Komodo doesn't know the rule "corner promotion+wrong B&

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

Uri Blass
Posts: 10297
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Komodo doesn't know the rule "corner promotion+wron

Post by Uri Blass »

kgburcham wrote:
I do not understand why stockfish cannot find Rxf4 fast in the following position but it is not because of lack of knowledge about the bad bishop.


Stockfish will find it, just takes a little time.

[D] 5r1k/8/2p1B2P/1p1n2K1/p2R1P2/8/1P6/8 b - -

Engine: Stockfish 090913 64 SSE4.2 (4096 MB)
by Tord Romstad, Marco Costalba and Joona

33/61 0:26 +2.30++ 1...Kh7 2.Re4 Nf6 (340.499.432) 12847
33/61 0:33 +2.03++ 1...Rxf4 2.Rxf4 Nxf4 (439.052.024) 13215
33/61 0:33 +1.62++ 1...Rxf4 2.Rxf4 Nxf4 (451.208.770) 13343
33/61 0:34 +1.01++ 1...Rxf4 2.Rxd5 cxd5 3.Kxf4 b4 4.Bf5 a3 5.Bb1 axb2 (474.442.424) 13614
33/61 0:35 +0.09++ 1...Rxf4 2.Rxf4 Nxf4 (490.935.278) 13742
33/85 1:30 +2.84-- 1...Rxf4 2.Rxf4 Nxf4 (1.297.964.657) 14293
33/85 1:31 0.00 1...Rxf4 2.Rxf4 Nxf4 (1.317.064.656) 14343
34/85 1:32 0.00 1...Rxf4 2.Rxf4 Nxf4 (1.324.578.683) 14352
35/85 1:33 0.00 1...Rxf4 2.Rxf4 Nxf4 (1.341.953.028) 14396
36/85 1:35 0.00 1...Rxf4 2.Rxf4 Nxf4 (1.375.668.335) 14474
37/85 1:36 0.00 1...Rxf4 2.Rxf4 Nxf4 (1.407.709.794) 14539
38/85 1:38 0.00 1...Rxf4 2.Rxf4 Nxf4 3.Kxf4 b4 4.Ba2 c5 5.Bc4 a3 6.bxa3 bxa3 7.Kg5 a2 8.Bxa2 c4 9.Bb1 Kg8 10.Kh4 Kh8 11.Bc2 Kg8 12.Kg5 Kh8 13.Kg6 Kg8 14.Kf6 Kh8 (1.444.122.404) 14625
39/85 1:41 0.00 1...Rxf4 2.Rxf4 Nxf4 3.Kxf4 b4 4.Ba2 c5 5.Bc4 a3 6.bxa3 bxa3 7.Kg5 a2 8.Bxa2 c4 9.Bb1 Kg8 10.Kh4 Kh8 11.Bc2 Kg8 12.Kg5 Kh8 13.Kg6 Kg8 14.Kf6 Kh8 (1.496.151.915) 14732
I am not sure about it and it may be a question of luck.

I think that it is better to use only 1 thread so we have deterministic results.

I see non draw score for the following position for some reason
when at small depth it see a draw score(I use stockfish from 29.9 that is not the latest version but I do not know about specific bug that people fixed later).

[D]7k/8/7P/3p1B2/pp3K2/8/1P6/8 b - - 1 66

I use clear hash and 1024 mbytes so everybody who use stockfish from 29.9 should get the same result.

Note that the move that stockfish suggest is a3 and if I give it the position after a3 from clear hash stockfish see a constant draw score.

The score at depth 4-18 is a draw score and the score at higher depths is winning for white for some reason.
Top
User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 4:27 pm

Re: Komodo doesn't know the rule "corner promotion+wron

Post by Don »

Komodo DOES know that the wrong rook pawn is a dead draw against a bishop. I suspect Stockfish also knows this.

What Komodo doesn't have is rules that generalize this concept whether there are other things on the board, such as more pawns. It's currently handles as a very specific ending.

