Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka vs Zappa in Mexico City
1 2 Previous Next  
Parent - - By Alkelele (****) Date 2007-09-29 19:17 Edited 2007-09-29 19:22
Yes, ok, but one problem is that this is just one exit from book. Our repertoire is based on thousands of different exits from book. Naturally, we can't test each of these 20 times. We can try to figure out ourselves if Rybka can handle them all right (based on our experience with positions of a similar type), and I wouldn´t see any reason to predict that Rybka would not only fail to win this ending, but even outright lose it. Interestingly, Jeroon and I had discussed this exact ending on MSN before the match, agreeing that this was way better than Rybka and other engines initially think. Furthermore, Zappa often has a problem in overestimating having a piece for 3 pawns (here, the bishop), and, sure enough, during the game, Rybka realised far earlier than Zappa how critical this was for black. In one sense, this comes down to stuff like: Can the bookmakers be responsible for not realising that Rybka may miss winning continuatins like Qe1 (instead of gxh5). I just don´t think this would be fair with regard to this particular line.

In other lines, sure, you have to take into consideration systematical middlegame or endgame weaknesses and not just the objective evaluation. This is a constant situation of applying subtle judgement to 100s of different lines and then extract a single conclusion about what to play 5, 10, or 15 moves earlier. I think the last game is a good example for this discussion. White got a huge advantage out of the opening, and Zappa even followed up with the horrible b5-b4. On the other hand, this was a closed position prompting a king-attack, and this is not Rybka´s strongest suit. Let´s say you are not satisfied with this position as white, due to this. Then the question is, where the do you want to play something else? Would you be ready to discard 1.Nf3 because black may play into a position that Rybka will not win 100% of the time? My opinion is that unless you have devine abilities, you have to, before a Rybka-Zappa match, be more than happy about the opening of game 10 arising in one of the games. We are only humans, and chess does not give you options to just find and pick one even more promising line whenever you feel like it.
Parent - By Bouddha (****) Date 2007-09-29 21:09 Edited 2007-09-29 21:54
I do 100% agree.

I would really like to to see Rybka improve its king attack for version 3.

It would have won last game if it would have been clearly improved in this sector.
I also already showed on this forum by posting c11 and c14 games were Rybka was playing to passive and not dealling as good with king attack than other engines like e.g. Deep Fritz 10 or Naum 2.2

So good news, there is room for improvement !  ;-)

best regards


1st example

[Event "QX6700, Blitz:5'"]
[Site "Privé"]
[Date "2007.09.29"]
[Round "14"]
[White "Zappa Mexico X64"]
[Black "Rybka 2.3.2a mp"]
[Result "1-0"]
[ECO "C11"]
[WhiteElo "3000"]
[BlackElo "3000"]
[Annotator "0.61;0.00"]
[PlyCount "151"]
[TimeControl "300"]

