Komodo Syzygy evals

[Hurnavich]

Hi what went wrong here? Komodo says winning for black white has won


[Event "Rated game, 3 min"]
[Site "Engine Room"]
[Date "2014.05.21"]
[Round "?"]
[White "JPN, StockfishRW 180514 "]
[Black "Hurnavich, Komodo 7 64-bit MP"]
[Result "1-0"]
[ECO "C88"]
[WhiteElo "2611"]
[BlackElo "2643"]
[Annotator "0.39;0.54"]
[PlyCount "125"]
[EventDate "2014.05.21"]
[TimeControl "180"]

{StockfishRW 180514 SSE4.2 AVX. (12 cores): 47.7 plies; 24,522kN/s Intel(R)
Xeon(R) CPU X5650 @ 2.67GHz 4000MHz, Experimental 4.2e.ctg, 512 MB}
1. e4 {B 0} e5 {B 0} 2. Nf3 {B 0} Nc6 {B 0} 3. Bb5 {B 0} a6 {B 0} 4. Ba4 {B 0}
Nf6 {B 0} 5. O-O {B 0} Be7 {B 0} 6. Re1 {B 0} b5 {B 0} 7. Bb3 {B 0} O-O {B 0}
8. d3 {B 0} h6 {B 0} 9. a4 {B 0} b4 {B 0} 10. Bd5 {B 0} Bd6 {B 0} 11. Bxc6 {B 0
} dxc6 {B 0} 12. b3 {B 0} Be6 {B 0} 13. Nbd2 {B 0} a5 {B 0} 14. Bb2 {B 0} Nd7 {
B 0} 15. d4 {B 0} f6 {B 0} 16. Qe2 {B 0} Qc8 {B 0} 17. dxe5 {B 0} Nxe5 {0.54/
19 3} 18. Nxe5 {0.39/29 4} fxe5 {0.49/22 1} 19. Nf3 {0.25/28 10} Bg4 {0.49/23 7
} 20. Qc4+ {0.43/31 0} Be6 {0.50/22 4} 21. Qd3 {0.34/31 1} Qd8 {0.50/23 8} 22.
Nxe5 {0.23/33 3} Qf6 {0.46/23 5 (Bc5)} 23. Qg3 {0.46/30 4} h5 {0.61/22 5 (Rae8)
} 24. h4 {0.52/30 5} Rae8 {0.74/21 4} 25. Rad1 {0.52/32 1 (Rab1)} Bg4 {0.78/21
11 (Bc8)} 26. Rxd6 {0.73/29 6} Qxd6 {0.75/21 11 (cxd6)} 27. f3 {0.70/30 4} Qc5+
{0.70/22 2 (Bc8)} 28. Kh2 {0.73/29 5} Qxc2 {0.75/22 8} 29. Ba1 {0.65/31 5} Qxb3
{0.80/22 0} 30. fxg4 {0.74/30 2} Qxg3+ {0.65/23 10} 31. Kxg3 {0.81/32 4} Rd8 {
0.80/22 9 (hxg4)} 32. gxh5 {0.87/26 5} Rd2 {0.64/21 1} 33. Ng6 {0.79/27 4
(Nxc6)} Rfd8 {0.84/18 2 (Rff2)} 34. Be5 {0.88/26 4} Rd1 {0.98/20 2 (R8d3+)} 35.
Ne7+ {0.92/29 6 (Rxd1)} Kh7 {1.00/21 2} 36. Rxd1 {0.97/32 1} Rxd1 {0.96/21 0}
37. Nxc6 {0.97/33 2} Rc1 {0.96/22 1} 38. Nd4 {0.97/35 1} Rc3+ {1.10/22 1} 39.
Kf4 {1.07/35 0} b3 {1.09/22 1} 40. Nxb3 {1.20/36 1} Rxb3 {1.12/23 1} 41. Bxc7 {
1.22/38 1} Rb2 {1.18/23 2 (Rb4)} 42. g3 {1.34/30 3} Rf2+ {1.38/24 0 (Rb4)} 43.
Ke5 {1.46/34 2 (Ke3)} Rf3 {1.02/21 3 (Ra2)} 44. Bxa5 {2.13/35 3} Rxg3 {1.32/20
1 (Ra3)} 45. Bb6 {1.76/30 4} Kh6 {2.50/21 3} 46. Kd5 {6.50/31 2} Rd3+ {2.89/20
0} 47. Kc4 {6.74/32 1} Rd7 {4.78/19 4 (Rh3)} 48. a5 {10.39/29 6} Re7 {6.91/21
1 (Rb7)} 49. Kd5 {25.05/30 6} Re8 {10.36/22 1 (Kxh5)} 50. a6 {58.60/38 8} Kh7 {
17.51/21 1 (Rf8)} 51. a7 {129.36/34 13} Kh6 {-#1780/26 0 (g6)} 52. Kc6 {151.60/
33 10} Kh7 {-#1781/28 0} 53. Kb7 {30.36/30 8} Re7+ {-#1784/27 0 (Rxe4)} 54. Bc7
{154.90/25 1} Rxe4 {-#1786/26 0 (Re8)} 55. a8=Q {154.92/26 1} Rb4+ {-#1788/26 1
} 56. Bb6 {154.93/28 0} Rb5 {-#1790/25 0 (Rf4)} 57. Qe8 {154.94/27 1 (Qf8)} Rb4
{-#1791/24 1} 58. Qg6+ {154.95/29 0} Kg8 {-#1794/25 0} 59. h6 {154.96/30 1} Kf8
{-#1796/26 0 (Rxb6+)} 60. h7 {#4/133 1} Rxb6+ {-#1798/29 6 (Ke7)} 61. Kxb6 {#3/
192 0} Ke7 {-#1799/9 0} 62. h8=Q {#2/192 0} Kd7 {-#1799/9 0} 63. Qhe8# {#1/192
0} 1-0