This is such a common theme however that would should indeed put some more work into this general class of ending.
Vinvin wrote:Move 61 to 66 :
Strelka eval : -0.60
Komodo 6 eval : -2.60
But it's clearly a draw endgame (even if white gives the 2 pawns) : white king to to a1.
[pgn]

[Site "GrahamCCRL.dyndns.org\Komodo Gauntlet.ele"]
[Date "10-09-2013"]
[White "Strelka 5.5 x64 "]
[Black "Komodo 6 64-bit "]
[Result "1/2-1/2"]
[TimeControl "4025:000 "]

1. c4 c5 2. g3 g6 3. Bg2 Bg7 4. Nc3 Nc6 5. Nf3 d6 6. O-O e6 7. b3
Nge7 8. Bb2 O-O 9. Qc2 b6 10. d4 cxd4 11. Nb5 e5 12. Ba3 f5 13. Nxd6
e4 14. Ng5 h6 15. Nh3 g5 16. Rad1 Be6 17. f4 g4 18. Nf2 Qd7 19. Nxg4
fxg4 20. Bxe4 Rf5 21. Nxf5 Nxf5 22. Bc1 Re8 23. e3 Qc8 24. Qg2 dxe3
25. Bxc6 e2 26. Bxe8 exf1=Q 27. Kxf1 Qxe8 28. Qe4 Qc8 29. Ba3 Bf7
30. Kf2 Bf6 31. Rd3 b5 32. cxb5 Qc2 33. Qe2 Bd4 34. Ke1 Qxe2 35. Kxe2
Be8 36. Rd2 Bxb5 37. Kd1 Bg1 38. Kc2 Nd4 39. Kc3 Nf3 40. Rd8 Kg7
41. Kb4 a6 42. Bb2 Kf7 43. Rh8 Bxh2 44. Rxh6 Bxg3 45. Kc5 Bf2 46. Kd5
g3 47. Rf6 Kg7 48. f5 Nh4 49. a4 Be8 50. Rg6 Kf8 51. Rg4 Bf7 52. Ke4
g2 53. Bf6 g1=Q 54. Rxg1 Bxg1 55. Bxh4 Bxb3 56. a5 Bh2 57. Kd3 Ba2
58. Kc2 Bc7 59. Be1 Be5 60. Bc3 Bxc3 61. Kxc3 Ke8 62. Kb2 Bg8 63. f6
Bf7 64. Kb1 Kd7 65. Kc2 Kc6 66. Kb2 Kb5 67. Kc2 Kxa5
1/2-1/2
[/pgn]
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
Top
Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 12:41 pm

Re: Komodo doesn't know the rule "corner promotion+wron

Post by Lyudmil Tsvetkov »

I do not know if this is the right thread, but probably it is still thematic.

I do not know if it is a wrong-coloured knight or something like this, but interesting how many engines immediately recognize the below position as drawn without tablebases?

[D]8/6kP/8/6N1/8/8/8/6K1 w - - 0 1

It is basically the same stuff as the wrong-coloured bishop and an edge pawn - stalemate motives, probably less frequent, but still equally objective. And of course, some more complex derivatives would also lead to a draw. So basically, a knight and an edge a or h pawn on the 7th rank with the enemy king controlling the promotion square of the pawn is a draw.
Top
User avatar
Laskos
Posts: 10948
Joined: Wed Jul 26, 2006 10:21 pm
Full name: Kai Laskos

Re: Komodo doesn't know the rule "corner promotion+wron

Post by Laskos »

Uri Blass wrote:
I am not sure about it and it may be a question of luck.

I think that it is better to use only 1 thread so we have deterministic results.

I see non draw score for the following position for some reason
when at small depth it see a draw score(I use stockfish from 29.9 that is not the latest version but I do not know about specific bug that people fixed later).

[D]7k/8/7P/3p1B2/pp3K2/8/1P6/8 b - - 1 66

I use clear hash and 1024 mbytes so everybody who use stockfish from 29.9 should get the same result.

Note that the move that stockfish suggest is a3 and if I give it the position after a3 from clear hash stockfish see a constant draw score.