{Intel(R) Core(TM)2 Quad CPU           @ 2.66GHz 2667 MHz  W=12.6 ply; 2'622kN/
s; 4 TBAs; RybkaII.ctg  B=13.8 ply; 447kN/s; Super.ctg} 1. e4 {B/0 0} e6 {B/0 0
} 2. d4 {B/0 0} d5 {B/0 0} 3. Nc3 {B/0 0} Nf6 {B/0 0} 4. e5 {B/0 0} Nfd7 {B/0 0
} 5. f4 {B/0 0} c5 {B/0 0} 6. Nf3 {B/0 0} Nc6 {B/0 0} 7. Be3 {B/0 0} cxd4 {
B/0 0} 8. Nxd4 {B/0 0} Bc5 {B/0 0} 9. Qd2 {B/0 0} O-O {B/0 0} 10. O-O-O {B/0 0}
a6 {B/0 0} 11. Qf2 {B/0 0} Bxd4 {B/0 0} 12. Bxd4 {B/0 0} b5 {B/0 0} 13. Be3 {
B/0 0} b4 {B/0 0} 14. Na4 {Blancs dernier coup de la bibliothèque B/0 0} a5 {
B/0 0} 15. Bb5 {0.61/13 11} Qc7 {B/0 0} 16. Nc5 {0.49/13 16} a4 {
Noirs dernier coup de la bibliothèque B/0 0} 17. Kb1 {0.38/13 21} a3 $2 {
(Ta5) 0.00/15 7} (17... b3) (17... Ra5) 18. b3 {0.48/13 5} Rb8 {0.00/16 10} 19.
Bd3 {0.55/14 5} Re8 {(d4) 0.00/15 5} 20. Rhe1 {(Cxd7) 0.52/13 9} Ra8 {
0.12/14 13} 21. Nxd7 {0.52/13 6} Qxd7 {0.19/15 3} 22. Qe2 {(Fc5) 0.49/13 4} Qc7
{0.00/15 10} 23. h4 {(g4) 0.48/13 7} Rd8 {(Cb8) 0.01/14 7} 24. Qh5 {
(Ff2) 0.48/12 3} g6 {(h6) -0.11/16 23} 25. Qe2 {0.55/14 7} d4 {(Ce7) 0.00/15 8}
26. Bd2 {0.50/14 5} Bb7 {(Ce7) 0.00/15 9} 27. h5 {0.74/14 8} Ne7 {0.15/15 6}
28. hxg6 {(Dg4) 1.02/14 7} fxg6 {0.18/15 8} 29. Qg4 {(Th1) 0.98/14 7} Bd5 {
(Cf5) 0.29/14 17} 30. Bxb4 {0.83/13 6} Rdb8 {0.33/14 2} 31. Bd2 {0.81/13 8} Qc6
{0.36/13 5} 32. Re2 {(Tg1) 0.77/12 4} Rb7 {(Ta7) 0.43/13 4} 33. Qh3 {0.76/12 6}
Rf8 {(Cf5) 0.48/12 4} 34. Rh1 {0.95/12 6} Rf7 {0.45/13 4} 35. Bc1 {
(Tc1) 0.87/12 4} Ra7 {(Dc5) 0.54/14 5} 36. Rh2 {1.08/13 4} Qc5 {(Rf8) 0.59/13 4
} 37. g4 {1.09/13 9} Bc6 {(Dc3) 0.65/13 11} 38. Ref2 {1.21/12 3} Bd7 {
(Fd5) 0.65/13 1} 39. Qh6 {(c4) 1.44/11 3} Qb6 {(Ta8) 0.65/12 3} 40. Rd2 {
(c4) 1.45/12 7} Qc5 {(Fc6) 0.83/12 5} 41. Rdf2 {(Dh4) 1.35/12 6} Qb6 {
(Fc8) 0.00/20 3} 42. Rd2 {(Fxa3) 1.61/13 7} Qc5 {(Fc6) 0.00/19 1} 43. Rh3 {
(Fxa3) 1.39/12 4} Bc8 {(Fc6) 0.91/11 2} 44. Rdh2 {(Dg5) 1.50/12 3} Rg7 {
1.13/13 4} 45. Qg5 {(Dh4) 1.54/12 7} Qd5 {(Tb7) 1.62/14 7} 46. Bc4 {
(Fxa3) 1.80/12 3} Qd7 {(Dc5) 1.79/15 5} 47. Qh6 {(Td3) 1.89/12 3} Qc6 {
(De8) 1.55/12 2} 48. Rd3 {1.82/11 2} Qc5 {(Db6) 1.71/13 7} 49. c3 {1.78/11 3}
Nd5 {1.76/12 1} 50. cxd4 {(Fxd5) 1.83/12 3} Qf8 {1.65/12 1} 51. Rf3 {
(Fxd5) 2.05/12 3} Nb6 {1.93/13 2} 52. Bd3 {2.05/11 0} Ba6 {(Fb7) 2.10/13 4} 53.
Bc2 {(Fxa6) 2.29/12 2} Bb7 {(Da8) 2.34/13 4} 54. Rf1 {(Tff2) 2.36/12 3} Nd5 {
2.79/13 3} 55. f5 {2.34/12 2} Nc3+ {2.88/13 3} 56. Ka1 {2.34/11 0} Be4 {
2.81/14 2} 57. Rhf2 {(Dh3) 2.67/11 3} exf5 {3.17/12 1} 58. gxf5 {3.27/12 2}
Rgf7 {3.32/13 3} 59. Qh4 {(e6) 3.02/12 6} gxf5 {3.96/12 2} 60. e6 {3.48/12 2}
Rf6 {4.33/13 1} 61. Qg3+ {3.48/11 0} Qg7 {4.21/13 0} 62. Qxc3 {3.48/10 0} Rc7 {
4.41/12 0} 63. e7 {3.84/11 2} Qxe7 {4.37/13 2} 64. Bxa3 {3.84/10 0} Rxc3 {
(Dd8) 5.39/13 7} 65. Bxe7 {5.64/11 3} Re6 {(Tfc6) 5.43/12 2} 66. Bxe4 {
6.38/12 1} Rxe4 {5.55/12 3} 67. Bf6 {(Fc5) 6.38/11 0} Kf7 {5.49/9 1} 68. Rxf5 {
6.72/11 1} Rc2 {5.64/8 1} 69. Re5 {6.72/10 0} Rxe5 {5.82/10 0} 70. Bxe5+ {
6.72/10 0} Kg6 {(Re6) 5.96/11 1} 71. a4 {(Rb1) 8.17/12 1} h5 {6.06/10 1} 72. a5
{8.17/11 0} Rc8 {(Rg5) 6.41/12 1} 73. a6 {(Rb2) 8.34/9 0} Ra8 {6.58/13 0} 74.
Rf6+ {9.90/13 1} Kg5 {6.74/13 1} 75. b4 {(Rb2) 9.90/13 0} h4 {7.00/10 1} 76. b5
{(Rb2) 9.90/12 0} 1-0

2nd games :

[Event "QX6700, Blitz:5'"]
[Site "Privé"]
[Date "2007.09.29"]
[Round "6"]
[White "Zappa Mexico X64"]
[Black "Rybka 2.3.2a mp"]
[Result "1-0"]
[ECO "C11"]
[WhiteElo "3000"]
[BlackElo "3000"]
[Annotator "0.57;0.11"]
[PlyCount "95"]
[TimeControl "300"]