Syzygy use bug?

[Hurnavich]

7Q/3k2p1/1K4Q1/8/7P/8/8/8 w - - 0 63

load up Stockfish sees mate to follow Komodo does not with Syzygy



Analysis by Komodo 7 64-bit MP:

63.h5
-+ (-#1799) Depth: 9 00:00:00 0kN, tb=110
(Hurnavich, 21.05.2014)

Analysis by SF 190514 BYO RW:

63.Qhe8#
+- (#1) Depth: 192/2 00:00:00 8kN, tb=38
(Hurnavich, 21.05.2014)

I think it is a six man syzygy bug with Komodo

uncheck syzygy and correct result


Analysis by Komodo 7 64-bit MP:

63.Qhe8#
+- (#1) Depth: 24 00:00:11 187MN
(Hurnavich, 21.05.2014)

[Ajedrecista]

Hello:

[d]7Q/3k2p1/1K4Q1/8/7P/8/8/8 w - - 0 63

load up Stockfish sees mate to follow Komodo does not with Syzygy



Analysis by Komodo 7 64-bit MP:

63.h5
-+ (-#1799) Depth: 9 00:00:00 0kN, tb=110
(Hurnavich, 21.05.2014)

Analysis by SF 190514 BYO RW:

63.Qhe8#
+- (#1) Depth: 192/2 00:00:00 8kN, tb=38
(Hurnavich, 21.05.2014)

I think it is a six man syzygy bug with Komodo

uncheck syzygy and correct result


Analysis by Komodo 7 64-bit MP:

63.Qhe8#
+- (#1) Depth: 24 00:00:11 187MN
(Hurnavich, 21.05.2014)

I think it is a particular output, something like ±#(1800 - X), where ± indicates the checkmate from the side to move (+ if it is in favour of Komodo or - if it is against it) and X means 'mate in X moves' (moves, not plies). For example, look at 59.- ..., Kf8 (-#1796) by Komodo and 60.- h7 (#4) by SF. More strange are 61th and 62th moves by black (both -#1799 instead of -#1798 and -#1799 respectively).