The score at depth 4-18 is a draw score and the score at higher depths is winning for white for some reason.
SF sees it at very high depth, after 18 minutes search of an i7 quad. Depth 43 and higher - draw 0.00

Don't know what happens with Stockfish (5 Oct. 2013)

FEN: 7k/8/7P/3p1B2/pp3K2/8/1P6/8 b - - 1 66

Stockfish_13100518_x64_modern_sse42:

27/2 00:00 2,592 518,400 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
28/2 00:00 2,688 448,000 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
29/2 00:00 2,784 464,000 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
30/2 00:00 2,880 56,470 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
31/2 00:00 2,976 58,352 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
32/2 00:00 3,072 60,235 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
33/2 00:00 3,168 62,117 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
34/2 00:00 3,264 64,000 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
35/2 00:00 3,360 65,882 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
36/2 00:00 3,456 67,764 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
37/2 00:00 3,552 69,647 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
38/2 00:00 3,648 71,529 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-c3 Kg8-f8 Kc3-d3 Kf8-g8 Kd3-e3 Kg8-h8 Ke3-f3 Kh8-g8 Kf3-g4 Kg8-f7 Kg4-g5 Kf7-g8 Kg5-f6 Kg8-h8 Kf6-g6 Kh8-g8 Bb1-d3 b2-b1Q Bd3xb1 b3-b2 Kg6-g5 Kg8-h8 Kg5-h4 Kh8-g8 Kh4-h5 Kg8-f8 Kh5-g4 Kf8-g8 Kg4-g5
39/45 00:12 151,440,302 12,433,522 -3.53 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-a3 Kf7-g8 Ka3-a4 Kg8-f7 Ka4-b4 Kf7-g8 Kb4-a4
40/48+ 00:41 542,148,645 13,109,944 -3.45 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c3 Kh8-g8 Kc3-b4 Kg8-f7 Kb4-b5 Kf7-g8 Kb5-a5 Kg8-f8 Ka5-b6 Kf8-g8 Kb6-c5 Kg8-f8 Kc5-c4 Kf8-g8 Kc4-d5 Kg8-h8 Kd5-d6 Kh8-g8 Kd6-c5
40/48+ 00:54 712,847,916 13,147,809 -3.37 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-c3
40/48+ 01:09 921,642,169 13,242,746 -3.25 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-d5 Kg8-f7 Kd5-d6 Kf7-g8 Kd6-d5
40/48+ 01:27 1,167,731,173 13,311,573 -3.07 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-b5 Kg8-f8 Kb5-a5 Kf8-f7 Ka5-a4 Kf7-g8 Ka4-b5
40/48+ 01:50 1,481,541,637 13,459,628 -2.79 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-e3 Kh8-g8 Ke3-f3 Kg8-h8 Kf3-e2 Kh8-g8 Ke2-d2 Kg8-f7 Kd2-e1 Kf7-g8 Ke1-d2
40/48 02:21 1,893,210,850 13,388,192 -2.76 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5-d4 Kg8-f7 Kd4xd5 b4-b3 Kd5-d4 Kf7-g8 Kd4-c4 Kg8-h8 Kc4-c3 Kh8-g8 Kc3-d4 Kg8-f7 Kd4-e5 Kf7-g8 Ke5-f6 Kg8-h8 Kf6-e7 Kh8-g8 Ke7-d6 Kg8-f7 Bb1-d3 b2-b1B Bd3xb1 Kf7-g8 Kd6-e7 b3-b2 Ke7-d7 Kg8-h8 Kd7-c6 Kh8-g8 Kc6-d5 Kg8-h8 Kd5-d4 Kh8-g8 Kd4-c5 Kg8-f8 Bb1-h7 b2-b1R Bh7xb1 Kf8-g8 Kc5-d5 Kg8-h8 Kd5-c4 Kh8-g8 Bb1-d3 Kg8-h8 Kc4-b3 Kh8-g8 Bd3-f5 Kg8-h8 Bf5-e4 Kh8-g8 Be4-b1 Kg8-h8 Kb3-b4 Kh8-g8 Bb1-e4 Kg8-h8 Kb4-a3 Kh8-g8 Be4-f5 Kg8-h8 Bf5-e4