{Intel(R) Core(TM)2 Quad CPU           @ 2.66GHz 2667 MHz  W=15.2 ply; 2'676kN/
s; RybkaII.ctg  B=11.6 ply; 445kN/s; Super.ctg} 1. e4 {B/0 0} e6 {B/0 0} 2. d4
{B/0 0} d5 {B/0 0} 3. Nc3 {B/0 0} Nf6 {B/0 0} 4. e5 {B/0 0} Nfd7 {B/0 0} 5. f4
{B/0 0} c5 {B/0 0} 6. Nf3 {B/0 0} Nc6 {B/0 0} 7. Be3 {B/0 0} cxd4 {B/0 0} 8.
Nxd4 {B/0 0} Bc5 {B/0 0} 9. Qd2 {B/0 0} O-O {B/0 0} 10. O-O-O {B/0 0} a6 {B/0 0
} 11. Qf2 {B/0 0} Bxd4 {B/0 0} 12. Bxd4 {B/0 0} b5 {B/0 0} 13. Be3 {B/0 0} b4 {
B/0 0} 14. Na4 {Blancs dernier coup de la bibliothèque B/0 0} a5 {B/0 0} 15.
Nc5 {0.57/12 6} Qc7 {(a4) 0.11/15 11} 16. Bb5 {(Rb1) 0.48/12 9} a4 {
Noirs dernier coup de la bibliothèque B/0 0} 17. g4 {0.48/12 12} Ra5 {0.00/14 6
} 18. Nxd7 {(Fd3) 0.31/13 7} Qxd7 {-0.10/11 0} 19. Qf1 {(Fd3) 0.09/14 10} Qc7 {
-0.31/15 11} 20. Kb1 {0.13/13 6} Bb7 {-0.31/14 3} 21. Bd3 {(Fc5) -0.02/13 7} d4
{(Tc8) -0.45/14 17} 22. Qh3 {(Fd2) 0.09/13 4} g6 {-0.58/15 3} 23. Bd2 {
-0.13/14 12} a3 $6 {(Tc5) -0.54/14 2} (23... b3) (23... Rc5) 24. b3 {0.11/13 5}
Rc8 {(Ce7) -0.33/14 27} 25. Rhe1 {(Dh6) 0.13/12 4} Qe7 {(Fa6) -0.41/13 2} 26.
Qg3 {0.27/12 3} Qc5 {-0.46/14 3} 27. h4 {(Df2) 0.28/13 4} Na7 {(Ce7) 0.00/14 12
} 28. h5 {(f5) 0.44/14 4} Nb5 {0.16/14 5} 29. Rc1 {(hxg6) 0.59/14 6} Nc3+ {
0.27/13 13} 30. Ka1 {0.59/13 0} Nd5 {(Fa6) 0.34/13 30} 31. hxg6 {1.07/13 6}
fxg6 {0.38/14 2} 32. f5 {0.98/13 4} Ne3 {0.33/13 2} 33. Qh4 {1.24/13 9} Qc7 {
2.22/14 27} 34. fxg6 {1.28/13 7} Rxe5 {(hxg6) 1.61/12 0} 35. Rf1 {
(gxh7+) 4.29/12 10} Nxf1 {(h5) 4.58/14 21} 36. Rxf1 {4.31/11 0} h5 {
(Te2) 7.70/13 16} 37. Rf7 {6.58/11 0} Qd8 {6.13/12 5} 38. Qf2 {6.58/10 0} Rc7 {
(Te3) 5.62/10 3} 39. Bh6 {(Fxb4) #10/13 3} Re1+ {(Ff3) 12.72/9 2} 40. Qxe1 {
#9/17 4} Rxf7 {#14/10 1} 41. Qxe6 {#8/16 0} Qe7 {#11/7 0} 42. Bc4 {
(gxf7+) #7/14 0} Bd5 {#12/6 0} 43. Qc8+ {#6/13 0} Qf8 {#5/3 0} 44. gxf7+ {
#5/13 0} Kxf7 {#4/3 0} 45. Bxd5+ {#4/14 0} Kg6 {#3/3 0} 46. Qxf8 {#3/14 0} hxg4
{#2/3 0} 47. Qg7+ {#2/63 0} Kh5 {#1/3 0} 48. Qg5# {#1/63 0} 1-0

3rd games

[Event "QX6700, Blitz:5'"]
[Site "Privé"]
[Date "2007.09.29"]
[Round "4"]
[White "Zappa Mexico X64"]
[Black "Rybka 2.3.2a mp"]
[Result "1-0"]
[ECO "C11"]
[WhiteElo "3000"]
[BlackElo "3000"]
[Annotator "0.63;0.08"]
[PlyCount "93"]
[TimeControl "300"]

{Intel(R) Core(TM)2 Quad CPU           @ 2.66GHz 2667 MHz  W=13.8 ply; 2'701kN/
s; RybkaII.ctg  B=14.4 ply; 495kN/s; Super.ctg} 1. e4 {B/0 0} e6 {B/0 0} 2. d4
{B/0 0} d5 {B/0 0} 3. Nc3 {B/0 0} Nf6 {B/0 0} 4. e5 {B/0 0} Nfd7 {B/0 0} 5. f4
{B/0 0} c5 {B/0 0} 6. Nf3 {B/0 0} Nc6 {B/0 0} 7. Be3 {B/0 0} cxd4 {B/0 0} 8.
Nxd4 {B/0 0} Bc5 {B/0 0} 9. Qd2 {B/0 0} O-O {B/0 0} 10. O-O-O {B/0 0} a6 {B/0 0
} 11. Qf2 {B/0 0} Bxd4 {B/0 0} 12. Bxd4 {B/0 0} b5 {B/0 0} 13. Be3 {B/0 0} b4 {
B/0 0} 14. Na4 {B/0 0} a5 {Les 2 dernier coup de la bibliothèque B/0 0} 15. Nc5
{0.63/12 6} a4 {0.08/16 22} (15... Qc7) 16. Bd3 {(Fb5) 0.48/13 10} a3 $6 {
(Dc7) 0.00/13 2} (16... Re8) 17. b3 {0.48/14 9} Re8 {0.01/15 4} 18. Bb5 {
0.46/14 10} Qc7 {0.07/15 4} 19. Nxd7 {(Cd3) 0.44/13 6} Bxd7 {-0.07/15 4} 20.
Bd3 {0.33/13 5} Rec8 {(Fc8) -0.01/15 3} 21. Bb6 {0.30/13 6} Qb7 {0.00/16 2} 22.
Rhe1 {(Fc5) 0.15/14 8} Nd8 {(Ca7) -0.03/15 7} 23. g4 {(Rb1) 0.61/13 5} g6 {
(Cc6) 0.10/13 11} 24. f5 {0.97/14 9} exf5 {(Cc6) 0.81/15 12} 25. gxf5 {
1.05/14 8} Rc6 {0.66/15 3} 26. Bxd8 {(e6) 1.08/14 9} Rxd8 {0.69/14 2} 27. e6 {
1.01/15 11} Be8 {0.84/15 11} 28. e7 {1.09/14 8} Rdc8 {0.82/16 15} 29. Qd4 {
1.12/13 8} R6c7 {(Db6) 1.00/16 38} 30. f6 {1.45/13 5} Qc6 {1.25/15 13} 31. Rf1
{1.74/14 7} Qc3 {(Dc5) 1.35/15 3} 32. Rf4 {(Dxc3) 2.12/14 4} Qc6 {
(Rh8) 1.86/14 5} 33. Rf5 {(Dxb4) 2.68/13 4} Qe6 {2.44/14 5} 34. Rxd5 {2.68/12 0
} Rc6 {(h6) 2.65/13 2} 35. Rd8 {3.23/13 5} h6 {2.96/13 2} 36. h4 {
(Rb1) 3.91/12 4} h5 {2.99/12 2} 37. Bc4 {(Rb1) 4.29/14 9} Rxc4 {4.89/13 2} 38.
bxc4 {4.29/12 0} Kh7 {5.14/13 2} 39. c5 {5.04/12 0} Qh3 {(b3) 5.28/12 4} 40.
Qxb4 {6.50/14 6} Qe3+ {5.49/13 2} 41. Kb1 {6.50/13 0} Qf3 {5.41/12 1} 42. Qb3 {
6.50/12 0} Qg4 {5.89/11 4} 43. Rc1 {(Dxa3) 7.13/12 3} Kh6 {5.75/11 5} 44. Re1 {
(Dxa3) 7.13/12 0} Qxh4 {9.80/13 18} 45. Qe3+ {7.13/12 0} Kh7 {9.96/12 2} 46.
Rxc8 {11.05/14 3} Qb4+ {10.35/11 1} 47. Kc1 {11.05/13 0} 1-0
Parent - By oudheusa (*****) Date 2007-10-08 10:18
I agree with Jeroen.

