Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Season 1

[BB+]

On move 48, Houdini finally pushes d5, and something might now happen. Black has 50+ minutes in reserve, while Rybka has only 2 minutes in addition to the 30-second increments. Rybka's time management is really poor this game. I don't know whether this is an R4 phenomenon, or maybe because of the plodding pace of the game with "many" choices (often 10 or move moves would be within 0.20 of the best move), or what.

[BB+]

4qbk1/5pp1/p7/P1p2b1p/7P/B1P2NP1/3r1P2/R4QK1 b - - 1 62Does your engine like Bd3 for Black here? The mainline is quite neat. How fast to find the winning move after 62... Bd3 63. Qg2 Be4 64. Re1? Nice win by Houdini.

[Uly]

Rybka's time management is really poor this game.

Have you checked other games? My impression is that this is something common from what I've read at RF, optimal settings have been found for repeating time controls, but no matter how much you tweak the TC settings, you never achieve the same performance for repeating time controls.

I think the time management of Rybka should be rewritten.

[MichaelIsGreat]

In the following position, Houdini 1.5a seems to get "carried away" by its knowledge of endgame databases and it plays a completely "illogical" but yet still logical (!) continuation.
Here is the position:
8/8/8/4N3/2K1k3/5n2/4r3/R7 b - - 0 1
Houdini 1.5 in this position plays 81. ... Ra2 82. Rxa2 Nxe5+ 83. Kb3 Nf3 84. Kc4 Kf4 85. Kc5 Ne5 86. Kd5 Nd3 87. Kd4 Ne5 ½-½
Houdini 1.5 assumes wrongly that its opponent knows the endgame tablebases for this particular continuation!!! That is a BIG mistake not only for opponents who are humans but even for many computer programs that even today do not support yet tablebases!! Moreover, implementations might be done differently from one program to another and one program might see earlier the result of an endgame database than its opponent!


If Robert Houdart (the programmer of Houdini 1.5) reads this post, I suggest that he changes his program to avoid its program to get "carried away" like that in such positions!
I also ask Robert Houdart to consider implementing in a future update the triplebases to have the possibility to have them in memory, as I think his program should benefit quite often and quite significantly by having these triplebases than by keeping using the Gaviota tablebases. A win by Shredder 12 against Rybka 4 in the division 1 was quite interesting on this matter. Shredder 12 is known to use a similar implementation of the triplebases that he calls Shredderbases and it paid off very well in this particular position.
5k2/2p5/5PP1/8/5n2/BK6/p7/8 b - - 18 114

[Event "TCEC - Division I - S1"]
[Site "http://www.tcec-chess.org"]
[Date "2011.01.19"]
[Round "13.1"]
[White "Shredder 12.0"]
[Black "Rybka 4.0"]
[Result "1-0"]
[SetUp "1"]
[FEN "5k2/2p5/5PP1/8/5n2/BK6/p7/8 b - - 18 114"]
[Variant "normal"]

1. d4 f5 2. Nf3 Nf6 3. g3 g6 4. Bg2 Bg7 5. c4 O-O 6. Nc3 d6 7. O-O Nc6 8. d5 Ne5 9. Nxe5 dxe5 10. Qb3 h6 11. e4 f4 12. c5 Kh7 13. Rd1 b6 14. c6 Ne8 15. a4 Nd6 16. f3 a5 17. Qc2 Ba6 18. b3 h5 19. Qf2 Rf7 20. Bf1 Qc8 21. Qg2 Bxf1 22. Rxf1 Qa6 23. Bb2 fxg3 24. hxg3 Qd3 25. Qe2 Qxe2 26. Nxe2 Raf8 27. Kg2 g5 28. Nc3 h4 29. g4 Kg6 30. Ba3 Rc8 31. Rfd1 Rcf8 32. Rd3 Rh8 33. b4 axb4 34. Bxb4 Ra8 35. Ra3 Rff8 36. Nd1 Ra6 37. Ra2 Rfa8 38. Nc3 Bf8 39. Rd1 Bg7 40. Rda1 Rb8 41. Rd1 Rba8 42. Rh1 Bf8 43. Rha1 Rb8 44. Kh3 Bg7 45. Rf1 Rba8 46. Re1 Rf8 47. Kg2 Rfa8 48. Rb1 Bf6 49. Rbb2 Rc8 50. Rb1 Rca8 51. Kh2 Rb8 52. Kh3 Rba8 53. Rd2 Rf8 54. Rf2 Rb8 55. Ra2 Rba8 56. Rc1 Bg7 57. Rf1 Rc8 58. Rff2 Rca8 59. Kh2 Bf8 60. Rf1 Kf6 61. Kh3 Rc8 62. Rb1 Nc4 63. Rc2 Raa8 64. Rbc1 Ra6 65. Rf2 Kg6 66. Ra2 Kf7 67. Rac2 Kf6 68. Ra2 Kf7 69. Nb5 Nd6 70. Rca1 Kf6 71. Nc3 Nf7 72. Rd1 Nd6 73. Rc1 Nc4 74. a5 bxa5 75. Bc5 Nd6 76. Na4 Raa8 77. Bf2 e6 78. Nc5 exd5 79. Nd7+ Kg7 80. exd5 Nb5 81. Nxe5 a4 82. Nd7 Kg6 83. Re1 Nc3 84. Re6+ Kf7 85. Rf6+ Kg8 86. Rd2 a3 87. Bd4 Nxd5 88. Rg6+ Kh7 89. Nxf8+ Rxf8 90. Rxg5 Nf4+ 91. Kxh4 Ne6 92. Rh5+ Kg6 93. Be3 Rfd8 94. Rxd8 Nxd8 95. Rh8 Rb8 96. f4 Nxc6 97. f5+ Kf7 98. Rxb8 Nxb8 99. Bc1 a2 100. Bb2 Nc6 101. g5 Ne7 102. Kg4 Kg8 103. g6 Kf8 104. Kg5 Nc8 105. f6 Nd6 106. Ba1 Kg8 107. Kf4 Kf8 108. Ke5 Nb7 109. Kd5 Nd8 110. Bd4 Ke8 111. Bb2 Kf8 112. Kc4 Ne6 113. Kb3 Nf4 114. Ba3+ 1-0


