's related to multithreading. I'm looking into it...

[Witek]

Hi, Unfortunately, Caissa loses 50% of its games by crashing !

Should be now fixed in newest release: https://github.com/Witek902/Caissa/releases/tag/0.3

[Rubinus]

Remaining tests of version 0.2:
ever 15m+10s
Caissa 0.2 - Fizbo 2 0:3 and crash - both cpu=2 (Even so, the efficiency seems to be weaker than the competition. We'll see with version 0.3, I'll try matches with cpu=6 and cpu=1)
Caissa 0.2 - Pedone 3.1 0:5=3 - both cpu=1
Caissa 0.2 - Chiron 5 0:8=4 both cpu=1
GUI DR3/F11-CB, book C30-39-LC0.ctg (My experimental book on King Gambit - a few dozen correspondence analyses of LC0 and later Stockfish, some games inserted from the database, where I now go through the played games and use Stockfish to correct obvious mistakes...)-depth 8.
games

[Rubinus]

First test of version 0.3:
15m+10s
Caissa 0.3 - Chiron 5 2:7=3 both cpu=6
GUI DR3/F11-CB, book C30-39-LC0.ctg, depth 8.
games
From a strategic point of view, game 11 is interesting, in my opinion both opponents play strategically weak - if I use the position after the 15th, 20th, 25th and 30th move as a starting position for a match against Stockfish, I have the impression that both opponents would lose such a match 0:8.
[pgn]
[Event "15m+10s"]
[Site "?"]
[Date "2021.11.04"]
[Round "11"]
[White "Caissa 0.3"]
[Black "Chiron 5 64bit"]
[Result "0-1"]
[ECO "C30"]
[Annotator "0.28;0.06"]
[PlyCount "118"]
[EventDate "2021.??.??"]
[TimeControl "900+10"]

{Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz W=21.2 plies; 854kN/s; C30-39-LC0.
ctg B=32.5 plies; 8 632kN/s; 18 125 509 TBAs; C30-39-LC0.ctg} 1. e4 {0.00/0 0}
e5 {0.00/0 0} 2. f4 {0.00/0 0} Bc5 {0.00/0 0} 3. Nf3 {0.00/0 0} d6 {0.00/0 0}
4. Bc4 {0.00/0 0} Nf6 {0.00/0 0} 5. d3 {0.00/0 0} O-O {0.00/0 0} 6. Nc3 {0.00/
0 0 Bílý mimo knihovnu} exf4 {0.00/0 0} 7. Bxf4 {0.28/19 52} Nc6 {0.00/0 0
Černý mimo knihovnu} 8. Qd2 {0.09/18 68} Bg4 {0.06/26 74 (Se6)} 9. a3 {-0.05/
17 35} Nd4 {0.01/27 46} 10. Nxd4 {0.17/19 26} Bxd4 {0.01/29 18} 11. Be3 {0.08/
20 52} Bxe3 {0.01/27 24} 12. Qxe3 {0.00/21 50} c6 {0.02/28 32} 13. h3 {0.10/20
56} Bc8 {0.02/27 26} 14. Bb3 {0.08/20 45 (Dd4)} d5 {0.00/29 26} 15. Qg3 {0.01/
20 38} Be6 {0.00/28 25 (dxe4)} 16. exd5 {0.53/20 40 (0-0)} Nxd5 {-0.09/28 28
(cxd5)} 17. Bxd5 {0.45/20 37 (Jxd5)} cxd5 {-0.20/28 35 (Sxd5)} 18. Qf4 {0.50/
19 45 (Df2)} Qb6 {-0.26/30 31} 19. Na4 {0.00/20 27} Qc6 {-0.24/29 39 (Da5+)}
20. Nc3 {-0.20/20 29} a5 {-0.24/28 28} 21. Qd4 {-0.22/19 40 (a4)} b5 {-0.20/29
26} 22. O-O {-0.19/19 26} Rab8 {-0.14/29 26} 23. Rac1 {-0.12/19 26} Rfe8 {-0.
28/29 25 (b4)} 24. b4 {0.00/19 22 (Jd1)} Rbc8 {-0.23/29 25} 25. Nb1 {-0.35/19
44 (Jd1)} a4 {-0.24/29 25} 26. Rf2 {-0.34/19 31} Qc7 {-0.24/32 69 (f6)} 27. c3
{-0.34/20 36 (Jd2)} h6 {-0.38/30 22 (f6)} 28. Nd2 {-0.46/19 27} Qg3 {-0.29/29
24 (f6)} 29. Rf3 {-0.53/20 75 (Ve1)} Qg5 {-0.45/31 22} 30. Qf2 {-0.52/20 19}
Bd7 {-0.44/30 52 (h5)} 31. Rg3 {-0.68/19 27 (d4)} Qe7 {-0.43/29 20} 32. Kh2 {
-0.50/19 27 (Jf1)} g5 {-0.46/29 20} 33. Qd4 {-0.57/19 32 (Jf3)} Qe5 {-0.58/30
19 (f5)} 34. Nf3 {-1.67/20 28} Qxd4 {-0.62/32 19} 35. Nxd4 {-1.87/23 35} f5 {
-0.72/34 20} 36. Rf3 {-2.05/22 16} f4 {-0.75/34 20} 37. g3 {-2.15/22 30} fxg3+
{-0.75/33 38} 38. Kxg3 {-2.14/21 17} Kg7 {-0.75/30 24} 39. Nf5+ {-0.82/21 29
(Kg2)} Bxf5 {-0.96/33 47} 40. Rxf5 {-0.80/22 14} Re3+ {-0.95/35 20} 41. Rf3 {
-1.94/22 36} Rxf3+ {-1.05/36 18} 42. Kxf3 {-0.73/24 15} Kf6 {-1.05/34 20} 43.
h4 {-0.69/24 23 (Kg4)} gxh4 {-1.25/32 20} 44. Ke3 {-0.69/24 13} Kg5 {-0.96/34
55 (Ke5)} 45. Kd4 {-2.30/24 19} Rd8 {-0.23/31 35 (h3)} 46. Kc5 {-0.99/23 17
(Vh1)} d4 {-2.14/31 22} 47. Kxb5 {-0.61/24 20} dxc3 {-0.82/30 42} 48. Rxc3 {-0.
70/23 11} h3 {-0.10/32 29} 49. d4 {-0.84/22 14} Kg4 {-0.10/34 14} 50. Rc2 {-1.
11/22 25} Rd5+ {0.00/35 24 (Vxd4)} 51. Ka6 {0.00/21 13 (Kc6)} Rh5 {-2.15/37 13}
52. Rh2 {-0.60/21 9 (Vc1)} Kg3 {-4.95/39 15} 53. Rh1 {-0.60/24 15} h2 {-6.63/
36 15 (Kg2)} 54. b5 {-2.14/24 10} Kg2 {-8.16/34 13} 55. Rd1 {-5.63/23 10 (Va1)}
h1=Q {-10.52/40 12} 56. Rxh1 {-5.82/25 10} Rxh1 {-12.72/40 12} 57. b6 {-8.35/
26 9} Rf1 {-15.02/43 29} 58. d5 {-8.35/26 10} Rf8 {-17.82/43 14} 59. Kb7 {-8.
54/27 11 (Ka5)} h5 {-21.22/46 16} 0-1
[/pgn]

[Witek]

First test of version 0.3:
15m+10s
Caissa 0.3 - Chiron 5 2:7=3 both cpu=6
GUI DR3/F11-CB, book C30-39-LC0.ctg, depth 8.
games
From a strategic point of view, game 11 is interesting, in my opinion both opponents play strategically weak - if I use the position after the 15th, 20th, 25th and 30th move as a starting position for a match against Stockfish, I have the impression that both opponents would lose such a match 0:8.

Once again - thanks for testing. The games you provide are good source of information. For example in this position the engine fails to find the only move that does not loose the game (Kc6). Stockfish finds a draw in couple of seconds on a single thread...
[fen]8/8/7p/1K1r4/pP1P2k1/P6p/2R5/8 w - - 3 51[/fen]


I need to go through more games and find positions like that and figure out why it fails.