Although we all 'feel' that Rybka is the stronger engine, she just made a couple of very bad evaluations in critical positions.

That is why she lost the match and what should be fixed.
Parent - - By George Tsavdaris (****) Date 2007-09-29 16:10

>Yes, we had very bad luck in two games, but that doesn't mean that the Zappa victory was not well-deserved!

I agree in many things you've said and i disagree also in many but that doesn't matter much, what i want to understand is about this luck you're referring.
Luck about something means and actually definitely should mean: Something random that humans can't predict its outcome AND humans can not do anything to affect its outcome.

I hope you agree on this. So what kind of bad luck do you think you had?

Shortly: I don't believe in luck in Chess!
If you write a Chess engine that would have avoid some kind of mistakes, then this engine will avoid that kind of mistakes.
If you write a Chess engine that doesn't know to avoid some kind of mistakes, then this engine will not avoid that kind of mistakes.
Same with opening books.
Same with everything.  You can control everything!
There is no luck inside....
Parent - - By Uri Blass (*****) Date 2007-09-29 16:17
Engines that use many processors are not deterministic so there may be luck about the choice of the move.

Even without that luck there may be luck in chess.

Imagine that you found a trick that the opponent did not see and win a pawn.
After winning the pawn you suddenly find that you fell into a trap and cannot prevent checkmate.

You see it earlier than the opponent who think for more time that you have the better position but it does not help you and you lose the game.

I think that in that case the opponent was lucky.

Parent - - By George Tsavdaris (****) Date 2007-09-29 16:34

>Engines that use many processors are not deterministic so there may be luck about the choice of the move.

I guess there is a programming method to make the multi-thread programs deterministic, so we can blame only the programmers that don't find this way and still use non-deterministic programs and leave some things in luck....

>Imagine that you found a trick that the opponent did not see and win a pawn.
>After winning the pawn you suddenly find that you fell into a trap and cannot prevent checkmate.
>You see it earlier than the opponent who think for more time that you have the better position but it does not help you and you lose the game.
>I think that in that case the opponent was lucky.

Well you played a blunder(the Pawn capture) and your opponent took advantage of it. You saw it earlier from your opponent but this is irrelevant since you played the blunder! You were not clever enough to avoid the blunder.
The opponent simply didn't do any mistake and although he didn't see as quickly as you saw it, he didn't made any mistake and took advantage from your mistakes. This is no luck as i see it.....
Parent - - By Vempele (Silver) Date 2007-09-29 17:05

>  I guess there is a programming method to make the multi-thread programs deterministic, so we can blame only the programmers that don't find this way and still use non-deterministic programs and leave some things in luck....

It's generally agreed that trying to write a deterministic parallel search would kill the scaling (think of it: no shared hash, history tables or killers, weird restrictions on splitpoints etc). You can't blame the programmers for refusing to settle for _deterministically_ bad pseudo-luck. :)
Parent - By George Tsavdaris (****) Date 2007-09-29 17:32

>It's generally agreed that trying to write a deterministic parallel search would kill the scaling.......

There is sure a way for writing a deterministic parallel search without killing the scaling. We humans, are just incapable of finding it. No luck involved.... :-)
Parent - By Uri Blass (*****) Date 2007-09-30 19:45
If I play a blunder that my opponent did not understand it was a blunder then I consider it as luck.

The same can happen with programs.

It is possible that program A simply outsearchs program B to get a position when both programs agree it is +1 pawn for program A but the position is winning for program B.

B can find the win but A could also find the win in case of getting the same position so the fact that B wins the game is clearly because of luck(assuming that it was not a book preperation).