Again, it is a mistake for Houdini 1.5 to play this way for several reasons:
1) against a human, Houdini 1.5 might win later on even with a just a knight as an advantage and even in drawn positions.
2) against computers, same story. The computer opponent to Houdini 1.5 might not have tablebases and again Houdini 1.5 could again win even in drawn positions! Especially if it is programmed to play the most complicated variations in such a case. Complicated of course should have to be defined clearly and programmed following the chosen definition of this expression "complicated variation".


This Elite tournament is great. Thanks to the organizer Martin. For this Elite match only, I regret that Martin did not choose to let the games end by either a mat or a forced draw. It might not have been possible to do that easily for you for this Elite match. It would have been more interesting to see the continuation up to the complete win (mat). Sometimes, programs might find drawn positions even in the most unusual lost positions, especially when a program has tablebases and the opponent does not or if the opponent's implementation is not done as well as the other program's implementation.
Martin, plan this idea for the next Elite match; just let the game finish either by a mat or by a forced draw.

[MichaelIsGreat]

Correcting the diagram of my previous post where the second diagram should be as follows at position 113 ... Nf4 :
5k2/2p5/5PP1/8/5n2/1K6/pB6/8 w - - 17 114
And if you want to review this particular game in full, here it is:
[SetUp "1"]
[Event "TCEC - Division I - S1"]
[Site "http://www.tcec-chess.org"]
[Date "2011.01.19"]
[Round "13.1"]
[White "Shredder 12.0"]
[Black "Rybka 4.0"]
[Result "1-0"]
[Variant "normal"]
1. d4 f5 2. Nf3 Nf6 3. g3 g6 4. Bg2 Bg7 5. c4 O-O 6. Nc3 d6 7. O-O Nc6 8. d5 Ne5 9. Nxe5 dxe5 10. Qb3 h6 11. e4 f4 12. c5 Kh7 13. Rd1 b6 14. c6 Ne8 15. a4 Nd6 16. f3 a5 17. Qc2 Ba6 18. b3 h5 19. Qf2 Rf7 20. Bf1 Qc8 21. Qg2 Bxf1 22. Rxf1 Qa6 23. Bb2 fxg3 24. hxg3 Qd3 25. Qe2 Qxe2 26. Nxe2 Raf8 27. Kg2 g5 28. Nc3 h4 29. g4 Kg6 30. Ba3 Rc8 31. Rfd1 Rcf8 32. Rd3 Rh8 33. b4 axb4 34. Bxb4 Ra8 35. Ra3 Rff8 36. Nd1 Ra6 37. Ra2 Rfa8 38. Nc3 Bf8 39. Rd1 Bg7 40. Rda1 Rb8 41. Rd1 Rba8 42. Rh1 Bf8 43. Rha1 Rb8 44. Kh3 Bg7 45. Rf1 Rba8 46. Re1 Rf8 47. Kg2 Rfa8 48. Rb1 Bf6 49. Rbb2 Rc8 50. Rb1 Rca8 51. Kh2 Rb8 52. Kh3 Rba8 53. Rd2 Rf8 54. Rf2 Rb8 55. Ra2 Rba8 56. Rc1 Bg7 57. Rf1 Rc8 58. Rff2 Rca8 59. Kh2 Bf8 60. Rf1 Kf6 61. Kh3 Rc8 62. Rb1 Nc4 63. Rc2 Raa8 64. Rbc1 Ra6 65. Rf2 Kg6 66. Ra2 Kf7 67. Rac2 Kf6 68. Ra2 Kf7 69. Nb5 Nd6 70. Rca1 Kf6 71. Nc3 Nf7 72. Rd1 Nd6 73. Rc1 Nc4 74. a5 bxa5 75. Bc5 Nd6 76. Na4 Raa8 77. Bf2 e6 78. Nc5 exd5 79. Nd7+ Kg7 80. exd5 Nb5 81. Nxe5 a4 82. Nd7 Kg6 83. Re1 Nc3 84. Re6+ Kf7 85. Rf6+ Kg8 86. Rd2 a3 87. Bd4 Nxd5 88. Rg6+ Kh7 89. Nxf8+ Rxf8 90. Rxg5 Nf4+ 91. Kxh4 Ne6 92. Rh5+ Kg6 93. Be3 Rfd8 94. Rxd8 Nxd8 95. Rh8 Rb8 96. f4 Nxc6 97. f5+ Kf7 98. Rxb8 Nxb8 99. Bc1 a2 100. Bb2 Nc6 101. g5 Ne7 102. Kg4 Kg8 103. g6 Kf8 104. Kg5 Nc8 105. f6 Nd6 106. Ba1 Kg8 107. Kf4 Kf8 108. Ke5 Nb7 109. Kd5 Nd8 110. Bd4 Ke8 111. Bb2 Kf8 112. Kc4 Ne6 113. Kb3 Nf4 114. Ba3+ 1-0