41/48- 02:21 1,893,853,693 13,388,382 -2.84 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5-d4 Kg8-f7 Kd4xd5 b4-b3 Kd5-d6 Kf7-g8 Kd6-e7 Kg8-h8 Ke7-d7 Kh8-g8 Kd7-d6 Kg8-f7 Kd6-e5 Kf7-g8 Ke5-d4 Kg8-f7 Kd4-c4 Kf7-g8 Kc4xb3 Kg8-h8 Kb3-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e5 Kg8-f7 Ke5-f5 Kf7-g8 Kf5-f6 Kg8-h8 Kf6-e6 Kh8-g8 Ke6-d5 Kg8-h8 Kd5-c4
41/48- 02:21 1,894,075,829 13,387,397 -2.94 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5-d4 Kg8-f8 Kd4xd5 b4-b3 Kd5-d4 Kf8-g8 Kd4-c5 Kg8-h8 Kc5-b5 Kh8-g8 Kb5-c5
41/48- 02:21 1,894,106,385 13,387,046 -3.08 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5-d4 Kg8-f8 Kd4xd5 b4-b3 Kd5-d4 Kf8-g8 Kd4-c5 Kg8-h8 Kc5-b5 Kh8-g8 Kb5-c4 Kg8-h8 Kc4xb3 Kh8-g8 Kb3-c3 Kg8-h8 Kc3-c4 Kh8-g8 Kc4-d5 Kg8-h8 Kd5-e6 Kh8-g8 Ke6-e5 Kg8-f7 Ke5-f5 Kf7-g8 Kf5-f6 Kg8-h8 Kf6-e6
41/48+ 03:05 2,491,243,333 13,443,870 -2.68 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5-d4 b4-b3 Kd4xd5 Kg8-h8 Kd5-c4 Kh8-g8 Kc4xb3 Kg8-f7 Kb3-c3 Kf7-g8 Kc3-d2 Kg8-h8 Kd2-e3 Kh8-g8 Ke3-f4 Kg8-f8 Kf4-g4 Kf8-f7 Kg4-f5 Kf7-g8 Kf5-g6 Kg8-h8 Kg6-g5 Kh8-g8 Kg5-f4
41/48+ 03:48 3,083,893,581 13,524,603 -2.41 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5-d4 Kg8-f7 Kd4xd5 Kf7-g8 Kd5-c4 b4-b3 Kc4-c3 Kg8-h8 Kc3xb2 Kh8-g8 Bb1-d3 Kg8-h8 Bd3-f5 Kh8-g8 Bf5-e4 Kg8-h8 Be4-d3 Kh8-g8 Kb2-a3 b3-b2 Bd3-b1 Kg8-f7 Bb1-h7 b2-b1R Bh7xb1 Kf7-g8 Bb1-f5 Kg8-h8 Bf5-e4 Kh8-g8 Be4-f5
41/48 04:31 3,728,210,092 13,724,365 -2.28 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5-d4 Kg8-f7 Kd4xd5 Kf7-g8 Kd5-c4 b4-b3 Kc4-c3 Kg8-h8 Kc3-d2 Kh8-g8 Kd2-e3 Kg8-h8 Ke3-d4 Kh8-g8 Kd4-d5 Kg8-f8 Kd5-e5 Kf8-f7 Bb1-d3 Kf7-g8 Ke5-f6 b2-b1Q Bd3xb1 Kg8-h8 Kf6-e5 Kh8-g8 Ke5-d4 Kg8-h8 Kd4-c4 Kh8-g8 Kc4-b5 b3-b2 Kb5-c4 Kg8-h8 Kc4-b4 Kh8-g8 Kb4-c5 Kg8-h8 Kc5-d5 Kh8-g8 Kd5-c4
42/48- 06:37 5,501,159,839 13,841,694 -2.36 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 b4-b3 Kd5-c4 Kg8-h8 Kc4-c3 Kh8-g8 Kc3xb3 Kg8-f7 Kb3-c3 Kf7-g8 Kc3-d4 Kg8-h8 Kd4-e3 Kh8-g8 Ke3-e2 Kg8-h8 Ke2-e1 Kh8-g8 Ke1-f2 Kg8-f7 Kf2-e1 Kf7-g8
42/48- 06:38 5,511,925,955 13,837,308 -2.44 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 b4-b3 Kd5-c4 Kg8-f8 Kc4xb3 Kf8-g8 Kb3-c4 Kg8-h8 Kc4-d5 Kh8-g8 Kd5-c4
42/48+ 07:35 6,347,366,280 13,939,747 -2.20 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 b4-b3 Kd5-e6 Kg8-h8 Ke6-e5 Kh8-g8 Ke5-f4 Kg8-f7 Kf4-f5 Kf7-g8 Kf5-f6 Kg8-h8 Kf6-e5
42/49- 07:37 6,366,580,355 13,929,419 -2.56 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 b4-b3 Kd5-e5 Kg8-h8 Ke5-d6 Kh8-g8 Kd6-c5 Kg8-h8 Kc5-b4 Kh8-g8 Kb4xb3 Kg8-h8 Kb3-a4 Kh8-g8 Ka4-b4 Kg8-h8 Kb4-a5 Kh8-g8 Ka5-b6 Kg8-f7 Bb1-e4 b2-b1B h6-h7 Kf7-g7
42/49- 07:52 6,556,645,590 13,881,875 -2.83 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 b4-b3 Kd5-d4 Kg8-h8 Kd4-c5 Kh8-g8 Kc5-b5 Kg8-h8 Kb5-a6 Kh8-g8 Ka6-b6 Kg8-h8 Kb6-c7 Kh8-g8 Kc7-b6