Parent - - By Alkelele (****) Date 2007-09-29 18:50
I generally agree with you about luck, that it often does not make much sense to discuss such a thing. If Rybka misses Qe1 or Zappa misses hxg7, it´s better just to consider this a consequence of the strength of the engines. But that´s why I was talking about freakishly bad luck :-)

In game 4, Rybka lost a half point due to pushing further once the 50 move rule kicked. In game 9, she failed to win a totally won position with an eval of more than +4. Let´s say, each of these things may occur in something like 1 out of 300 games. Then let´s also say, that somehow this is partly a consequence of a somewhat systematical endgame weakness in Rybka. So we may say, such freakish incidents may happen in 10 out of 300 Rybka games. Then, the bad luck here is having 2 of such incidents happen in a 10-game match.

Anyway, this is bound to happen. It´s like a soccer game where one team had 5 shots on the post and lost. This surely doesn´t mean that the winning team didn´t deserve to win! They performed as well as the circumstances dictated them to do in order to win. The only reason I brought this up was that somehow the match result has been used as evidence that we were overconfident, didn´t prepare properly etc. etc. My point is that there other more straightforward and truthful explanations for the match loss, like, for example, anything can and will happen in a 10-game match against a strong opponent.
Parent - By Banned for Life (Gold) Date 2007-09-29 19:21
Game 4 probably doesn't belong in the freakishly bad luck category. I would categorize it more as a slow motion train wreck. As I recall, many moves before the blunder, Erdo turned down a draw offer. One must assume that he did this because he felt there was the possibility that Rybka would blunder and lose. She did, and this isn't really a rarity in engine chess games and I recall Team Rajlich actually making use of this behavior in a freestyle game against Bjorn's unassisted Rybka (in that case the 50-move rule didn't cause a loss though). Some engines handle these "leave well enough alone" situations better than others.

So maybe you could put the transition from win to draw in the bad luck category, but the transition from draw to loss would have to go in the engine weakness category.

This was probably a great opening for Team Rybka, but its also possible that the expected outcome from the +4 score might end up being unexpectedly low if Rybka always ends up making the same mistakes.

Parent - By JohnL (***) Date 2007-10-01 21:03
I would definitely say there is luck in chess. The reason is that if you want to make a useful model of chess performance, you have to involve some "luck" or "randomness".

You can always say afterwards that there was no luck, you lost only because that and that factor. The problem is that this is ad hoc reasoning and you will introduce a lot of special (non-generalizable) factors,
which are not useful nor effective for prediction. And neither useful for what you should spend your finite chess practice time on, trying to improve yourself :-)

This can be seen as "only" a practical view, but I would say that it is also purely theoretically motivated...

Also look at the ELO system which is based on luck, the stronger player only has a given chance to win. It is meaningless to search for a 100% predictive model by analyzing why games ended with certain results.
Parent - - By Venator (Silver) Date 2007-09-29 07:43
Hi Dagh,

I can add some comments to your report:

1) In the 5 white games the Zappa team had 1 book draw (which was a very fine one IMO, so I must congratulate Erdo on finding it!) and was worse to lost in the remaining 4 games. In game 2 white was a sound pawn up three moves after Zappa left book. In game 4 the position was already won for white after the opening moves were over. In game 8 black misplayed the opening (Rae8, Rb8) after which white was better. In game 10 amazingly black chose a hedgehog with a very passive position and I consider the black position already critical before move 20 and lost after b5-b4?

2) The Zaitsev was very deeply prepared by me especially for this match. IMO it is a very suitable opening for Rybka, which was comfirmed all the way in my analysis: complicated position, imbalanced structure, potential pawn mass for black on the Q-side. In many lines I found out Zappa (and other progs) misplay it and also do not evaluate it correctly. I spent more than 100 hours on this line (using the Octa machines from Lukas, thanks Lukas!), finding many new beautiful idea's in the process. I firmly believe in this line for black and I still do.

3) Playchess statistics are nice, but mainly based on blitz games with specially prepared books. So the scores of certain lines (f.e. the Zaitsev) do not say much about the position itself. Only by digging into lines deeply, using strong hardware and many different programs, can tell you the truth about these lines. And how Rybka reacts on it. I am often amazed how a line is discarded just by looking into a database, seeing that is scores badly and 'so the line is bad'.

4) With black we obviously had a few problems, especially in game 3 (I regard the line in game 1 as not dangerous for black and so it was no surprise Erdo avoided it in game 3). Somehow Rybka misplayed it, by moving Na5-c6-a5-b7. I cannot recall my analysis of this line, but I stopped at a point where I was sure Rybka found the good moves herself. This game proved that this was probably 1 or 2 moves too early. In game 5 white played an amazingly timid line. We knew exactly that our opponent would go for this 3.Bb5+ line, as Rybka has some problems against Zappa in the lines where black allows c2-c4. When you look at Playchess you can see that the huge majority of the games went 5... Nc6, while IMO this line is absolutely nothing for white after the simple 5... Nf6. This is exactly the reason no top GM tries this anymore. So basically we were already even after the timid moves Re1, Rxe5 and Re1. Game 7 saw black easily equalise after Zappa's timid moves a3 and Qd3. In game 9 white played the non critical Bc4 line against the Karpov Caro-Kann, again not achieving much from the opening, in which Rybka simply outplayed Zappa.

5) I don't agree with the critics that say that Rybka plays the Ruy Lopez below par. My analysis shows otherwise and two games are of course far too little an amount to base an objective conclusion upon.

6) All in all I think that except for game 3 the lines we played were fine. But as the result was 5,5-4,5 in Zap's favour, of course it is logical people start to look for reasons for the loss and to blame the bookexpert is something I already witnessed in the past several times. I have no problems with that and learned to live with it :-). When you win you are the hero, when you lose you did a lousy job. All I can say is that we did our damned best, Dagh and I spent a huge amount of time analysing our lines and looking for Zappa weaknesses. And it was still not enough. But chess is more than a bunch of opening moves and Playchess statistics and even a winning position out of book can be lost. I made a joke about it: 'perhaps we should book until it is mate!?'.