Probably it is ±#(1801 - X) instead of ±#(1800 - X). Better help is needed.

But I see: 63.- h5? instead of 63.- Qhe8#. I guess it does not relate with Distance to Mate (DTM) versus Distance to Conversion (DTC), isn't it? I am not an expert on TB as everybody can see.

Here is the full game, posted at the original post:

[pgn][Event "Rated game, 3 min"]
[Site "Engine Room"]
[Date "2014.05.21"]
[Round "?"]
[White "JPN, StockfishRW 180514 "]
[Black "Hurnavich, Komodo 7 64-bit MP"]
[Result "1-0"]
[ECO "C88"]
[WhiteElo "2611"]
[BlackElo "2643"]
[Annotator "0.39;0.54"]
[PlyCount "125"]
[EventDate "2014.05.21"]
[TimeControl "180"]

{StockfishRW 180514 SSE4.2 AVX. (12 cores): 47.7 plies; 24,522kN/s Intel(R)
Xeon(R) CPU X5650 @ 2.67GHz 4000MHz, Experimental 4.2e.ctg, 512 MB}
1. e4 {B 0} e5 {B 0} 2. Nf3 {B 0} Nc6 {B 0} 3. Bb5 {B 0} a6 {B 0} 4. Ba4 {B 0}
Nf6 {B 0} 5. O-O {B 0} Be7 {B 0} 6. Re1 {B 0} b5 {B 0} 7. Bb3 {B 0} O-O {B 0}
8. d3 {B 0} h6 {B 0} 9. a4 {B 0} b4 {B 0} 10. Bd5 {B 0} Bd6 {B 0} 11. Bxc6 {B 0
} dxc6 {B 0} 12. b3 {B 0} Be6 {B 0} 13. Nbd2 {B 0} a5 {B 0} 14. Bb2 {B 0} Nd7 {
B 0} 15. d4 {B 0} f6 {B 0} 16. Qe2 {B 0} Qc8 {B 0} 17. dxe5 {B 0} Nxe5 {0.54/
19 3} 18. Nxe5 {0.39/29 4} fxe5 {0.49/22 1} 19. Nf3 {0.25/28 10} Bg4 {0.49/23 7
} 20. Qc4+ {0.43/31 0} Be6 {0.50/22 4} 21. Qd3 {0.34/31 1} Qd8 {0.50/23 8} 22.
Nxe5 {0.23/33 3} Qf6 {0.46/23 5 (Bc5)} 23. Qg3 {0.46/30 4} h5 {0.61/22 5 (Rae8)
} 24. h4 {0.52/30 5} Rae8 {0.74/21 4} 25. Rad1 {0.52/32 1 (Rab1)} Bg4 {0.78/21
11 (Bc8)} 26. Rxd6 {0.73/29 6} Qxd6 {0.75/21 11 (cxd6)} 27. f3 {0.70/30 4} Qc5+
{0.70/22 2 (Bc8)} 28. Kh2 {0.73/29 5} Qxc2 {0.75/22 8} 29. Ba1 {0.65/31 5} Qxb3
{0.80/22 0} 30. fxg4 {0.74/30 2} Qxg3+ {0.65/23 10} 31. Kxg3 {0.81/32 4} Rd8 {
0.80/22 9 (hxg4)} 32. gxh5 {0.87/26 5} Rd2 {0.64/21 1} 33. Ng6 {0.79/27 4
(Nxc6)} Rfd8 {0.84/18 2 (Rff2)} 34. Be5 {0.88/26 4} Rd1 {0.98/20 2 (R8d3+)} 35.
Ne7+ {0.92/29 6 (Rxd1)} Kh7 {1.00/21 2} 36. Rxd1 {0.97/32 1} Rxd1 {0.96/21 0}
37. Nxc6 {0.97/33 2} Rc1 {0.96/22 1} 38. Nd4 {0.97/35 1} Rc3+ {1.10/22 1} 39.
Kf4 {1.07/35 0} b3 {1.09/22 1} 40. Nxb3 {1.20/36 1} Rxb3 {1.12/23 1} 41. Bxc7 {
1.22/38 1} Rb2 {1.18/23 2 (Rb4)} 42. g3 {1.34/30 3} Rf2+ {1.38/24 0 (Rb4)} 43.
Ke5 {1.46/34 2 (Ke3)} Rf3 {1.02/21 3 (Ra2)} 44. Bxa5 {2.13/35 3} Rxg3 {1.32/20
1 (Ra3)} 45. Bb6 {1.76/30 4} Kh6 {2.50/21 3} 46. Kd5 {6.50/31 2} Rd3+ {2.89/20
0} 47. Kc4 {6.74/32 1} Rd7 {4.78/19 4 (Rh3)} 48. a5 {10.39/29 6} Re7 {6.91/21
1 (Rb7)} 49. Kd5 {25.05/30 6} Re8 {10.36/22 1 (Kxh5)} 50. a6 {58.60/38 8} Kh7 {
17.51/21 1 (Rf8)} 51. a7 {129.36/34 13} Kh6 {-#1780/26 0 (g6)} 52. Kc6 {151.60/
33 10} Kh7 {-#1781/28 0} 53. Kb7 {30.36/30 8} Re7+ {-#1784/27 0 (Rxe4)} 54. Bc7
{154.90/25 1} Rxe4 {-#1786/26 0 (Re8)} 55. a8=Q {154.92/26 1} Rb4+ {-#1788/26 1
} 56. Bb6 {154.93/28 0} Rb5 {-#1790/25 0 (Rf4)} 57. Qe8 {154.94/27 1 (Qf8)} Rb4
{-#1791/24 1} 58. Qg6+ {154.95/29 0} Kg8 {-#1794/25 0} 59. h6 {154.96/30 1} Kf8
{-#1796/26 0 (Rxb6+)} 60. h7 {#4/133 1} Rxb6+ {-#1798/29 6 (Ke7)} 61. Kxb6 {#3/
192 0} Ke7 {-#1799/9 0} 62. h8=Q {#2/192 0} Kd7 {-#1799/9 0} 63. Qhe8# {#1/192
0} 1-0[/pgn]