42/55+ 08:40 7,291,098,510 14,001,687 -2.02 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 Kg8-f8 Kd5-e4 b4-b3 Ke4-d3 Kf8-g8 Kd3-c4 Kg8-h8 Kc4-c3 Kh8-g8 Kc3-d2 Kg8-f8 Kd2-d3
42/55+ 09:44 8,317,665,247 14,219,908 -1.40 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 b4-b3 Kd5-c5 Kg8-h8 Kc5-b4 Kh8-g8 Kb4-c3 Kg8-f8 Kc3-c4 Kf8-f7 Kc4-d4 Kf7-g8 Kd4-e3 Kg8-h8 Ke3-e2 Kh8-g8 Ke2-d1 Kg8-f7 Kd1-e1 Kf7-f6 Ke1-f2 Kf6-f7 Kf2-g3 Kf7-g8 Kg3-f3 Kg8-h8 Kf3-e2
42/55 11:19 9,883,105,426 14,540,928 -1.23 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 b4-b3 Kd5-c5 Kg8-h8 Kc5-b4 Kh8-g8 Kb4-c3 Kg8-f8 Kc3-c4 Kf8-f7 Kc4-b4 Kf7-g8
43/55+ 14:32 13,142,066,440 15,070,157 -1.15 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 b4-b3 Kd5-d4 Kg8-h8 Kd4-e3 Kh8-g8 Ke3-e2 Kg8-h8 Ke2-f3 Kh8-g8 Kf3-g2 Kg8-h8 Kg2-f3
43/55+ 15:38 14,242,913,894 15,183,404 -1.07 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 Kg8-f7 Kd5-c4 b4-b3 Kc4-d4 Kf7-g8 Kd4-e5 Kg8-h8 Ke5-f4 Kh8-g8 Kf4-f3 Kg8-h8 Kf3-e3 Kh8-g8 Ke3-f4 Kg8-h8 Kf4-f3 Kh8-g8 Kf3-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-f3
43/55- 15:38 14,243,068,089 15,183,244 -1.31 a4-a3 Bf5-b1 a3xb2 Kf4-e5 Kh8-g8 Ke5xd5 b4-b3 Kd5-c6 Kg8-f8 Kc6-b6 Kf8-f7 Kb6-b5 Kf7-g8 Kb5-a4 Kg8-f8 Ka4-a5 Kf8-g8 Ka5-b6 Kg8-h8 Kb6-b5 Kh8-g8
43/55+ 16:42 15,261,062,390 15,227,181 -0.94 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c5 Kh8-g8 Kc5-c4 Kg8-h8 Kc4-c3 Kh8-g8 Kc3-b4 Kg8-h8 Kb4-c3
43/55+ 17:13 15,745,255,980 15,239,133 -0.67 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c5 Kh8-g8 Kc5-b5 Kg8-f8 Kb5-a4 Kf8-f7 Bb1-f5 b2-b1R Bf5xb1 Kf7-g8 Ka4-b5 Kg8-h8 Kb5-c4 Kh8-g8 Kc4-c3 b3-b2 Kc3-b4 Kg8-h8 Kb4-c5 Kh8-g8 Kc5-c6 Kg8-h8 Kc6-b5 Kh8-g8 Kb5-c5 Kg8-h8 Kc5-c4 Kh8-g8 Kc4-d4 Kg8-h8 Kd4-c4
43/55+ 17:37 16,146,794,196 15,264,188 -0.26 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
43/55 17:44 16,235,142,834 15,250,938 0.00 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
44/55 17:48 16,281,937,284 15,237,455 0.00 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
45/55 17:50 16,301,970,897 15,230,134 0.00 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
46/55 17:55 16,359,050,667 15,211,608 0.00 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
47/55 18:00 16,418,967,104 15,189,414 0.00 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
48/55 18:06 16,483,647,597 15,168,284 0.00 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
49/55 18:14 16,568,028,212 15,139,717 0.00 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
50/55 18:22 16,655,503,823 15,111,761 0.00 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
51/55 18:32 16,771,719,839 15,075,554 0.00 a4-a3 Bf5-b1 a3xb2 Kf4-e5 d5-d4 Ke5xd4 b4-b3 Kd4-c4 Kh8-g8 Kc4-c3 Kg8-h8 Kc3-d4 Kh8-g8 Kd4-e3 Kg8-h8 Ke3-d2 Kh8-g8 Kd2-e2 Kg8-h8 Ke2-d1 Kh8-g8 Kd1-d2 Kg8-h8 Kd2-e2 Kh8-g8 Ke2-e1 Kg8-f7 Ke1-d2 Kf7-g8
Top
User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 4:27 pm