Well, first time I post diagrams. It definitively is nicer with these diagrams

[Martin Thoresen]

Michael,

Thanks for your praise!

As for letting the games continue until mate, that is not going to happen simply because it takes too long and people get bored when the engines makes another 20 moves in a totally won position.
I agree that watching mates are fun, but it's unpractical.
As a slight compensation, I have increased the adjudication threshold starting with this ongoing Elite Match from +/- 5.25 to +/- 6.50

I might also consider raising it another notch for the upcoming Season 2.

Best,
Martin

[MichaelIsGreat]

Hello Martin,

Yes, I think you could raise the adjudication threshold up to say +/- 8.00 I think (at least +/- 7.50 is necessary I believe). Just to be on the safe side because unfortunately computer evaluations are far from being accurate even when they reach high values. I could even say sometimes especially when they reach high values, not seeing a drawn position for instance and thinking that the position is won, wrongly!
For the other points of your remarks, you might be right, games might be too long for quite a few impatient people, I do not know for sure on this point. I do not understand that because people could always come back later on at the end of the game to check how it was won up to the mat or to a forced draw.

Great matches at the Elite tournament between Houdini 1.5 and Rybka 4.0. Highly recommended for those who have not checked them out. That will give you a good idea of how good Houdini 1.5 has become even against such a worth opponent like Rybka 4.0! Check either Martin's excellent site at http://www.tcec-chess.org/elite_match.php or the ChessBomb.com site at http://livechess.chessdom.com/site/#tab ... 11-tcecs1e

[BB+]

Rybka has dodgy time management again (Game 12), but can someone explain the accounting in the PGN output?

50.Rf3 { ev=-0.01, d=23, pd=Rh5, mt=00:01:17, tl=00:41:13, s=16647 kN/s,   }
Rh5 { ev=-0.15, d=21, pd=Re1, mt=00:01:06, tl=00:06:15, s=627 kN/s,   }
51.Re1 { ev=-0.01, d=24, pd=Qd7, mt=00:01:04, tl=00:40:42, s=17234 kN/s,   }
Rc7 { ev=-0.15, d=21, pd=Rf1, mt=00:02:11, tl=00:04:36, s=627 kN/s,   }
52.Bg1 { ev=-0.01, d=24, pd=Kf8, mt=00:01:51, tl=00:39:23, s=16839 kN/s,   }
Qh8 { ev=-0.15, d=21, pd=Rh3, mt=00:01:37, tl=00:03:31, s=611 kN/s,   }
53.h3 { ev=-0.01, d=24, pd=Qc8, mt=00:03:18, tl=00:36:37, s=17193 kN/s,   }
Qc8 { ev=-0.31, d=20, pd=Ref1, mt=00:01:27, tl=00:02:36, s=605 kN/s,   }
54.Rff1 { ev=-0.01, d=25, pd=Qd7, mt=00:01:22, tl=00:35:47, s=17169 kN/s,  }
Qd7 { ev=-0.27, d=20, pd=Rb1, mt=00:01:25, tl=00:01:43, s=593 kN/s,   }

So here are the move times (mt), and time left (tl):

.     mt    tl    mt    tl
50  01:17 41:13  01:06 06:15
51  01:04 40:42  02:11 04:36
52  01:51 39:23  01:37 03:31
53  03:18 36:37  01:27 02:36
54  01:22 35:47  01:25 01:43

It seems to me that there's usually an extra ~2 seconds floating around? For instance, White's move 51, started with 41:13, took 01:04, so should have 40:09+0:30 or 40:39 left. Three extra seconds added. Black's move 54, started with 02:36, took 01:25, so should have 01:11+0:30 or 01:41 left. Two extra seconds added. Is this a feature of ChessGUI?

[Martin Thoresen]

BB+,

I am using the "Lag-free timing" feature so that might have something to do with it. I am sure Matthias can elaborate further.

Best,
Martin

[Matthias Gemuh]

BB+,

I am using the "Lag-free timing" feature so that might have something to do with it. I am sure Matthias can elaborate further.

Best,
Martin

Correct. Engines receive extra time to compensate for measured lag. See GUI log file.
The extra time is always _average_ lag per move in milliseconds.

Matthias.