Regards from Spain.

Ajedrecista.

[Hurnavich]

The thing is Stockfish and Houdini see the mate using Syzygy

Komodo appears to not see it so it is a problem with Komodo is it not?

[syzygy]

Just a small difference in implementation.

[Hurnavich]

Hi

Not understanding this SF and H. Work fine K appears to have the probem

[Hurnavich]

when set to 5 man probe works no problem so looks like a 6 man K problem

Hope a fix can be found



Analysis by Komodo 7 64-bit MP: 5 man probe set

63.Qhe8#
+- (#1) Depth: 28 00:00:11 149MN, tb=7661491
(Hurnavich, 21.05.2014)

[lkaufman]

Quote from our Syzygy implementer Jeremy Bernstein:

"OK I do see the problem with the 6-man bases installed. This is the DTZ problem: the Syzygy bases are preferring a pawn move (resetting the DTZ clock) to a queen move, even if the queen move is mate! The solution to this would be to continue to search on the move list returned, but that might result in some performance drops. It would, in any case, need to be tested.

The "nonsense score" is a tablebase win, as explained in the documentation."

[Hurnavich]

I have sent you a private message to respect the forum rules

Stockfish and Houdini appear to have No problem with 6 man same situation

so must be the way K implements its use no?

[lkaufman]

I have sent you a private message to respect the forum rules

Stockfish and Houdini appear to have No problem with 6 man same situation

so must be the way K implements its use no?

Yes, as Jeremy explained there is a fix but it would have to be tested.