Re: Komodo doesn't know the rule "corner promotion+wron

Post by Don »

michiguel wrote:
Laskos wrote:Just happened in TCEC, both Komodo and Stockfish were blind of the wrong bishop, Stockfish blundered and lost the game. In this drawn position

[D]8/6k1/2p2r2/1p1n1BKP/p2R1P2/8/1P6/8 b - - 1 61[/D]

Stockfish (black) had played b4 with -2.56 evaluation and lost. Houdini 3 (without TBs) plays Rf8 (with a correct drawish evaluation) and easily draws Komodo

The draw, evaluated by Komodo at 2.76, but by Houdini at 0.03 is:
[D]7k/8/6K1/2p4P/2B5/p7/8/8 b - - 1 69[/D]

So, the game Komodo-Stockfish, won by Komodo, was a game of two blind engines of the wrong bishop.
This is a critical position after Rf8, h6+, Kh8. If Be6 we reach
[D]5r1k/8/2p1B2P/1p1n2K1/p2R1P2/8/1P6/8 b - - 6 69

Then, Rxf4 draws and it needs to be detected quickly in the tree.

For instance, Gaviota outputs

Code: Select all

    201217  13       0.2    -0.16  Rxf4 36.Rxd5 cxd5 37.Kxf4 b4 38.Bf5 a3
                                   39.Bb1 axb2 40.Ke5 d4 41.Kxd4 Kg8
                                   42.h7+ Kg7
Miguel
We just added code to handle these type of endings in a more general way.

Now Komodo plays Nxf4 with close to a draw score although it still takes 5-10 seconds.

The previous version was not finding it even with a very deep search and I did not let it run long enough to see if it would find it in any reasonable amount of time.

Don
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
Top
User avatar
Laskos
Posts: 10948
Joined: Wed Jul 26, 2006 10:21 pm
Full name: Kai Laskos

Re: Komodo doesn't know the rule "corner promotion+wron

Post by Laskos »

Don wrote:
We just added code to handle these type of endings in a more general way.

Now Komodo plays Nxf4 with close to a draw score although it still takes 5-10 seconds.

The previous version was not finding it even with a very deep search and I did not let it run long enough to see if it would find it in any reasonable amount of time.

Don
Does it see now this as a draw?
[D]7k/8/6K1/2p4P/2B5/p7/8/8 b - - 1 69[/D]

The previous Komodo didn't, with 2.80 for white.
Top
User avatar
michiguel
Posts: 6401
Joined: Thu Mar 09, 2006 8:30 pm
Location: Chicago, Illinois, USA

Re: Komodo doesn't know the rule "corner promotion+wron

Post by michiguel »

Don wrote:
michiguel wrote:
Laskos wrote:Just happened in TCEC, both Komodo and Stockfish were blind of the wrong bishop, Stockfish blundered and lost the game. In this drawn position

[D]8/6k1/2p2r2/1p1n1BKP/p2R1P2/8/1P6/8 b - - 1 61[/D]

Stockfish (black) had played b4 with -2.56 evaluation and lost. Houdini 3 (without TBs) plays Rf8 (with a correct drawish evaluation) and easily draws Komodo

The draw, evaluated by Komodo at 2.76, but by Houdini at 0.03 is:
[D]7k/8/6K1/2p4P/2B5/p7/8/8 b - - 1 69[/D]

So, the game Komodo-Stockfish, won by Komodo, was a game of two blind engines of the wrong bishop.
This is a critical position after Rf8, h6+, Kh8. If Be6 we reach
[D]5r1k/8/2p1B2P/1p1n2K1/p2R1P2/8/1P6/8 b - - 6 69

Then, Rxf4 draws and it needs to be detected quickly in the tree.