In any case it was a close match which could have gone either way. I want to congratulate Erdo and Anthony for this win, in the end Zappa made fewer mistakes and I think that was the main reason Zappa won the match.

Best regards, Jeroen
Parent - - By Harvey Williamson (*****) Date 2007-09-29 09:20
Hi Jeroen,

I made a joke about it: 'perhaps we should book until it is mate!?'.


This is clearly wrong. You only need to book until your 6 man tbs take over. ;-)

See you in Leiden,

Parent - By Venator (Silver) Date 2007-09-29 11:58
Hi Harvey,

Not really, because mate can be delivered long before the endgame :-)

See you in Leiden!

Parent - - By Venator (Silver) Date 2007-09-29 12:25
Hi Alan,

You are wrong here. Neither were we overconfident, nor did we neglect the strength of Zappa. Dagh and I put a lot of hard work in preparing for this match, using Zappa in the process. If you think the Zappa preparation was better, then I think you didn't look at the positions after the opening objectively. But as I already mentioned: if you win you are the hero, if you lose you are the sucker. If Rybka had managed to pull away the last two games winning 5,5-4,5 we'd read other comments. But apparently this is a world of statistics and we have to live with that: opening great but you lost, then you did a bad job.

Personally I would feel very unhappy having two almost book losses in game 4 and game 10, and not gaining anything in the opening in white games 5, 7 and 9. And that in only 10 games. If that would happen to me, you *surely* have a reason to put the blame on me.

Kind regards, Jeroen
Parent - - By Banned for Life (Gold) Date 2007-09-29 14:54
Actually, I don't remember ever blaming either you or Dagh for the outcome of the match. In a long sequence of games, even with an engine with an ~75 Elo advantage, there will be no problem finding sequences of 10 games where the underdog had a better score.  Team Rybka's book play was lacking only based on expectations. The question that many people had going into the match was whether Team Rybka would win on book alone, or whether the engine would need to play a part too.  So to some degree, you were a victim of very high expectations.

As far as the out of book results go, the appropriate methodology would be to run each position a number of times with Rybka and Zappa playing their lines with the match hardware and time controls, and collect the results. It is very possible that if this experiment were done, the resulting statistics would contradict the otherwise reasonable assessment of the positions that you and Dagh have put forward. On the one hand, if you compared the match results with the statistics from these tests, and you got fewer points than the test statistics would predict, you could say you were unlucky. On the other hand, its possible that the statistics would show that some of the lines that look equal or better actually end up with a poor percentage when played by Rybka vs Zappa. In my opinion, these results would not go into the unlucky column, but would actually be bad lines for Rybka to be playing against Zappa, despite any apparent advantage at the end of the line. If your book was full of these type of lines, I would say you were badly prepared. Furthermore, I would content that there is no way you could determine whether this was the case or not without actually testing against Zappa.

Anyway, I'm glad to hear that you actually did test against Zappa and I suspect that if each game was replayed a bunch of times with the same engines on the same hardware and from the same openings, Team Rybka would probably win the match more often than not.

Parent - By Venator (Silver) Date 2007-09-29 15:24
Hi Alan,

Basically you want to test your book the way you describe it. Try as many lines as possible against the opponent you are playing against. That would be the ideal situation. However, you will never be able to do so, because of various reasons:

1. Limited time.
2. No insight in the lines the opponent is going to play.
3. No hardware simulation possible.
4. No access to both program versions that actually play.
5. The test will take a huge amount of time, as the opponent might play virtually anything.

So all you do is trust your judgment and trust what you know about Rybka and Zappa and patch the book the best you can, in the limited time available. Try the lines, try to figure out if they work well or not. Use both Rybka and Zappa. Analyse Erdo's lines. Look for mistakes in them, for weak lines and book them. That is exactly the way Dagh and I proceeded. Well, it didn't work. It is simply the way it is. All of team Rybka did their best to win the match, but it didn't happen. Once there would be an event in which Rybka wouldn't win. We all knew that. We also knew this was going to be a tough match, as various other internet matches between Zappa and Rybka already indicated a very close business. And it turned out that we were at the wrong side of the score. OK, that is all about chess, all about computer chess. Sometimes you win, sometimes you lose.
Parent - - By Alkelele (****) Date 2007-09-29 18:59

> Actually, I don't remember ever blaming either you or Dagh for the outcome of the match.

You posed questions about our method of preparation. I readily answered, and then you continued with baseless assertations about our preparation methods, quality, overconfidence etc. When I countered this with factual arguments, you finished off the conversation with an insult essentially saying that as long I would be on board, I would handicap Rybka with 75 Elo. You can call it blaming or not, either way, I find this kind of unprovoked attack worthy of an apology (and that, even if you were right).
Parent - - By Banned for Life (Gold) Date 2007-09-29 20:39

That's not quite right. From my perspective, your first response was already pretty high on the rudeness meter (and off the scale for someone from your part of the world). I will apologize for my third comment though.

As far as preparation goes, I continue to believe that Erdo made a very conscientious decision to concentrate as much on the weaknesses of Rybka as on the strengths of Zappa when putting his book together. This is not surprising behavior for a team that goes in as a significant underdog in both the book and engine categories. I never claimed that he was playing his tournament book in a public forum, although I suspect that he used this as a starting point in deciding which lines to pursue. I got to play a number of games against Erdo's account running Zappa and we generally came out of book with both engines thinking they were better off and Zappa's assessment being much more accurate based on the continuation. The fact that this preparation wasn't enough to put the Zappa book on the same plane as the Rybka book doesn't indicate that Erdo's effort wasn't a worthwhile endeavor. Without it, the match might have been over before it began.

