New engine: Caissa

[Rubinus]

15m+10s, both cpu=6, GUI DR3/F11-CB, book C30-39-LC0.ctg, depth 8.

Caissa 0.3 -Andscacs 0.95 1:4=7
Caissa 0.4 -Andscacs 0.95 3:3=6

games
It's only 12 games so far, but the shift is +89. Even if it ends up being less, it doesn't look bad. Version 0.4 had the upper hand in that game, but they blew a lot of chances. I see two weaknesses.
1.King safety and the ability to threaten a foreign king with an attack.
2.Still those unfortunate free pawns.
Both can be demonstrated in the following game. After 21 moves a nice position, but then it turns out that the program can't attack. Even a mediocre player like me would probably finish the job. Open a column, or prepare a sacrifice on h6 breaking the pawn cover. 22.Qh5 is a loss of tempo and 26.g6 definitely throws away the chances. There's no way to get to the king, White's black bishop is missing and it's too far for the knight. Finally, a pawn escapes on the other wing.

At the same time, those moves could have been rejected by a good network, although there will be more of a problem that the network is designed for a more efficient program. Maybe I should try with another network, there is a link here in the forum to a harmon network for SF 13, the size is the same as yours. The current score in my tests is SF 13-Harmon - SF 14.1 0:1=21 (-16ELO), so it might be at a decent level. Although Stockfish probably has, maybe just for now, significantly better selectivity. Perhaps I could try it with another network, after all yours is a bit old.

[pgn][Event "15m+10s"]
[Site "?"]
[Date "2021.11.12"]
[Round "7"]
[White "Caissa 0.4"]
[Black "Andscacs 0.95"]
[Result "0-1"]
[ECO "C30"]
[Annotator "0.04;-0.01"]
[PlyCount "78"]
[EventDate "2021.??.??"]
[TimeControl "900+10"]

{Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz W=18.7 plies; 499kN/s; C30-39-LC0.
ctg B=25.4 plies; 7 597kN/s; 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. Nc3 {0.00/0 0} Nc6 {0.00/0 0} 4. Nf3 {0.00/0 0}
d6 {0.00/0 0} 5. Bc4 {0.00/0 0} Nf6 {0.00/0 0} 6. d3 {0.00/0 0} a6 {0.00/0 0}
7. f5 {0.00/0 0} Na5 {0.00/0 0} 8. Bb3 {0.00/0 0} b5 {0.00/0 0 Oba mimo
knihovnu} 9. Qe2 {0.04/20 49} O-O {-0.01/28 81 (Jxb3)} 10. Bg5 {1.10/20 32
(Se3)} c6 {-0.01/25 76 (Jxb3)} 11. g4 {0.96/19 27 (h4)} Re8 {0.38/25 71 (Jxb3)}
12. h4 {0.94/19 32} d5 {0.23/23 67 (Jxb3)} 13. Rh3 {0.74/17 31 (0-0-0)} Bb7 {
-0.15/23 63 (Jxb3)} 14. Kf1 {1.29/18 29 (Sxf6)} Bb6 {-0.12/24 59 (Sd4)} 15. Nd2
{1.07/19 52 (Sxf6)} Ra7 {-0.36/22 55 (Dd6)} 16. Kg2 {1.45/18 39 (Sxf6)} Qd6 {
-0.14/24 52} 17. Bxf6 {1.36/19 42} Qxf6 {0.00/24 49} 18. Rg3 {1.49/19 34 (g5)}
Bc8 {0.14/25 46 (Dd8)} 19. Rf1 {1.31/19 41 (exd5)} Qd8 {0.01/24 19} 20. g5 {1.
88/19 82} Kh8 {0.08/24 19 (Jxb3)} 21. Nd1 {1.90/18 30 (Df3)} f6 {0.00/24 43
(Jxb3)} 22. Qh5 {2.37/19 46} Rf8 {0.00/24 30 (Jxb3)} 23. Ne3 {1.95/19 32} Bxe3
{-0.27/24 27 (Jxb3)} 24. Rxe3 {1.99/17 4} Nxb3 {-0.01/25 15} 25. axb3 {1.10/19
33 (Jxb3)} dxe4 {-0.53/28 18} 26. g6 {0.66/20 25} h6 {-0.58/29 13} 27. Nxe4 {
0.22/21 46 (Vxe4)} Qd4 {-1.11/29 12} 28. Re2 {0.00/21 26 (De2)} Qxb2 {-1.16/18
23} 29. Qf3 {-0.20/20 60} Rd8 {-1.12/22 16 (Dd4)} 30. Kh2 {0.10/19 41} Kg8 {-0.
97/27 21 (c5)} 31. Ref2 {0.03/18 96 (Vd1)} a5 {-1.24/27 16} 32. Qh5 {-0.89/18
28 (Kg3)} c5 {-1.74/24 18 (a4)} 33. Ng5 {-1.72/17 39 (Vg1)} a4 {-3.13/25 25}
34. bxa4 {-2.99/19 38} bxa4 {-3.04/27 15} 35. Rg2 {-3.93/20 119 (Je4)} Bb7 {-4.
51/27 19 (Vdd7)} 36. Rgf2 {-3.84/19 14 (Je4)} a3 {-6.26/28 28 (Vd7)} 37. Nf7 {
-7.29/19 77 (Je6)} Rd7 {-9.79/28 31} 38. c3 {-8.72/16 15 (De2)} Qxc3 {-23.25/
25 55} 39. Nxh6+ {-11.57/16 18} gxh6 {-90.10/22 47} 0-1
[/pgn]

[Rubinus]

At the same time, those moves could have been rejected by a good network, although there will be more of a problem that the network is designed for a more efficient program. Maybe I should try with another network, there is a link here in the forum to a harmon network for SF 13, the size is the same as yours. The current score in my tests is SF 13-Harmon - SF 14.1 0:1=21 (-16ELO), so it might be at a decent level. Although Stockfish probably has, maybe just for now, significantly better selectivity. Perhaps I could try it with another network, after all yours is a bit old.

15m+10s, both cpu=6, GUI DR3/F11-CB, book C30-39-LC0.ctg, depth 8.

Caissa 0.3 -Andscacs 0.95 1:4=7
Caissa 0.4 -Andscacs 0.95 3:3=6

Caissa 0.4-Harmon NN-Andscacs 0.95 0:9=1 !!!
Ohh. I didn't see that coming. While with Stockfish 13 the network program will deteriorate only minimally, at a rapid pace, in the case of your program the result is tristic. Depth matters.

So I can think of another test. I've been testing programs actually for over 20 years now, because I enjoy looking at batches of them. And after some time I have defined a way that even a layman can recognize a quality program in a few minutes. By how quickly it gets deep, by the multiple of time it takes to extend one half stroke. In the days of MChess Pro circa 1995, that was five times. With Lang's Genius, it was worse. Stockfish, if it doesn't have to change the main variant, does it about 3 times faster. So I'll try to compare it on a couple of positions to see how Caissa is doing.

[Rubinus]

1:1=4 against Schooner so far. But I guess you can look for another mistake. Apparently, under certain circumstances, the plus and minus signs change. Or a mistake in the handling of the Syzygy. Otherwise, it looked like the program went crazy. It could have been a nice win, and instead it's a loss - and it still drove my GUI crazy, which closed the game with a 1:0 rating. But I sat there because I got home early. 63.Vd5??, I've never seen that with my own eyes on the computer.

[pgn][Event "15m+10s"]
[Site "?"]
[Date "2021.11.13"]
[Round "5"]
[White "Caissa 0.4"]
[Black "Schooner 2.2 SSE POPCNT"]
[Result "0-1"]
[ECO "C33"]
[Annotator "-0.03;0.10"]
[PlyCount "131"]
[TimeControl "900+10"]

{Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz W=20.8 plies; 521kN/s; C30-39-LC0.
ctg B=23.9 plies; 14 761kN/s; C30-39-LC0.ctg} 1. e4 {B 0} e5 {B 0} 2. f4 {B 0}
exf4 {B 0} 3. Bc4 {B 0} d5 {B 0} 4. Bxd5 {B 0} Nf6 {B 0} 5. Nc3 {B 0} Nxd5 {B 0
} 6. Nxd5 {B 0} Bd6 {B 0} 7. Nf3 {B 0} c6 {B 0} 8. Nc3 {B 0} Bg4 {B 0 Oba mimo
knihovnu} 9. d4 {-0.03/20 35} Bc7 {0.10/22 26} 10. Bd2 {-0.15/20 75 (h3)} Nd7 {
0.00/24 102} 11. Qe2 {-0.30/20 33} Qe7 {-0.05/22 81 (0-0)} 12. O-O-O {-0.39/19
27} O-O {-0.16/22 51} 13. Qf2 {-0.19/18 74 (e5)} Rad8 {-0.07/21 56 (Vae8)} 14.
Rhe1 {-0.01/18 40 (Kb1)} f6 {-0.15/21 35 (Vde8)} 15. Kb1 {0.06/19 64} Nb6 {-0.
22/23 76 (Sh5)} 16. b3 {0.25/19 35} Bh5 {-0.24/22 55 (Vfe8)} 17. Bc1 {0.49/19
33} g5 {-0.26/20 30 (Vfe8)} 18. Bb2 {0.74/17 64 (h3)} Rfe8 {-0.35/21 19 (Sg6)}
19. g3 {0.97/18 27 (a3)} Nd7 {-0.15/21 50 (fxg3)} 20. gxf4 {2.31/19 40} Bxf4 {
-0.02/23 24} 21. h4 {2.91/18 47} g4 {0.05/21 44 (Vf8)} 22. Ng1 {3.39/20 27} Bc7
{0.00/21 20} 23. Nge2 {3.17/20 34} Bf7 {0.30/21 45 (b5)} 24. h5 {4.67/20 37
(Jg3)} Bxh5 {0.79/23 71} 25. Qh4 {5.16/20 25} Qf7 {0.72/22 38} 26. Rg1 {4.96/
20 36 (Vh1)} Bg6 {1.09/21 26 (b5)} 27. Qxg4 {5.88/19 45} b5 {1.38/21 53 (Kh8)}
28. Nf4 {7.48/18 21} Bxf4 {1.29/23 23 (b4)} 29. Qxf4 {7.59/17 11} Qe7 {1.54/22
14 (Jb6)} 30. Rde1 {6.78/19 38 (d5)} b4 {1.61/23 24 (a5)} 31. Nd1 {7.19/21 38}
c5 {1.55/22 15} 32. d5 {7.30/20 36 (dxc5)} a5 {1.95/24 72} 33. Nf2 {7.23/19 28}
Ne5 {1.98/22 17} 34. Ref1 {7.62/19 43} a4 {2.64/22 20} 35. Ng4 {7.93/20 23}
Nxg4 {2.92/23 27} 36. Rxg4 {7.64/16 7} axb3 {2.47/23 27 (Vd7)} 37. Bxf6 {7.85/
17 5} bxc2+ {2.63/24 10} 38. Kxc2 {8.49/20 46} Qxe4+ {2.67/27 10} 39. Qxe4 {8.
09/21 10} Rxe4 {2.91/27 9} 40. Rxg6+ {10.14/21 15} hxg6 {3.13/26 8} 41. Bxd8 {
11.05/21 8} Rc4+ {3.34/26 11 (g5)} 42. Kb3 {10.68/26 30} Rd4 {3.28/27 16} 43.
Be7 {10.66/27 36 (Vc1)} Rxd5 {3.21/26 13} 44. Kc4 {10.57/24 19 (Vc1)} Rd4+ {3.
90/25 34 (Vd2)} 45. Kb5 {10.99/26 22} Rd2 {4.15/25 10} 46. Ra1 {11.16/27 23} b3
{4.35/26 22 (Vb2)} 47. axb3 {13.75/19 3} Rb2 {4.63/24 9} 48. Ra3 {13.82/24 28
(Kc4)} c4 {4.80/24 7 (g5)} 49. Kxc4 {13.89/23 20} Kf7 {4.95/25 27 (Vc2+)} 50.
Bg5 {14.04/23 50 (Sc5)} Ke6 {4.30/25 9} 51. Ra5 {14.12/23 25 (b4)} Rc2+ {4.03/
26 34} 52. Kb4 {14.13/25 36 (Kd4)} Rb2 {3.99/25 15} 53. Rc5 {14.14/24 10} Kd6 {
4.34/27 14 (Kd7)} 54. Bf4+ {14.23/24 10} Kd7 {4.28/25 19 (Ke6)} 55. Rg5 {14.16/
22 10 (Kc4)} Ke6 {4.24/24 11 (Kc6)} 56. Be5 {14.60/23 8 (Kc4)} Rb1 {4.66/23 11
(Vd2)} 57. Bd4 {14.76/23 12 (Kc4)} Rc1 {4.93/26 20} 58. Kb5 {14.88/23 10
(Vxg6+)} Rc2 {4.96/25 16 (Vd1)} 59. b4 {14.95/22 10} Rc8 {5.15/25 13 (Vd2)} 60.
Ka6 {15.49/22 9} Rc4 {4.70/23 9 (Va8+)} 61. b5 {15.78/22 11} Ra4+ {2.92/30 10
(Kf7)} 62. Kb7 {15.97/22 9} Rxd4 {4.53/26 8 (Vc4)} 63. Rd5 {-173.38/19 11
(Vxg6+)} Kxd5 {#2744/24 6 (Vxd5)} 64. Ka6 {0.01/0 0 (Kc7)} Kc5 {#2752/28 6} 65.
Ka5 {0.01/0 0 (b6)} Rb4 {#2754/34 7} 66. Ka6 {0.01/0 0 (b6)} 0-1
[/pgn]

[Rubinus]

So I can think of another test. I've been testing programs actually for over 20 years now, because I enjoy looking at batches of them. And after some time I have defined a way that even a layman can recognize a quality program in a few minutes. By how quickly it gets deep, by the multiple of time it takes to extend one half stroke. In the days of MChess Pro circa 1995, that was five times. With Lang's Genius, it was worse. Stockfish, if it doesn't have to change the main variant, does it about 3 times faster. So I'll try to compare it on a couple of positions to see how Caissa is doing.

A Czech enthusiast has been working on the variant below for many years. Here.
1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. c3 Nf6 5. d4 exd4 6. cxd4 Bb4+ 7. Nc3 Nxe4 8.
O-O Bxc3 9. d5 Bf6 10. Re1 Ne7 11. Rxe4 d6 12. Bg5 Bxg5 13. Nxg5 h6 14. Qa4+
Bd7 15. Qb3
[fen]r2qk2r/pppbnpp1/3p3p/3P2N1/2B1R3/1Q6/PP3PPP/R5K1 b kq - 0 15[/fen]
All CPUs=6
Caissa 0.4 2,014x
0:16 depth 20
0:39 depth 21
1:21 depth 22 (Fluctuating between Bf5 and 0-0.)
1:55 depth 23
3:24 depth 24
8:50 depth 25 => 530/16=33,125 => Average efficiency to extend depth by half a stroke = 2,014 (five half-taxes, i.e. the fifth root)
Stockfish 14.1 1,28x
0:16 depth 30
8:44 depth 44 => 524/16=32,75 => 1,28!!! (the average multiple of the previous depth extension time)
Of course, it's a long way to the effectiveness of a world leader. Some kind of veteran.
Deep Hiarcs WSC2011 1,84x
0:15 depth 22
9:34 depth 28 => 38,267 => 1,84.
Hiarcs is an old program, but it has over 20 years of work behind it. So the efficiency is clear. If Mark Uniacke had finally committed to 64bits and included a knowledge network, we would probably have a strong program again, though maybe not the best.

If Caissa 0.4 is now able to play against Andscacs and Schooner, it would probably beat the 200 ELO weaker HIARCS according to CCRL. But the latter has better efficiency, so for some longer time, maybe 90+30 instead of 15+10, it might not be so certain.

15. - O-O 16. Rae1 b5 17. Bd3 Re8
[fen]r2qr1k1/p1pbnpp1/3p3p/1p1P2N1/4R3/1Q1B4/PP3PPP/4R1K1 w - - 0 18[/fen]
The next victim was found I think first on my computer - some Athlon3200 64bit - by Shredder10 and it took him about 10 minutes. A little later Rybka managed to find it and the other engines were completely blind here then, at least in some normal time, not overnight.
Caissa 0.4 1,81
0:10 depth 19
5:49 depth 25 => 1,81 (Here the main move Ne6! was not changed, so the chopping was faster.)
Stockfish 14.1 1,34
0:13 depth 30
7:17 depth 42 => 1,34
Hiarcs WCSC2011 1,38
0:21 depth 18
6:47 depth 28 => 1,38 (Even he could have cut faster.)
Well, apart from some mistakes see above, there is room for improvement here, in the ability to advance faster in depth.

[Rubinus]

15m+10s, both cpu=6, GUI DR3/F11-CB, book C30-39-LC0.ctg, depth 8.

Caissa 0.4 -Schooner 2.2 1:2=9 A significant improvement over version 0.3. And it could have been better, too bad. A lot of wasted chances.
Caissa 0.4 -LCO 0.28.0 0:7=5 Nothing has changed here, although I'm only running the weaker CUDNN variant, this engine is currently out of range.


Two of the least successful games. Free pawns, open columns against the king. Against LC0 the library was mainly a rejectedgambit and that's where Caissa is brutally outplayed since LC0. In time, I think it's going to take a better net.
[pgn][Event "15m+10s"]
[Site "?"]
[Date "2021.11.15"]
[Round "3"]
[White "Caissa 0.4"]
[Black "Lc0 v0.28.0"]
[Result "0-1"]
[ECO "C30"]
[Annotator "0.09;-0.14"]
[PlyCount "72"]
[EventDate "2021.??.??"]
[TimeControl "900+10"]

{Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz W=21.8 plies; 617kN/s; C30-39-LC0.
ctg B=16.7 plies; 8kN/s; 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. Nc3 {0.00/0 0} Nc6
{0.00/0 0} 5. Bb5 {0.00/0 0} Nge7 {0.00/0 0} 6. Na4 {0.00/0 0} Bg4 {0.00/0 0}
7. d3 {0.00/0 0} O-O {0.00/0 0} 8. Nxc5 {0.00/0 0} dxc5 {0.00/0 0 Oba mimo
knihovnu} 9. Be3 {0.09/20 34} Nd4 {-0.14/19 60} 10. Bxd4 {-0.07/21 28} cxd4 {
-0.14/15 0} 11. f5 {-0.10/22 53} a6 {-0.16/19 42 (c5)} 12. Ba4 {-0.92/22 33
(Sc4)} b5 {-0.19/18 30} 13. Bb3 {-0.65/23 66} c5 {-0.20/19 0} 14. O-O {-0.84/
22 56} Rc8 {-0.18/20 47 (Dc7)} 15. c4 {-1.06/21 65 (De2)} dxc3 {-0.20/21 43}
16. bxc3 {-1.09/21 41} Qb6 {-0.21/17 0} 17. Kh1 {-1.30/20 30} c4 {-0.22/14 0}
18. dxc4 {-1.20/21 28} bxc4 {-0.21/19 49} 19. Ba4 {-1.31/21 25} Rfd8 {-0.25/20
97} 20. Qc2 {-1.17/21 30 (De1)} Bxf3 {-0.31/15 43 (Da5)} 21. Rxf3 {-1.28/22 48}
Kh8 {-0.34/17 75 (Vd6)} 22. Rb1 {-1.15/22 29} Qh6 {-0.34/13 0} 23. f6 {-1.15/
22 32} gxf6 {-0.32/11 0} 24. Rg1 {-1.46/21 29} f5 {-0.27/15 71 (Dh4)} 25. exf5
{-0.88/22 26} f6 {-0.28/15 0} 26. Qb1 {-0.28/21 29 (De2)} Nd5 {-0.39/15 39} 27.
Qb7 {-0.46/22 28} Qg7 {-0.38/17 24} 28. Qxa6 {-1.10/24 36 (Sc6)} Qc7 {-0.58/19
38} 29. Bb5 {-1.62/23 26} Rd6 {-0.67/23 20} 30. Qa4 {-2.02/23 69 (Da3)} e4 {-1.
68/16 38} 31. Rg3 {-3.82/22 27} e3 {-1.95/17 20} 32. Qa3 {-6.04/22 104 (Dc2)}
Rb6 {-3.26/16 39} 33. Qb2 {-6.23/22 19} Nf4 {-3.82/17 23} 34. Rxe3 {-6.85/21 21
} Nd3 {-3.99/14 0} 35. Rxd3 {-8.42/22 27} cxd3 {-4.06/14 30} 36. c4 {-8.32/23
26} Qf4 {-4.39/13 19} 0-1
[/pgn]
[pgn][Event "15m+10s"]
[Site "?"]
[Date "2021.11.15"]
[Round "9"]
[White "Caissa 0.4"]
[Black "Lc0 v0.28.0"]
[Result "0-1"]
[ECO "C32"]
[Annotator "-0.42;-0.06"]
[PlyCount "60"]
[EventDate "2021.??.??"]
[TimeControl "900+10"]

{Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz W=21.3 plies; 536kN/s; C30-39-LC0.
ctg B=19.5 plies; 9kN/s; 2 TBAs; C30-39-LC0.ctg} 1. e4 {0.00/0 0} e5 {0.00/0 0}
2. f4 {0.00/0 0} d5 {0.00/0 0} 3. exd5 {0.00/0 1} e4 {0.00/0 0} 4. d3 {0.00/0 0
} Nf6 {0.00/0 0} 5. c4 {0.00/0 0} Bc5 {0.00/0 0} 6. d4 {0.00/0 0} Bb4+ {0.00/0
0} 7. Nc3 {0.00/0 0} Bxc3+ {0.00/0 0} 8. bxc3 {0.00/0 0} O-O {0.00/0 0 Oba
mimo knihovnu} 9. Ne2 {-0.42/20 39} c6 {-0.06/15 61} 10. dxc6 {-0.27/21 34}
Nxc6 {-0.08/19 29} 11. Ng3 {-0.41/21 82 (Vb1)} Qa5 {-0.21/21 40} 12. d5 {-0.05/
21 45} e3 {-0.21/25 0 (Vd8)} 13. Qd3 {-0.16/20 27} h5 {-0.22/23 29} 14. Bxe3 {
-0.16/20 34 (dxc6)} h4 {-0.28/24 48} 15. Ne2 {-1.44/21 64} Re8 {-0.28/22 0} 16.
dxc6 {-1.26/20 27} Bf5 {-0.23/26 41 (bxc6)} 17. Qd4 {-0.19/21 42} bxc6 {-0.23/
20 1} 18. Qc5 {-0.58/21 36} Qa4 {-0.24/20 0} 19. Bd4 {-0.68/21 39} Qc2 {-0.23/
30 55} 20. Bxf6 {-0.95/21 40 (Da3)} gxf6 {-0.80/25 67} 21. Rd1 {-1.47/21 38
(Kf2)} Bd3 {-1.14/20 45} 22. Rxd3 {-1.56/22 28 (Dh5)} Qxd3 {-1.31/20 34} 23.
Qd4 {-1.59/23 40} Qf5 {-1.33/13 0 (Dg6)} 24. g4 {-1.59/22 26} Qc2 {-1.56/20 44}
25. Qd2 {-2.52/23 31} Qe4 {-1.58/16 1} 26. Rg1 {-2.90/21 6} Rad8 {-1.76/18 37}
27. Qc1 {-6.05/22 51} Rd3 {-1.89/15 15} 28. g5 {-6.73/22 82} f5 {-1.90/10 0}
29. g6 {-7.64/22 40} f6 {-1.94/13 1} 30. h3 {-9.16/22 73 (Vg2)} Rxh3 {-4.16/13
100} 0-1
[/pgn]

[Rubinus]

15m+10s, both cpu=6, GUI DR3/F11-CB, book C30-39-LC0.ctg, depth 8.

