Examples of much weaker programs beating stronger ones?
raylopez99
search engines at the brute force level, and the only intelligence is
the evaluation function. So unless a stronger computer walks into a
mating net that lies beyond it's search ply, and those must be rare (I
would imagine), I don't see how a much weaker chess program (say 300
Elo points weaker) can beat a stronger one.
Any examples?
RL
Here's what I found from the net...I'd like to see how Super Const.
beat Mephisto, probably though the three points are mostly draws.
http://ssdf.bosjo.net/rlwww101.txt
217 Mephisto Lyon 68000 12 MHz 2004 21 -21 1121
57% 1951
294 Novag Super Constellation 6502 4 MHz 1629 18 -18 1583
34% 1746
217 Mephisto Lyon 68000 12 MHz, 2004 match against Super Const.
17-3
The Master
> I don't think they exist. That's because computers use the same
> search engines at the brute force level, and the only intelligence is
> the evaluation function. So unless a stronger computer walks into a
> mating net that lies beyond it's search ply, and those must be rare (I
> would imagine), I don't see how a much weaker chess program (say 300
> Elo points weaker) can beat a stronger one.
You appear to have forgotten about the fact that chess programs
often 'override' their calculation routines and eval functions while
in
the opening book. So then, theoretically at least, it is possible
for
a program with a great eval function and fine calculation speed to
blindly stumble into a lost position right in the opening! Of
course,
cases of this happening in real life are likely to be few and far
between since commercial programmers may do extensive testing.
But lest we forget, the current top program once saw its creator
answer a question regarding the program's endgame knowledge by
saying he thought it wasn't important for results. Now, imagine a
programmer whose attitude is that 'any decent opening book will
do.' Such a program could be subject to occasional (if infrequent)
upsets by much weaker programs before the brute-force engine
even begins to crank out torque.
Now, I expect such upsets to be quite rare since so many people
are obsessed with the chess openings and they are so very heavily
studied. But 300 points is not a huge gap. I recently was sent a
game in which a gap twice this large was surmounted between
human players for a draw, but just one simple improvement in the
endgame would have made this upset draw into an upset win!
raylopez99
> Now, I expect such upsets to be quite rare since so many people
> are obsessed with the chess openings and they are so very heavily
> studied. But 300 points is not a huge gap. I recently was sent a
> game in which a gap twice this large was surmounted between
> human players for a draw, but just one simple improvement in the
> endgame would have made this upset draw into an upset win!
We're talking computers Minor, not humans. Can you find me a game
with a 600 Elo difference that was won by the weaker machine?
RL
Sanny
> We're talking computers Minor, not humans. Can you find me a game
> with a 600 Elo difference that was won by the weaker machine?
GetClub once beat Jester 2 years back when Jester was arround 20 times
stronger than GetClub.
Today, even after 20 fold improvements GetClub finds it difficult to
beat Jester.
So Luck may favour a weak engine.
Bye
Sanny
If you have not played Chess at GetClub
You have never played any nice game in your life.
So, Play Chess at: http://www.GetClub.com/Chess.html
Tony M
<raylo...@gmail.com> wrote:
>I don't think they exist. That's because computers use the same
>search engines at the brute force level, and the only intelligence is
>the evaluation function.
I realize you're sort of joking here, but this statement is untrue.
Evaluation and search together provide "intelligence", and a program
without search (eg CPP1) is not going to be very intelligent. Plus,
searches are different in every program. Most programs use some sort
of forward pruning, so most modern programs are not pure "brute force"
programs. A stronger program may prune a good move that a weaker
program doesn't, and that could be a possible downfall for the
stronger program.
>So unless a stronger computer walks into a
>mating net that lies beyond it's search ply, and those must be rare (I
>would imagine), I don't see how a much weaker chess program (say 300
>Elo points weaker) can beat a stronger one.
>
>Any examples?
Here are three examples of my weak program beating programs 300 ELO
above it, and one example of a draw against a program over 1000 ELO
above it (a lucky draw by threefold repetition against the strong
program Thinker). The win against Freyr was on time in a drawn
position, so I'm not sure if that counts using your criteria.
[Event "GoodEnginesAgain"]
[Site "TONY-SUA7L25HR1"]
[Date "2005.06.04"]
[Round "1"]
[White "TonysChess 0.01a"]
[Black "TSCP 1.81"]
[Result "1-0"]
[ECO "A70"]
[PlyCount "118"]
[EventDate "2005.??.??"]