For instance, Gaviota outputs

Code: Select all

    201217  13       0.2    -0.16  Rxf4 36.Rxd5 cxd5 37.Kxf4 b4 38.Bf5 a3
                                   39.Bb1 axb2 40.Ke5 d4 41.Kxd4 Kg8
                                   42.h7+ Kg7
Miguel
We just added code to handle these type of endings in a more general way.

Now Komodo plays Nxf4 with close to a draw score although it still takes 5-10 seconds.

The previous version was not finding it even with a very deep search and I did not let it run long enough to see if it would find it in any reasonable amount of time.

Don
I need to make a correction. It takes 22 seconds for Gaviota to fail high in my3.2 Ghz AMD (1 thread). The really fast time I reported before must have been because I was analyzing the game and the hash tables must have been full.

It is a very nice instructive endgame. After
... Rxf4/
Rxf4 Nxf4/
Kxf4 b4 (threatens a3 and if b3, a2 wins so..)
Ba2 (if a3 now b3 winning, so at this point black need to get the B out of a2)
... c5/
Ke4 c4!/
Bxc4 a3 draws.

Miguel
Top
User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 4:27 pm

Re: Komodo doesn't know the rule "corner promotion+wron

Post by Don »

Laskos wrote:
Don wrote:
We just added code to handle these type of endings in a more general way.

Now Komodo plays Nxf4 with close to a draw score although it still takes 5-10 seconds.

The previous version was not finding it even with a very deep search and I did not let it run long enough to see if it would find it in any reasonable amount of time.

Don
Does it see now this as a draw?
[D]7k/8/6K1/2p4P/2B5/p7/8/8 b - - 1 69[/D]

The previous Komodo didn't, with 2.80 for white.
With the change it now see's this as -8:

position fen 7k/8/6K1/2p4P/2B5/p7/8/8 b - - 1 69
go depth 12
info depth 1 time 14 nodes 4 nps 285 hashfull 0
info depth 1 time 14 nodes 21 score cp 2 nps 1499 hashfull 0 pv a3a2 c4a2
info depth 2 time 14 nodes 21 nps 1499 hashfull 0
info depth 2 time 14 nodes 59 score cp 2 nps 4214 hashfull 0 pv a3a2 c4a2
info depth 3 time 14 nodes 59 nps 4214 hashfull 0
info depth 3 time 14 nodes 82 score cp -8 nps 5857 hashfull 0 pv a3a2 c4a2 c5c4
info depth 4 time 14 nodes 82 nps 5857 hashfull 0
info depth 4 time 15 nodes 160 score cp -8 nps 10666 hashfull 0 pv a3a2 c4a2 c5c4 h5h6 c4c3 a2b1
info depth 5 time 15 nodes 160 nps 10666 hashfull 0
info depth 5 time 15 nodes 376 score cp -8 nps 25066 hashfull 0 pv a3a2 c4a2 c5c4 a2b1 c4c3 b1c2
info depth 6 time 15 nodes 376 nps 25066 hashfull 0
info depth 6 time 15 nodes 541 score cp -8 nps 36066 hashfull 0 pv a3a2 c4a2 c5c4 a2b1 c4c3 b1c2 h8g8 h5h6 g8h8
info depth 7 time 15 nodes 541 nps 36066 hashfull 0
info depth 7 time 15 nodes 944 score cp -8 nps 62932 hashfull 0 pv a3a2 c4a2 c5c4 a2b1 c4c3 b1c2 h8g8 c2a4
info depth 8 time 15 nodes 944 nps 62932 hashfull 0
info depth 8 time 16 nodes 1758 score cp -8 nps 109874 hashfull 0 pv a3a2 c4a2 c5c4 a2b1 c4c3 b1c2 h8g8 c2a4 c3c2 a4c2
info depth 9 time 16 nodes 1758 nps 109874 hashfull 0
info depth 9 time 16 nodes 3337 score cp -8 nps 208562 hashfull 0 pv a3a2 c4a2 c5c4 a2b1 c4c3 b1c2 h8g8 c2a4 c3c2 a4c2 g8h8 h5h6 h8g8
info depth 10 time 16 nodes 3337 nps 208562 hashfull 0
info depth 10 time 16 nodes 5608 score cp -8 nps 329883 hashfull 0 pv a3a2 c4a2 c5c4 a2b1 c4c3 b1c2 h8g8 c2a4 c3c2 a4c2 g8h8 c2b1 h8g8 b1a2 g8h8
info depth 11 time 16 nodes 5608 nps 329883 hashfull 0
info depth 11 time 19 nodes 9589 score cp -8 nps 504682 hashfull 0 pv a3a2 c4a2 c5c4 a2b1 c4c3 b1c2 h8g8 c2a4 c3c2 a4c2 g8h8 c2b1 h8g8 h5h6
info depth 12 time 19 nodes 9589 nps 504682 hashfull 0
info depth 12 time 21 nodes 15470 score cp -8 nps 736661 hashfull 0 pv a3a2 c4a2 c5c4 a2b1 c4c3 b1c2 h8g8 c2a4 c3c2 a4c2 g8h8 c2b1 h8g8 b1e4 g8h8 e4c2
bestmove a3a2 ponder c4a2
Capital punishment would be more effective as a preventive measure if it were administered prior to the crime.
Top
gladius
Posts: 568
Joined: Tue Dec 12, 2006 10:10 am
Full name: Gary Linscott

Re: Komodo doesn't know the rule "corner promotion+wron

Post by gladius »

Uri Blass wrote:I am not sure about it and it may be a question of luck.

I think that it is better to use only 1 thread so we have deterministic results.

I see non draw score for the following position for some reason
when at small depth it see a draw score(I use stockfish from 29.9 that is not the latest version but I do not know about specific bug that people fixed later).

[D]7k/8/7P/3p1B2/pp3K2/8/1P6/8 b - - 1 66

I use clear hash and 1024 mbytes so everybody who use stockfish from 29.9 should get the same result.

Note that the move that stockfish suggest is a3 and if I give it the position after a3 from clear hash stockfish see a constant draw score.

The score at depth 4-18 is a draw score and the score at higher depths is winning for white for some reason.
It appears related to futility pruning. With this patch: http://tests.stockfishchess.org/tests/v ... 3ba5ae713d, SF now sees this as a draw starting at ply 1, up to ply 35 in single thread mode.

SF (with the patch) now finds Nxf4, which causes a fail high at depth 26 single threaded:
[d]5r1k/8/2p1B2P/1p1n2K1/p2R1P2/8/1P6/8 b - - 6 69[/d]

info depth 26 seldepth 42 score cp -206 lowerbound nodes 27120293 nps 2669845 time 10158 multipv 1 pv d5f4 d4f4 f8f4 g5f4 b5b4 e6a2 c6c5 f4g5 h8h7 a2b1 h7g8 g5g6 a4a3 b2b3 c5c4 b3c4 b4b3

It does switch over to Rxf4 once it resolves the fail high, that happens at 30 seconds:
info depth 26 seldepth 50 score cp 0 nodes 85830792 nps 2759122 time 31108 multipv 1 pv f8f4 d4d5 c6d5 g5f4 b5b4 e6d5 a4a3 b2a3 b4a3 f4e5 a3a2 d5a2 h8h7 e5d5 h7h6
Top
Vinvin
Posts: 5228
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Komodo doesn't know the rule "corner promotion+wron

Post by Vinvin »

Don wrote:We just added code to handle these type of endings in a more general way.

Now Komodo plays Nxf4 with close to a draw score although it still takes 5-10 seconds.

The previous version was not finding it even with a very deep search and I did not let it run long enough to see if it would find it in any reasonable amount of time.

Don
Thanks for the improvement, Don !
Does it play Bc7 with a winning score here ?
[d]5k2/8/p7/P3bP2/8/2B5/b1K5/8 b - - 0 1
Top

Return to “Computer Chess Club: General Topics”