Of course when you are the front runner you have more options. You can concentrate on your opponents, or decide to ignore them. In my opinion, Vas is correct in ignoring other engines and just trying to make Rybka the best that it can be, but I don't believe the Rybka bookmakers should ignore how the other engine plays an opening in a match situation. My question was intended only to ascertain whether you guys were preparing for Zappa or just trying to be the best you could be. Your response was a) Erdo didn't do this, b) its impossible to do, and c) maybe we did a little bit of it.

My Aristotelian conclusions, based on all the discussions are as follows:

1) Zappa focused more on Rybka than vice versa. This isn't really surprising given that Rybka was a heavy favorite going in. The natural result was that Team Zappa was extremely focussed on finding any Rybka weaknesses, while the Rybka team was partially focused on Zappa and partially focused on being the best it could be. This is not by any means an indictment of the Rybka Team or approach.
2) Team Rybka had an advantage coming out of book, but less of an advantage than most expected. Some people castigated you guys for this but I don't think I was one of them. This is the price you pay for being the best.
3) In only 10 games, Rybka ended up in quite a few positions that it had a tendency to play poorly. This might have been just bad luck or maybe it had something to do with Team Zappa's opening preparation. The bad luck argument is easier to defend, but I suspect that if the match were much longer, the score wouldn't reflect the difference in Elo between the two engines.

Parent - - By Alkelele (****) Date 2007-10-01 12:27
Dear Alan,

Thanks. You write:

> My question was intended only to ascertain whether you guys were preparing for Zappa or just trying to be the best you could be. Your response was a) Erdo didn't do this, b) its impossible to do, and c) maybe we did a little bit of it.

I suspect we misunderstood each other here. I never meant to say something like a. Indeed, what you earlier wrote was:

> I suspect that Team Zappa has a much better appreciation of the types of positions that Zappa plays well against Rybka than vice versa.

THIS is what I replied to, saying that I firmly believe this to be wrong. Both Jeroen and I have said repeatedly now that we spent a significant amount of time analysing Rybka-Zappa issues, and that we took this analysis into account during all our bookwork. I have also given you several concrete examples from the opening choices in the games now. What more could we do to convince you that your Aristotelian conclusion that "Zappa focused more on Rybka than vice versa" is wrong? Indeed, looking back, I can safely say that the one single main theme of our preparation was Rybka-Zappa issues.

Btw, this leads to a catch 22 situation. We can either be criticised for not focusing on Zappa, or when we inform that we indeed did do so, we are suckers for apparently not doing it in the right way. That's ok, criticism is fine, especially when we lost. On the other hand, it would be nice if it is not assumed that we're lying through our teeth when we engage in post-mortem discussions here in the forum.

Kind regards,
Parent - - By Banned for Life (Gold) Date 2007-10-01 15:32

My intent wasn't to be critical, only to ascertain your method of preparation. As far as the match is concerned, I think Anthony got it right when he said that Zappa won because the match format allowed Zappa to be in a time-processing speed region that it favors, hard work by the opening group to keep you and Jeroen from blowing them out of the water before the engines engaged, and luck.

I don't really see the Catch 22 situation here. When the favorite loses in any endeavor, there is bound to be criticism, some well intentioned and well grounded, some malicious and far fetched. Being overly defensive and attacking your critics is a probably a poor response to any of these permutations. Even if you are sure the result was really 100% due to bad luck, it would be better all around to talk about lessons learned and things that will be improved in the future, rather than trying to show that theoretically Team Rybka should have won the match. Vas is a master at this by way. His summation of the match includes congratulations for the opponents, commendations for his own team, a promise to return with a better product, and no excuses.

Parent - - By Alkelele (****) Date 2007-10-01 15:59
Thanks, you are right. Indeed, after a little while, it's obvious to me that the best way to respond to any criticism is to try and make sure to win convincingly the next time around. Anyway, take it easy on me, I'm still just learning how to handle a loss ;-) 
Parent - By Nelson Hernandez (Gold) Date 2007-10-01 17:14
You are right, losing is a serious omission in your psychological repertoire.  We must try to remedy that with lots of practice.  :)
Parent - By Linus (***) Date 2007-09-29 07:38
As has been pointed out here before, 10 games is not enough to give us a statistically sound information about which engine is better. The rating lists do that for us anyway.

I would compare events like Rybka - Zappa with Formula 1. A public test lab with audience. New things are developed, improved and tested under they eyes of a large audience (ok, maybe not *that* large when it comes to chess). The opening books are edited every day, new engine versions were tried (at least by the Rybka team). That is, what makes these battles interesting.

And I believe we saw some incredible computer chess in this event. I am looking forward to a rematch.
Parent - By turbojuice1122 (Gold) Date 2007-09-29 01:55
Actually, I saw a number of Ruy Lopez games played by engines running Zappa Mexico on Playchess, and Rybka tended to have a bit of success when playing the Breyer Variation, if I remember correctly; perhaps this is something to look at.
Parent - - By revengeska (**) Date 2007-09-29 02:40
I love how in every video you were either ready to fall asleep or actually asleep:P
Parent - - By Alkelele (****) Date 2007-09-29 07:07
I fell asleep during Rybka-GM Robert Fontaine after 36 hours of no sleep. I probably stayed awake for so long as a consequence of my alleged overconfidence, thinking that Rybka would easily beat Zappa in the next couple of games... ;-)
Parent - By revengeska (**) Date 2007-09-29 07:14
Yeah, I guess so.  To keep playing the spanish against Zappa had to be overconfidence:)
Parent - By Lukas Cimiotti (Bronze) Date 2007-09-28 06:34
Zappa ran on Harvey Williamson's computer - an Intel Xeon 2x X5355 - 8 X 2.66 GHz.
For fairness and as equal hardware was announced before the match,we didn't use my best computer, an Intel Xeon 2x X5365 - 8 X 3 GHz, but my 2x X5355. There were no technical problems, except 2 disconnects in the last game.