1. Nf3 Nf6 2. d4 e6 3. c4 c5 4. d5 exd5 5. cxd5 d6 6. Nc3 g6 7. e4 Bg4
8. Bb5+
Nbd7 9. O-O Bg7 10. Bf4 Qe7 11. h3 Bxf3 12. Qxf3 h5 13. Bg5 O-O-O 14.
Bxd7+
Rxd7 15. Rae1 Rhd8 16. e5 dxe5 17. Ne4 Rxd5 18. Nxf6 Rd3 19. Re3 Rxe3
20. fxe3
Qe6 21. Ne4 f6 22. Bxf6 Bxf6 23. Qxf6 Qxf6 24. Rxf6 Rd1+ 25. Kh2 Ra1
26. Nc3 b5
27. Rc6+ Kd8 28. Rxc5 b4 29. Nd5 Rxa2 30. Nxb4 Rxb2 31. Nc6+ Ke8 32.
Rxe5+ Kd7
33. Nxa7 Rb4 34. e4 h4 35. Re7+ Kxe7 36. Nc6+ Kd6 37. Nxb4 Ke5 38. g3
g5 39.
Nd3+ Kxe4 40. Ne1 Kf5 41. gxh4 gxh4 42. Nf3 Ke4 43. Nxh4 Kf4 44. Ng2+
Kf3 45.
Kg1 Kg3 46. h4 Kg4 47. Kh2 Kf5 48. Kh3 Ke5 49. Kg4 Kf6 50. h5 Kf7 51.
Kf5 Kg7
52. Kg5 Kh7 53. h6 Kg8 54. Kg6 Kh8 55. Nf4 Kg8 56. Ng2 Kh8 57. Nh4 Kg8
58. Nf5
Kh8 59. Nd4 Kg8 1-0
[Event "GoodEnginesAgain"]
[Site "TONY-SUA7L25HR1"]
[Date "2005.06.04"]
[Round "1"]
[White "TonysChess 0.01a"]
[Black "SmallPotato 0.6.1"]
[Result "1-0"]
[ECO "D14"]
[PlyCount "90"]
[EventDate "2005.??.??"]
1. c4 c6 2. Nc3 d5 3. cxd5 cxd5 4. d4 Nf6 5. Nf3 Nc6 6. e3 a6 7. Bd3
Bg4 8. O-O
e5 9. Be2 Bxf3 10. Bxf3 e4 11. Be2 Bd6 12. h3 O-O 13. f3 b5 14. fxe4
dxe4 15.
Qc2 Nb4 16. Qb1 Re8 17. a4 bxa4 18. Nxa4 Nbd5 19. Nc5 Bxc5 20. dxc5
Qc7 21. Qc2
a5 22. Bd2 Reb8 23. Kf2 Qe5 24. Rxa5 Rxa5 25. Bxa5 Nxe3 26. Kxe3 Rxb2
27. Qc4
Qg3+ 28. Kd4 Qxg2 29. Qa6 Rb8 30. h4 Qg3 31. Qd6 Qxd6+ 32. cxd6 Kf8
33. Bc4 Ke8
34. Re1 Kf8 35. Ra1 h6 36. Bc7 Rc8 37. Bb5 Kg8 38. Bc6 h5 39. Rc1 Rf8
40. d7
Nxd7 41. Bxd7 f5 42. Ke5 g6 43. Bd6 Rf7 44. Be6 Kg7 45. Bxf7 e3 1-0
[Event "GoodEnginesAgain"]
[Site "TONY-SUA7L25HR1"]
[Date "2005.06.04"]
[Round "1"]
[White "TonysChess 0.01a"]
[Black "Freyr 1.0.67"]
[Result "1-0"]
[ECO "D38"]
[PlyCount "191"]
[EventDate "2005.??.??"]
1. d4 Nf6 2. c4 e6 3. Nc3 Bb4 4. Nf3 d5 5. Qa4+ Nc6 6. Ne5 Bd7 7. Nxd7
Qxd7 8.
e3 e5 9. dxe5 Bxc3+ 10. bxc3 Nxe5 11. Qxd7+ Nexd7 12. Rb1 O-O-O 13.
Ba3 Ne4 14.
Bb2 dxc4 15. Bxc4 Nd2 16. Kxd2 Ne5+ 17. Kc2 Nxc4 18. Bc1 Rhe8 19. Rb4
Re4 20.
Kb3 Nd2+ 21. Kb2 Re6 22. Rf4 f6 23. Ra4 Kb8 24. Rd4 Rxd4 25. cxd4 Rb6+
26. Kc2
Ne4 27. f3 Rc6+ 28. Kb3 Nf2 29. Rg1 f5 30. Bd2 g6 31. a3 g5 32. Rf1
Nd3 33. e4
f4 34. g3 fxg3 35. hxg3 Rb6+ 36. Kc3 Nb2 37. Bxg5 Na4+ 38. Kc4 Nb2+
39. Kc3
Na4+ 40. Kd3 Rb3+ 41. Kc4 Rb2 42. e5 h5 43. e6 b5+ 44. Kd5 Nb6+ 45.
Ke4 Re2+
46. Kf5 Nc4 47. Rc1 Nd6+ 48. Kf6 Rf2 49. f4 Rd2 50. Kg6 Rxd4 51. Kxh5
Rd3 52.
a4 Rxg3 53. Bd8 Ne8 54. axb5 Ng7+ 55. Kh6 Nxe6 56. Bg5 Nd4 57. Rb1 Rg2
58. Bf6
Nf5+ 59. Kh5 Ne3 60. Re1 Nc4 61. Bd4 Rd2 62. Re8+ Kb7 63. Be5 Na3 64.
f5 Nxb5
65. f6 Rf2 66. Kg6 Rg2+ 67. Kh5 Rf2 68. Kg6 Rf1 69. f7 Kc6 70. f8=Q
Rxf8 71.
Rxf8 a5 72. Ra8 Kb6 73. Bf6 Nd6 74. Bd8 Nb5 75. Kg7 Nd4 76. Kf7 Nc6
77. Bf6 Kc5
78. Ke8 Kb6 79. Kf8 Nb4 80. Kg7 Kb5 81. Bd8 Nd5 82. Rb8+ Kc6 83. Bg5
a4 84. Rb2
Kc5 85. Ra2 Kb5 86. Rd2 c6 87. Rb2+ Nb4 88. Bd2 c5 89. Rb1 a3 90. Kg6
Kc4 91.
Kf5 Kd3 92. Bxb4 cxb4 93. Rxb4 Kc2 94. Rc4+ Kb1 95. Ke6 Ka1 96. Kd7
1-0
[Event "GoodEnginesAgain"]
[Site "TONY-SUA7L25HR1"]
[Date "2005.06.04"]
[Round "1"]
[White "TonysChess 0.01a"]
[Black "Thinker 4.7a"]
[Result "1/2-1/2"]
[ECO "D02"]
[PlyCount "113"]
[EventDate "2005.??.??"]
1. Nf3 d5 2. d4 c6 3. Nc3 Nf6 4. e3 e6 5. Bd3 Nbd7 6. Ng5 h6 7. Nf3
Bd6 8. O-O
O-O 9. e4 dxe4 10. Nxe4 Nxe4 11. Bxe4 e5 12. dxe5 Nxe5 13. Nxe5 Bxe5
14. Qxd8
Rxd8 15. Re1 Re8 16. g3 Bg4 17. c3 Bc7 18. Bd2 Rad8 19. Bc2 Bb6 20.
Rxe8+ Rxe8
21. Re1 Rxe1+ 22. Bxe1 Bf3 23. a3 Kf8 24. b3 Ke7 25. c4 Bd4 26. a4 Ke6
27. h3
f5 28. Kf1 a6 29. Bd2 h5 30. a5 g6 31. h4 c5 32. Bc1 Bc3 33. Be3 Bb4
34. Kg1
Kd6 35. Bh6 Bxa5 36. Bf8+ Kc6 37. Be7 Bb4 38. Kf1 Bc3 39. Bf8 Bb4 40.
Be7 Bc3
41. Bf8 Bd4 42. Bh6 b6 43. Ke1 Kd7 44. Bf8 Bc3+ 45. Kf1 Ke6 46. Bh6
Kd6 47.
Bf4+ Kd7 48. Bh6 Kd6 49. Bf4+ Kd7 50. Bh6 Kc8 51. Bf8 Kc7 52. Be7 Kd7
53. Bf8
Ke8 54. Bd6 Kd8 55. Bf8 Kc7 56. Be7 Kd7 57. Bf8 1/2-1/2
Tony
JVMe...@aol.com
> On Wed, 8 Sep 2010 17:12:15 -0700 (PDT), raylopez99
>
> >I don't think they exist. That's because computers use the same
> >search engines at the brute force level, and the only intelligence is
> >the evaluation function.
>
> I realize you're sort of joking here, but this statement is untrue.
> Evaluation and search together provide "intelligence", and a program
> without search (eg CPP1) is not going to be very intelligent. Plus,
> searches are different in every program. Most programs use some sort
> of forward pruning, so most modern programs are not pure "brute force"
> programs. A stronger program may prune a good move that a weaker
> program doesn't, and that could be a possible downfall for the
> stronger program.
>
> >So unless a stronger computer walks into a
> >mating net that lies beyond it's search ply, and those must be rare (I
> >would imagine), I don't see how a much weaker chess program (say 300
> >Elo points weaker) can beat a stronger one.
It is indeed very rare for a weaker program to beat a considerably
stronger one, but it does happen. My program, Myrddin, has never done
it (unless you allow for the stronger engine crashing, which has
happened twice). However, it does have perfectly legitimate DRAWS
against engines that are MUCH stronger.
Here are two. The first is against Komodo, probably rated 1000 points
stronger. The other is against Hermann, probably about 600 points
stronger.
[Event "BASEMENT_QUAD_2009-11-7 (1)"]
[Site "BASEMENT_QUAD"]
[Date "2010.08.03"]
[Round "14.1"]
[White "Komodo 1.2"]
[Black "Myrrdin 0.83k 64bits"]
[Result "1/2-1/2"]
[TimeControl "3600"]
[Annotator "29. +0.44 25... -1.51"]
[Number "262"]
1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 O-
O 8.
c3 d5 9. exd5 Nxd5 10. Nxe5 Nxe5 11. Rxe5 c6 12. d4 Bd6 13. Re1 Qh4
14. g3
Qh3 15. Be3 Bg4 16. Qd3 Rae8 17. Nd2 Re6 18. a4 bxa4 19. Rxa4 f5 20.
Qf1
Qh5 21. f4 Rfe8 22. Rxa6 Rxe3 23. Rxe3 Rxe3 24. Rxc6 Qe8 25. Rxd6 Re1
{-1.51/16} 26. Bxd5+ Kf8 {-1.23/18} 27. Qxe1 Qxe1+ {-1.11/18} 28. Nf1
Be2
{-1.34/17 1:16} 29. Bg2 {+0.44/26 1:21} Bxf1 {-1.15/17 25} 30. Bxf1
{+0.19/26 1:19} h5 {-1.15/14 1:59} 31. Rd8+ {+0.60/20 1:17} Kf7
{-1.03/17 58} 32. Rh8 {+0.38/23 1:15} Kf6 {-1.18/15 2:12} 33. d5
{+0.06/21 1:14} Qe3+ {-0.82/15 53} 34. Kg2 {+0.00/25 1:12} Qd2+
{-1.00/16 23} 35. Kh3 {+0.00/28 1:10} Qxd5 {-0.84/14 2:48} 36. Re8
{+0.00/26 1:08} g5 {-0.62/15 55} 37. fxg5+ {+0.00/23 1:06} Kxg5
{-0.76/15 1:10} 38. c4 {+0.00/25 1:05} Qd2 {-0.49/16 4:42} 39. Rg8+
{+0.00/21 1:03} Kf6 {-0.16/14 1:47} 40. Bg2 {+0.00/23 1:02} Qxb2
{-0.04/13 1:26} 41. Rf8+ {-0.09/22 1:00} Kg5 {+0.00/18 4} 42. Rg8+
{-0.09/25 58} Kf6 {+0.00/19 45} 43. Rf8+ {+0.00/31 57} Kg5 {+0.00/15
39}
44. Rg8+ {+0.00/37 55} Kf6 {+0.00/21 56}
{Draw by Repetition} 1/2-1/2
[Event "BASEMENT_QUAD_2009-11-7 (1)"]
[Site "BASEMENT_QUAD"]
[Date "2010.08.04"]
[Round "15.1"]
[White "Myrrdin 0.83k 64bits"]
[Black "Hermann 2.536"]
[Result "1/2-1/2"]
[TimeControl "3600"]
[Annotator "18. +0.21 17... -0.17"]
[Number "282"]
1. d4 d5 2. c4 e6 3. Nc3 c6 4. Nf3 dxc4 5. e3 b5 6. a4 Bb4 7. Bd2 a5
8.
axb5 Bxc3 9. Bxc3 cxb5 10. b3 Bb7 11. bxc4 b4 12. Bb2 Nf6 13. Bd3 Nbd7
14.
O-O O-O 15. Qc2 Qc7 16. e4 e5 17. Rfe1 exd4 {-0.17/14 1:30} 18. Bxd4
{+0.21/12} Nh5 19. c5 {+0.04/11 5:02} Nf4 20. Bb5 {+0.27/13 54} Ba6
{+0.05/14 1:24} 21. Bxd7 {+0.15/13 2:55} Qxd7 {-0.02/13 1:21} 22. Be5
{-0.02/15 22} Nd3 23. Red1 {+0.03/15} Qb5 24. Bd6 {-0.04/14} Rfe8 25.
Nd4
{+0.22/15} Qc4 26. Qxc4 {+0.00/15 1:18} Bxc4 {+0.13/17 3:53} 27. c6
{-0.21/15 5} Rxe4 28. Nf5 {-0.20/14 1:37} Ree8 {+0.00/15 1:04} 29. c7
{-0.56/13 1:18} b3 {+0.42/15 1:03} 30. Ne7+ {-0.56/14 1:01} Kh8
{+0.45/16 1:01} 31. Ra3 {-0.60/14 2:13} a4 {+0.76/15 59} 32. f4
{-0.58/14 1:32} b2 {+0.78/16 58} 33. Rc3 {+0.03/14 33} Nc1 {+0.88/17
56}
34. Rxc4 {-0.78/14 40} b1=Q {+0.21/15 55} 35. Kh1 {-0.78/11 34} a3
{+0.90/16 54} 36. Rcxc1 {-0.26/14 30} Qb7 {+0.89/16 52} 37. Bxa3
{-0.53/14 3:21} h6 38. Bd6 {-0.55/14} f6 39. h3 {-0.53/10 3:01} Kh7
40. Rd2
{+0.00/17 2:42} Rxe7 {+0.04/17 47} 41. Bxe7 {+1.21/13 16} Rc8
{+0.00/15 46}
42. Bd6 {+1.13/14 1:10} Qa6 {-0.04/15 45} 43. Kh2 {+1.35/13 15} Qb5
{-0.24/15 44} 44. Rcc2 {+1.19/13 4} Qb3 45. Rb2 {+1.12/12 2.4} Qc3 46.
g3
{+1.13/15 24} h5 {-0.21/17 41} 47. Rg2 {+1.21/13} h4 48. gxh4
{+1.63/14 59}
Qd3 {-0.22/18 39} 49. Rbd2 {+1.51/14 48} Qc4 {-0.22/15 38} 50. Rg3
{+1.33/12 36} Qe6 {-0.22/14 37} 51. Ra3 {+1.41/14 1:45} Qe1 {-0.08/17
36}
52. Rg2 {+1.33/14 42} Qd1 {+0.00/15 35} 53. Ra6 {+1.30/14 31} Re8
{+0.00/13 34} 54. h5 {+1.02/12 16} Qxh5 {+0.00/14 33} 55. Rc6
{+0.81/12 35}
Qf5 {+0.27/13 32} 56. Rc5 {+0.66/12 1:24} Qd7 {+0.00/15 31} 57. Rd5
{+0.81/12 28} Qc6 {+0.27/13 31} 58. Rd3 {+0.73/12 41} Rg8 {+0.07/14
58} 59.
Rgd2 {+0.68/12 25} Ra8 {+0.14/14 28} 60. Rg3 {+0.79/12 1:09} Qd7
{+0.03/15 28} 61. Rc3 {+0.97/12 35} Rc8 {+0.07/13 27} 62. Rcc2
{+0.79/13 36} Qa4 {+0.06/14 26} 63. Rc5 {+0.72/13} Qb3 64. Rh5+
{+0.87/13 14} Kg6 {+0.00/18 25} 65. Rhd5 {+0.79/14 1:14} Qc4 66. f5+
{+1.07/12} Kh5 67. R5d4 {+1.03/14 22} Qc6 {+0.00/18 23} 68. Bg3
{+0.83/14 21} Rxc7 {+0.00/16 23} 69. Bxc7 {+0.80/15} Qxc7+ 70. Kg2
{+0.80/16 19} Qc6+ 71. Rd5 {+0.80/16 12} Qb7 {+0.00/21 21} 72. Rd3
{+0.83/14 17} Qc6 {+0.00/20 20} 73. Kf2 {+0.83/14 11} Qb6+ {+0.00/17
20}
74. Ke2 {+0.82/15 25} Qb1 {+0.00/17 19} 75. R5d4 {+0.63/14 11} Qc2+
{+0.00/13 19} 76. Kf1 {+0.23/13 24} Qc1+ {+0.00/16 18} 77. Kg2
{+0.17/14 1:13} Qc5 {+0.00/17 18} 78. Rd8 {+0.18/13 11} Qc2+ {+0.00/17
18}
79. Kf1 {+0.00/13 12} Qc1+ {+0.00/18 17} 80. Rd1 {+0.00/14 9} Qf4+
{+0.00/17 17} 81. Kg2 {+0.00/15} Qe4+ 82. Kg3 {+0.00/15} Qe3+ 83. Kg2
{+0.00/16} Qe4+ 84. Kh2 {+0.00/14 19} Qf4+ {+0.00/20 15} 85. Kh1
{+0.00/15 8} Qf3+ {+0.00/16 15} 86. Kh2 {+0.00/17} Qf2+ 87. Kh1
{+0.00/16}
Qf3+ 88. Kh2 {+0.00/20} Qf2+ 89. Kh1 {+0.00/5 58} Qf3+
{Draw by repetition} 1/2-1/2
Myrddin has also drawn DanaSah and Amyan, and both are about as strong
as Hermann.
jm
The Master
That depends... on how much you are willing to pay.
BTW, that was very sneaky of you, substituting '600' where you
originally had '300'! Sneaky tactical tricks like that may work on
the weak-minded (i.e. TK, MH, and their ilk) but your Jedi mind
tricks won't work on me. (Oh, and you are supposed to use The
Force for *good*, not evil. Duh.)
The Master
> > We're talking computers Minor, not humans. Can you find me a game
> > with a 600 Elo difference that was won by the weaker machine?
>
> GetClub once beat Jester 2 years back when Jester was arround 20 times
> stronger than GetClub.
>
> Today, even after 20 fold improvements GetClub finds it difficult to
> beat Jester.
>
> So Luck may favour a weak engine.
In this case, it may be that the operator had something to do
with the outcome. I think what Phil-Ray is looking for is a game
in which a significantly weaker program won on its own, sans any
hanky-panky or interference or incompetence by the computer
operator.
One way to find such games is to use the google search engine.
Advanced techniques like typing two words into a search engine
are far beyond the abilities of Phil-Ray, because after typing them
in correctly you must press the 'enter' key and then examine the
results. Note that this process involves more than one easy step,
thus exceeding his limited abilities. :)
raylopez99
> One way to find such games is to use the google search engine.
> Advanced techniques like typing two words into a search engine
> are far beyond the abilities of Phil-Ray, because after typing them
> in correctly you must press the 'enter' key and then examine the
> results. Note that this process involves more than one easy step,
> thus exceeding his limited abilities. :)
Why don't you Google it and post then? I'll shut up if you do.
Cordially,
Phil (but you can call me Ray)
raylopez99
> GetClub once beat Jester 2 years back when Jester was arround 20 times
> stronger than GetClub.
>
> Today, even after 20 fold improvements GetClub finds it difficult to
> beat Jester.
>
> So Luck may favour a weak engine.
>
> Bye
> Sanny
Please post a game where your weak "GetClub" beats your strong
"Jester".
That would be interesting. Maybe if more people visit your site (if
you make it interesting, like showing a weak engine beating a strong)
you might get more popular. Less people visiting, less popular. So
try something unique like what I suggest.
RL
chip...@gmail.com
>
> Please post a game where your weak "GetClub" beats your strong
> "Jester".
I don't recall Sanny *ever* posting a game that GitClub won against
another computer. I could be proven wrong of course.
The Master
As far as what you can recall, that only you know for sure.
But as far as what Sanny has posted here in rgc, I can recall
games in which he *claimed* GetClub defeated Rybka and Jester.
Of course, those claims were preposterous and the explanation
turned out to be, as far as we could tell, a sort of lock-up by Rybka
when she spotted a forced checkmate. In other words, Rybka
would not actually play the game out but merely displayed the
forced mate. Sanny, not having a clue about chess, interpreted
this as a resignation and claimed victory for his program.
Similarly,
when Sanny faced off his program against Jester, strange results
were reported -- results which can most easily be explained as
operator incompetence. When others play Jester online, they tend
to report that it plays much stronger than does GetClub. These
strange, anomalous reports of GetClub's great success seem to
emanate from just one source: Sanny.
raylopez99
Good points. I notice in computer vs computer play, they always have
"evenly matched" computers.
I think there's a reason for this. Since computers will never make a
mistake within their event horizon (the number of moves they can
calculate by brute force, typically four or five), the only way they
can lose is if they incorrectly rank the move at the end of their
event horizon (i.e., the four or fifth move). This is of course the
intelligence of the machine--the scoring function of the last move.
If they score a promising looking move and play that line, that
ultimately leads to mate, then they lose to the other machine, which
may not have scored that line. But suppose GetClub can only evaluate
by brute force three moves ahead (and scores the third move). Now
supposed Rybka can calculate by brute force five moves ahead.
Logically, unless some unusual trap exists in the chess tree (which
I'm sure they exist, but are rare), GetClub will never beat Rybka
since GetClub's best hunch is within the event horizon of Rybka.
That's to say: GetClub might think it's found a promising line, but
Rybka, since it calculates further ahead, knows for sure this line
loses.
The only way GetClub can beat Rybka under this scenario is as
follows: a line of moves exist in the chess tree such that the third
move looks promising (i.e., open files for a rook, no pawn island
created, etc), that GetClub choses, but on the fifth move (two moves
later) under best play is inferior (as seen by Rybka). But, unknown
to Rybka, on the sixth, seventh, eighth or subsequent moves, this line
leads to a decisive mating net (or massive loss of material, same
thing). Hence Rybka, with its superior knowledge, actually is at a
disability to GetClub. Better to have been 'simple minded' and seen
what GetClub, with its limited intelligence, saw. A corollary: Rybka
sees a line that GetClub cannot see, that looks superior, but in fact
walks into a mating net. So again Rybka would 'beat itself' through
its superior intelligence.
But how common are such lines in the chess tree? Very rare I would
imagine. Hence the seemingly non-existent wins by a much inferior
chess program to a superior chess program.
RL
MikeMurray
<raylo...@gmail.com> wrote:
>The only way GetClub can beat Rybka under this scenario is as
>follows: a line of moves exist in the chess tree such that the third
>move looks promising (i.e., open files for a rook, no pawn island
>created, etc), that GetClub choses, but on the fifth move (two moves
>later) under best play is inferior (as seen by Rybka). But, unknown
>to Rybka, on the sixth, seventh, eighth or subsequent moves, this line
>leads to a decisive mating net (or massive loss of material, same
>thing). Hence Rybka, with its superior knowledge, actually is at a
>disability to GetClub. Better to have been 'simple minded' and seen
>what GetClub, with its limited intelligence, saw. A corollary: Rybka
>sees a line that GetClub cannot see, that looks superior, but in fact
>walks into a mating net. So again Rybka would 'beat itself' through
>its superior intelligence.
>But how common are such lines in the chess tree? Very rare I would
>imagine. Hence the seemingly non-existent wins by a much inferior
>chess program to a superior chess program.
So is there a way to quantify this ?
I first thought one might get an empirical feel for the rarity of such
lines (relative to the event horizons of the respective programs) by
checking the score of matches between programs of disparate ratings.
If such lines didn't exist, it seems reasonable to assume the higher
rated program would always win or draw. So, in a given match, if the
score were 10 wins, 2 losses, 4 draws, one might guess these anomalous
positions occurred in 1/8 of the games. BUT, winning lines outside
both programs' event horizons and ignored by BOTH would not have
been counted, so this doesn't seem to work.
johnny_t
chess database programs...
http://computerchess.org.uk/ccrl/4040/games.html
I am sure tons of games match your criteria.
The first is the issue of event horizon. This surely has happened. But
this can also be caused by similar searches having slightly different
lengths of time in certain positions that would exaggerate the problem.
But the contention that they all have the same search is simply not true.
One of the biggest expansions of computer chess is selective search.
First really shown off in Fruit, then Rybka, possibly Fritz, and now in
Stockfish.
Event horizon problems are the number one reason for winning and losing
in computer chess, and pushing out the horizon as far as possible, and
enveloping your opponents horizon as much as possible simply leads to
more wins.
Selective search was the way to dramatically pare back the problem of
geometric increase in search compared to brute strength.
The speed of the search and selectivity derived from eval is what has
largely pushed the envelope.
This means that there is an opening for brute force, which out of a
quarter of a million games, has likely happened. And that is a move
sequence that wins, but is cut from selective search until it is too
late to recover.
But anyways there is a quarter of a million games, go practice your
chess database skills.
Martin Brown
The canonical example where this sort of thing happens is where a
machine with 6 men tablebases can see that its position is lost with
perfect play and so plays the line that maximises time to losing. This
may not be an optiimal strategy against an imperfect opponent.
However, in some of these circumstances only another machine armed with
perfect play could win and there is a very good chance for an engine or
human to make a mistake that turns the tables. It follows a line which
whilst not optimal keeps the position complicated and provides
opportunities for the opponent to make a mistake.
Regards,
Martin Brown
JVMe...@aol.com
wrote:
> On 10/09/2010 16:00, MikeMurray wrote:
>
>
>
>
>
> > On Fri, 10 Sep 2010 02:25:59 -0700 (PDT), raylopez99
I have some examples of this (or other weirdness caused by
tablebases).
Here's a position in which White is clearly lost even though White is
up a pawn and there isn't much material left on the board
8/8/3Q1rk1/4R3/2q5/6P1/8/7K w - -
Black is threatening Qf1+ followed by Rf2+, and there is no way for
White to avoid trading queen for rook. My engine (using 5-piece tbs)
saw this quickly, and avoided reducing down to five pieces by making
the seemingly horrible blunder -- Rg5+?? -- giving up a rook instantly
but allowing the queen to stay on the board longer -- in effect, past
the search horizon.
But here are some examples of what I think Martin is talking about.
8/3k4/8/8/P1P5/1PK5/r7/2B5 w - -
The thing to do here is avoid Bb2. Your program may be tempted into
blocking off the rook with the bishop, then walking the king around to
win the rook. But this would allow the drawing Rxa4! bxa4 Kxc4.
8/8/8/2k5/P1P5/1P6/rBK5/8 w - - 0 48
Best move here is Kc3 for the same reason.
In other words, although it will still be a hard fight with BPPP v R,
you have to let the rook free to have any chance at a win.
jm
Mark Houlsby
Greg, this is Ray, not Phil. Phil is as deranged as you are. Ray is
sane, and smart, and he's running rings around you.
If your intention is to continue to appear to be a deranged fuckwit,
you
should continue to reply to Ray, who is not Phil.
If you'd rather that didn't happen, stop responding to him.
That's free advice.
Mark Houlsby
Do you have the complete games scores, please?
mh
None
> Greg, this is Ray, not Phil.
No shit?
> That's free advice
Worth the cost.
Mark Houlsby
Why thank you, Stan, I like to think so.
Whether Greg will take it in is quite another matter.
raylopez99
> Here is where you can get 270k+ games for downloading into your favorite
> chess database programs...
>
> http://computerchess.org.uk/ccrl/4040/games.html
>
> I am sure tons of games match your criteria.
Perhaps. Thanks for the link. You seem to know a thing or two about
computer chess, like Dr. Hyatt. The games are just collections, but
I'll check them out and if anything interesting appears I'll post. I
quickly checked out one database (ChessTiger) and just like I thought
it's mostly against equal rated opponents. I'll keep looking though.
>
> The first is the issue of event horizon. This surely has happened. But
> this can also be caused by similar searches having slightly different
> lengths of time in certain positions that would exaggerate the problem.
>
> But the contention that they all have the same search is simply not true.
>
> One of the biggest expansions of computer chess is selective search.
> First really shown off in Fruit, then Rybka, possibly Fritz, and now in
> Stockfish.
Wow! This is shocking. I have followed computer chess back in the
days when they published a paper magazine out of Canada (forget the
CSci PhD who published it, but it was pretty famous) and I was under
the impression that everybody, even ChessMaster 2000, did selective
search?! Are you sure about this? Fruit was first? I find that hard
to believe.
> But anyways there is a quarter of a million games, go practice your
> chess database skills.
I can't. The site says "No aggressive download managers allowed".
Can you, a member in good standing (?) put in a special request of the
kind I inferred in my OP? Thanks in advance if you can.
RL
JVMe...@aol.com
>
> - Show quoted text -
I do not, I'm afraid. The second and third positions were given to me
by another chess engine programmer who I can no longer recall.
jm
Mark Houlsby
Ok, thank you anyway.