Caissa 0.4 -Fizbo 2 5:2=5
Caissa 0.4 -Pedone 3.1 2:3=7
Against five of the same opponents, missing actually Chiron, out of 60 games a gain of 45%. Version 0.3 32%, played two games less against LCO. Approximately less than 100 ELO plus. In the CCRL rankings, I'd estimate somewhere around 3250 with 4CPU and a tempo of 40/15. A longer tempo would probably be a bit weaker due to the slower progression to depth described earlier.

In addition, an error is found where apparently in the root of the syzygyvine the wrong sign plays not the best but the worst move, an example somewhere earlier.

games
Other games against Chiron 5, Fritz 18 (so far =2), Revenge.

[pohl4711]

Caissa 0.4 does not run in cutechess-cli. The default Hash-value is complete bogus and setting the Hash-option failes.

[Guenther]

Caissa 0.4 does not run in cutechess-cli. The default Hash-value is complete bogus and setting the Hash-option failes.

It seems the default size is set in Bytes!
https://github.com/Witek902/Caissa/blob ... I.cpp#L162
https://github.com/Witek902/Caissa/sear ... aultTTSize

The min/max values are instead in MB it seems. This looks a bit quirky. I will try to get some logs for a possible missbehaviour.

Edit:
From cmd there is no problem at least to change to different hash sizes (in MB), checked with taskmanager

Code: Select all

Large page support enabled. Minimum page size: 2048 KB
Caissa 0.4 by Michal Witanowski
Loading NNUE : C:\Engines\UCI\Caissa_04-64/nn-04cf2b4ed1da.nnue
NNUE loaded!
uci
id name Caissa 0.4
id author Michal Witanowski
option name Hash type spin default 16777216 min 1 max 1048576
option name MultiPV type spin default 1 min 1 max 255
option name Threads type spin default 1 min 1 max 64
option name Ponder type check default false
option name EvalFile type string default nn-04cf2b4ed1da.nnue
option name SyzygyPath type string default <empty>
option name UCI_AnalyseMode type check default false
option name UseSAN type check default false
uciok
go
info depth 1 seldepth 1 time 0 score cp 74 nodes 22 nps 32280 pv d2d4
info depth 2 seldepth 3 time 1 score cp 21 nodes 64 nps 51386 pv d2d4 d7d5
info depth 3 seldepth 5 time 7 score cp 49 nodes 331 nps 43846 pv d2d4 d7d5 g1f3
info depth 4 seldepth 9 time 12 score cp 39 nodes 777 nps 62558 pv e2e4 c7c5 g1f3 g8f6
info depth 5 seldepth 11 time 43 score cp 45 nodes 2312 nps 53614 pv d2d4 d7d5 g1f3 g8f6 c2c4
info depth 6 seldepth 11 time 44 score cp 44 nodes 2536 nps 57387 pv d2d4 d7d5 c2c4 d5c4 g1f3 g8f6
info depth 7 seldepth 15 time 131 score cp 49 nodes 6429 nps 48868 pv d2d4 d7d5 c2c4 e7e6 g1f3 g8f6 e2e3
info depth 8 seldepth 22 time 392 score cp 36 nodes 19596 nps 49886 pv c2c4 c7c5 g1f3 e7e6 b1c3 g8f6 d2d4 c5d4
info depth 9 seldepth 20 time 885 score cp 53 nodes 42500 nps 48022 pv d2d4 d7d5 c2c4 e7e6 b1c3 g8f6 g1f3 b8d7 e2e3
info depth 10 seldepth 20 time 944 score cp 33 upperbound nodes 44942 nps 47603 pv d2d4 d7d5 c2c4 e7e6 b1c3 g8f6 g1f3 b8d7 f3e5
info depth 10 seldepth 22 time 629 score cp 29 nodes 85274 nps 135391 pv d2d4 d7d5 c2c4 e7e6 b1c3 g8f6 g1f3 b8d7 c4d5 e6d5
info depth 11 seldepth 22 time 551 score cp 44 lowerbound nodes 25522 nps 46251 pv d2d4 d7d5 c2c4 e7e6 b1c3 g8f6 g1f3 g7g6 c4d5 e6d5
info depth 11 seldepth 22 time 3198 score cp 45 nodes 181707 nps 56811 pv e2e4 e7e5 c2c4 g8f6 b1c3 b8c6 g1f3 f8b4 d2d4 b4c3 b2c3
info depth 12 seldepth 24 time 5491 score cp 40 nodes 263666 nps 48015 pv d2d4 d7d5 c2c4 e7e6 b1c3 g8f6 g1f3 f8e7 c1f4 b8d7 c4d5 e6d5
f2f3
stop
bestmove e2e4 ponder e7e5
setoption name hash value 256
setoption name hash value 16
setoption name hash value 1024

[Guenther]

Code: Select all

Large page support enabled. Minimum page size: 2048 KB
Caissa 0.4 by Michal Witanowski
Loading NNUE : C:\Engines\UCI\Caissa_04-64/nn-04cf2b4ed1da.nnue
NNUE loaded!
uci
id name Caissa 0.4
id author Michal Witanowski
option name Hash type spin default 16777216 min 1 max 1048576

I checked it now in the CuteChessGUI and actually 'Hash' isn't displayed at all in the uci options.
May be it doesn't like that the default value appears to be 'higher' than max size (though it is in Bytes, but Cute cannot know this).

Also the uci specs demand that it is given in MB anyway.

Code: Select all

        * <id> = Hash, type is spin
            the value in MB for memory for hash tables can be changed,
            this should be answered with the first "setoptions" command at program boot
            if the engine has sent the appropriate "option name Hash" command,
            which should be supported by all engines!
            So the engine should use a very small hash first as default.

[Rubinus]

Caissa 0.4 - Fritz 18 so far 1:2=5.
Attractive game. First, a wasted win with moves 20.-Re7? (instead of Be6!) and maybe 22.-Be6? (instead of a6!). Then Fritz's greater effectiveness played a role, getting into greater depths and holding the draw. On the other hand, it should be appreciated that Caissa didn't spoil the position, even though time was already short and her assessment of the position was rather off.
Fritz deepens faster, he would probably be significantly better for tournament time.

[pgn][Event "15m+10s"]
[Site "?"]
[Date "2021.11.17"]
[Round "4"]
[White "Fritz 18"]
[Black "Caissa 0.4"]
[Result "1/2-1/2"]
[ECO "C33"]
[Annotator "-0.72;-2.02"]
[PlyCount "181"]
[TimeControl "900+10"]

1. e4 {B 0} e5 {B 0} 2. f4 {B 0} exf4 {B 0} 3. Bc4 {B 0} Nf6 {B 0} 4. d3 {B 0}
d5 {B 0} 5. exd5 {B 0} Nxd5 {B 0} 6. Bxd5 {B 1 Bílý mimo knihovnu} Qxd5 {B 0}
7. Qe2+ {-0.72/26 37} Be6 {-2.02/20 51} 8. Nc3 {-0.66/28 61} Qa5 {-2.55/20 26}
9. Nf3 {-0.57/28 30 (Sxf4)} Nc6 {-3.28/20 31} 10. Bxf4 {-0.57/28 32} Ba3 {-3.
06/21 47} 11. Bc1 {-0.78/26 31 (Sxc7)} O-O-O {-3.55/21 35} 12. Qf2 {-0.82/28 29
} Rhe8 {-3.35/20 46} 13. O-O {-0.84/29 19} Bc5 {-3.09/22 80} 14. Be3 {-0.87/31
23} Bxe3 {-3.27/22 25} 15. Qxe3 {-0.87/30 17} Bg4 {-3.22/21 26 (Sd5)} 16. Qd2 {
-0.89/30 35 (Dc1)} Qb6+ {-3.74/20 25 (f5)} 17. Kh1 {-0.99/28 49} Qxb2 {-3.85/
21 24} 18. Rab1 {-1.06/27 39} Qa3 {-4.21/19 9} 19. Nb5 {-1.09/29 22} Qa4 {-4.
11/20 32} 20. Ng1 {-1.10/28 18 (Df2)} Re7 {-4.49/18 56 (Se6)} 21. Rf4 {-0.81/
29 48 (h3)} Qxa2 {-3.82/18 5} 22. Qc1 {-0.40/31 57} Be6 {-3.04/20 33 (Sd7)} 23.
Nc3 {0.01/31 30 (Va1)} Qa5 {-0.08/21 8} 24. Rb5 {0.01/31 14} Qxc3 {-0.15/20 4}
25. Ne2 {0.01/32 16} Qf6 {-0.17/22 45 (De5)} 26. Rxf6 {0.01/25 28} gxf6 {-0.38/
19 4} 27. Nf4 {0.01/28 90 (Df1)} f5 {-0.21/21 42 (b6)} 28. c3 {0.01/26 21 (h3)}
a5 {-0.30/21 36} 29. h3 {-0.01/25 11} b6 {0.02/21 33} 30. d4 {-0.01/24 14 (Kh2)
} Na7 {0.00/22 26} 31. Rb2 {-0.01/27 5 (Ve5)} Bc4 {-0.48/23 55} 32. Qc2 {-0.01/
29 11} Re4 {-0.28/22 30} 33. Nh5 {-0.18/29 21} Nb5 {-0.30/21 44} 34. Qa4 {-0.
01/30 20} Nd6 {-0.31/23 29} 35. Qc6 {-0.01/30 20} Kb8 {-0.22/23 38} 36. Nf6 {
-0.01/31 24} Re3 {-0.49/23 41} 37. Nxh7 {-0.01/30 17 (Kh2)} Bd3 {-0.94/21 41}
38. Ng5 {-0.01/31 18} Nc4 {-0.94/21 85 (Vg8)} 39. Rf2 {-0.01/28 9} Nd6 {-0.58/
22 23 (Ve1+)} 40. Rb2 {-0.01/33 3} Nc4 {-0.58/23 49 (Vg8)} 41. Rf2 {-0.01/34 9}
Rd6 {-0.39/23 47 (Ve1+)} 42. Qa4 {-0.01/29 24} Rg6 {-1.21/22 33} 43. Nxf7 {-0.
01/31 19} Rg7 {-0.28/21 24} 44. Rf3 {-0.01/32 14 (Je5)} f4 {-1.76/21 17} 45.
Rxe3 {-0.01/31 11 (Vxf4)} fxe3 {-3.46/19 3} 46. Qe8+ {-0.01/33 11} Ka7 {-1.48/
21 15} 47. Qe7 {-0.01/34 15} Ka6 {-0.73/22 32} 48. Qxc7 {-0.01/32 10} Nd6 {-1.
23/22 15} 49. Qxd6 {-0.01/32 24} e2 {-1.25/23 12} 50. Qe6 {-0.01/34 12} Rxf7 {
-1.16/22 3} 51. Kh2 {-0.01/34 20} a4 {-0.57/22 14} 52. Qc8+ {-0.01/34 17 (d5)}
Rb7 {-1.92/22 21 (Ka7)} 53. Qa8+ {-0.01/31 12} Ra7 {-1.99/22 10} 54. Qc8+ {-0.
01/31 10} Rb7 {-1.99/23 10 (Kb5)} 55. Qa8+ {-0.01/34 18} Ra7 {-1.84/23 10} 56.
Qc8+ {0.01/36 4} Ka5 {-1.40/21 10 (Kb5)} 57. Qf8 {-0.01/30 12} Bb5 {-0.79/21 10
} 58. Qb4+ {-0.01/31 16} Ka6 {0.01/0 0} 59. c4 {-0.01/32 10} Be8 {-1.51/22 17}
60. Kg3 {-0.01/31 30 (h4)} Rf7 {-1.96/19 2} 61. c5 {-0.01/30 14} bxc5 {-2.06/
21 10} 62. dxc5 {-0.01/33 11} Bb5 {-1.98/21 10} 63. Qe4 {-0.01/29 12} Ka5 {-1.
88/21 21 (a3)} 64. Qe3 {-0.01/31 14} Bc4 {-1.79/20 10 (Kb4)} 65. c6 {-0.01/29
10} Kb5 {-1.80/19 9} 66. Qe5+ {-0.01/30 12} Kxc6 {-1.22/18 12 (Kb4)} 67. Qe4+ {
-0.01/31 12 (Dc3)} Kb5 {-1.06/19 10} 68. Qe5+ {-0.01/31 10} Ka6 {-0.97/18 10
(Kb6)} 69. Qc3 {-0.01/32 8 (Da1)} Bb5 {-0.98/20 9 (Kb6)} 70. Qe5 {-0.01/32 11}
Bc4 {-0.98/20 11 (a3)} 71. Qc3 {0.01/37 5 (Da1)} Bb5 {-0.98/20 10 (Kb6)} 72.
Qe5 {0.01/38 5} Rd7 {-0.60/16 9 (a3)} 73. Qf6+ {-0.01/31 9 (Kf2)} Ka7 {-0.51/
19 10 (Kb7)} 74. Kf2 {-0.01/32 9 (Df2+)} Bc4 {-0.61/20 10 (Vd1)} 75. Qa1 {-0.
01/31 9 (Dc3)} Rd1 {-1.71/19 10 (Vf7+)} 76. Qg7+ {-0.01/34 11} Kb6 {-1.79/21 10
} 77. Qf6+ {-0.01/35 14} Kc5 {-1.79/20 10 (Kb7)} 78. Qe5+ {-0.01/34 32 (De7+)}
Kc6 {-1.79/21 10} 79. Qe8+ {-0.01/36 17 (Df6+)} Kb7 {-1.79/21 10 (Kb6)} 80.
Qe7+ {-0.01/36 16} Kc6 {-1.79/21 10 (Ka6)} 81. Qe8+ {-0.01/43 4 (Df6+)} Kb6 {
-1.79/15 10} 82. Qb8+ {-0.01/36 13} Ka5 {-1.71/15 10} 83. Qe5+ {-0.01/35 8} Kb4
{-1.47/16 9 (Ka6)} 84. Qb2+ {-0.01/34 10} Kc5 {-1.47/17 12} 85. Qe5+ {-0.01/45
5} Kb4 {-1.48/16 10 (Kc6)} 86. Qb2+ {0.01/44 8} Ka5 {-1.48/18 10 (Kc5)} 87.
Qe5+ {-0.01/41 12} Kb6 {-1.48/17 10 (Ka6)} 88. Qb8+ {0.01/43 9 (Df6+)} Ka6 {0.
00/28 8 (Ka5)} 89. Qa8+ {-0.01/43 14} Kb5 {0.00/31 9} 90. Qb8+ {-0.01/42 65}
Kc6 {0.00/31 13 (Kc5)} 91. Qe8+ {0.01/43 17 Přijímám remis} 1/2-1/2
[/pgn]