Parent - By Father (**) Date 2007-09-28 03:18
Congratullations Vasik.
Rybka is great.
Parent - - By VP (*) Date 2007-09-28 05:46
Congratulation Vas and Rybka team for the wonderful tournament. The expectations from Rybka were sky high, and that is why the result hurts.
It does not matter how low you fall- what matters is how high you bounce!
I am sure Vas will bounce back and touch new heights!
Parent - - By M ANSARI (*****) Date 2007-09-28 08:45
Yes congrats to Anthony and Erdo for doing a really great job in maximizing their effort.  This could have easily gone either way and for a while it looked like Rybka would come back as it always does .... unfortunately it ran out of games.  Personally I think that the book used by Rybka was more than adequate ... maybe Rybka's first loss was not the best book line since Rybka did not handle it properly ... but the others were fine.  Rybka missed Qe1 in game 4 which would have won on the spot and played the  f4? allowing a draw in game 9 ... those moves cannot be blamed on books but on Rybka the engine and both moves could have changed the final outcome.  I think the Rybka team will learn a lot from this match since some glaring weaknesses have been discovered that most surely will be worked on now. Rybka for so long has somehow managed to put these defects under the carpet because other engines were on a level much lower and thus somehow these faults were not exposed.  Zappa on 8 core has shown that it can punish Rybka for these defects and it might just be the impetus that is needed to get Rybka stronger.  What I have learned the most is that no matter how impressed we are at the strength of engines with big hardware they can still play amazingly poorly in some positions ... especially in some endgames.  Game 4 was just ugly and should probably be rated as XXX pornography with an adult rating ... I hope we never see such a game again from any of the top engines.
Parent - By Bouddha (****) Date 2007-09-29 07:07
I do 100% agree.

I would personally prefered to see other openings than Spanish, but we can not blame the opening book.
Was Rybka that didn't play well, and it is the engine that must be improved here and maybe not the book changed.

Parent - - By Laszlo (***) Date 2007-09-28 09:02
Bravo Vas, and bon courage!
Rybka created two "immortel" games: n.2 and n.8. (Zappa did not created such games.) Pity for the bishop disasters.
Parent - - By JhorAVi (***) Date 2007-09-28 09:57
We Rybka fans are so lucky that it wasn't a world title match :)   Vas still have unlimited opportunities to prove that Rybka is still and always the best! :)
Parent - By Uly (Gold) Date 2007-09-28 16:50
We Rybka fans are so lucky that it didn't happen against Junior
Parent - By saxon (**) Date 2007-09-28 11:12
Vasik,I'm so sorry becouse of the "f4 blunder".
I just couldn't beleive it when I saw it...
Endgames,endgames.. Work more on  engames.Not on openings.
And Rybka will be untouchable again.

Best wishes.
Parent - - By PAKman (**) Date 2007-09-28 13:27
Vas your are a class act.
Keep up the good work.

Larry thanks for keeping up posted.
Dagh thanks for being there on short notice.
Jeroen hope you are better.
Lukas the hardware did good looking in from North Carolina :)

Norm Pruitt
Parent - - By onursurme (***) Date 2007-09-28 15:22
I wrote an idea before, but no one replied (confirmatory or opposite).
I will keep repeating till I get an answer from the officials of Rybka.

if the pieces on board are less than a number, use a seperate program modul which is designed for endgames.
teach endgames to Rybka like this :
"if the pieces on the board are 2 kings plus
   3 pawns and 2 opposite color bishops
         evaluate the position with these criteria : ...
   1 pawn
         evaluate the position with these criteria : ...
   2 pawns and a rook
and so on."

impossible ? doing this also may help you to discover new evaluation criteria for the main program modul.
Parent - - By George Tsavdaris (****) Date 2007-09-28 23:24
And how do you know that Rybka does not do this or something similar....?
Parent - By onursurme (***) Date 2007-09-29 13:33
I didn't say "I know". this is just an idea.
it is up to the Rybka team to say if Rybka allready does this or not, and why not.
Parent - By turbojuice1122 (Gold) Date 2007-09-29 01:58
I believe that some engines do this; Rybka might do this, I don't know (though there are obviously a couple of people who post regularly on this forum who know this very well :-) )--though I'm not sure how successful it is.
Parent - - By Hetman (*****) Date 2007-09-29 05:01 Edited 2007-09-29 05:07
Congratulations for the Rybka team for the fight to the end. Congrats to Zappa Team.

Rybka is still Tournament WCC. I wonder... Playing the match it is the other matter, as in the human chess. Here the special preparation is more important then the  universal one and the theoreticly weaker opponent could win due to the better preparation. It reminds Botwinnik revange matches vs Smyslov, Tal.
I do not mean that Zappa has been weaker in the match. I think that every win and every loss is deserved. Sometimes the loss is better step to improve then win. The level of the games was high. That is most important - waiting for 3.0. :-)

Parent - - By Ankan (*) Date 2007-09-30 13:35
Congratulations to Zappa Team for the impressive win!! ... congrats to Rybka team also for their efforts. I do agree that 10 games are not sufficient for deciding which engine is the best when you win with the narrowest possible margin, but winner is winner and that is what counts.
I am a bit disappointed by the result but I still feel Rybka is a better engine - it lost only due to some very bad luck. I feel this is the best outcome of this match even though I am a Rybka fan as this will raise the level of competition and Vas will hopefully work harder on Rybka's weak points ;-)
Parent - By mariaclara (**) Date 2007-10-08 09:49
when will the rematch be?
...excitedly waiting.......
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka vs Zappa in Mexico City
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill