Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Perfect "God" Play ELO = 3800?
1 2 3 Previous Next  
Poll Perfect "God" Play ELO... (Closed)
< 3800. 5 8%
<= 3800. 0 0%
approx 3800. 5 8%
>= 3800. 4 7%
> 3800. 35 57%
No Idea!! 12 20%
Parent - - By turbojuice1122 (Gold) Date 2010-09-19 20:15
Obviously a few would disagree with this, but I think that a strong enough engine like Cluster Rybka thinking at a certain ply or beyond will get a draw against ANY engine thinking at ANY ply.

Queen odds games are generally seen as something that someone rated X elo above someone else should be able to do.  I think that the inability to extend this beyond a certain elo is due to the "win boundary" of these odds, and that by looking at it from the opposite angle, one concludes that the draw boundary prevents one from being able to match engines against each other with the 2x ply scheme and expect a win beyond a certain value x.  I would guess that this value is between about 30 and 40 for Rybka.
Parent - - By Banned for Life (Gold) Date 2010-09-19 20:51
Obviously a few would disagree with this, but I think that a strong enough engine like Cluster Rybka thinking at a certain ply or beyond will get a draw against ANY engine thinking at ANY ply.

This was the same argument you made several years ago for R2.3.2a until it was disproven by R3. It was wrong then, and it's wrong now, because engines are nowhere near as close to perfection as you hypothesize.
Parent - - By turbojuice1122 (Gold) Date 2010-09-21 11:08
How was this wrong then?  How was it disproven?  Put Rybka 2.3.2a at 40 ply from the starting position, and have it play against Rybka Cluster at any ply, and I guarantee the result will be a draw.  The drawish nature of chess is simply too high.  Of course, obviously this would not occur if you have a very difficult starting position where things can go wildly either way.
Parent - By Banned for Life (Gold) Date 2010-09-21 13:14
Put Rybka 2.3.2a at 40 ply from the starting position, and have it play against Rybka Cluster at any ply, and I guarantee the result will be a draw.

You can only guarantee the results of this experiment because it cannot be carried out. Any realizable experiment, e.g. running both for 24 hours a move, or a week a move, or a month a move would show the contrary in a decisive manner. The reality is that R2.3.2a has serious flaws in its search and evaluation algorithms that can be exploited by the current engines which have less serious flaws.

Of course an engine that uses pure alpha-beta search and runs until each node reaches an unambiguous conclusion (win, lose, or draw) will play perfect chess, but even this engine would be beaten if it were limited to a ply depth that was insufficient to see unambiguous node values.
Parent - By BankShots (***) Date 2010-09-20 18:09
Here's a short Odds Tournament:
Elos:  White ~ 1597, Black ~ 3308.
http://computerchess.org.uk/ccrl/404/

DpR4x64 vs Pir0.5, Qs Odds for Black, B  0

                         12345678
1   Piranha 0.5           ½1½1½0½0   4.0/8  0.00
2   Deep Rybka 4 x64 p1  ½0½0½1½1    4.0/8  0.00

Rybka's two wins:
[Event "DpR4x64 vs Pir0.5, Qs Odds for Black, B"]
[Site ""]
[Date "2010.09.20"]
[Round "6"]
[White "Deep Rybka 4 x64 p1"]
[Black "Piranha 0.5"]
[Result "1-0"]
[Annotator "-9.91;-10.93"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[PlyCount "109"]
[TimeControl "240"]

{AMD Phenom(tm) II X4 955 Processor 3214 MHz  W=14.6 plies; 299kN/s  B=5.9
plies; 93kN/s} 1. Nf3 {-9.91/16 8} d5 {-10.93/7 5} 2. d4 {-9.86/16 10} Nc6 {
-11.04/6 5 (Bc8-f5)} 3. h3 {-9.87/16 8 (e2-e3)} Bf5 {-11.32/6 4 (Ng8-f6)} 4. c3
{-9.92/17 6 (Nb1-c3)} Bxb1 {-11.64/6 4 (e7-e6)} 5. Rxb1 {-9.41/14 1} e6 {-11.
26/6 4 (Ng8-f6)} 6. g3 {-9.59/17 20 (e2-e3)} Qf6 {-12.01/6 4 (Qd8-d7)} 7. Be3 {
-9.53/15 7 (a2-a3)} Qf5 {-12.09/6 4 (h7-h6)} 8. Rc1 {-9.63/17 14 (Rb1-d1)}
O-O-O {-11.99/5 4 (h7-h6)} 9. Bg2 {-9.31/14 6} Bd6 {-11.44/5 4} 10. Nh4 {-9.35/
14 3 (0-0)} Qf6 {-11.55/6 4} 11. O-O {-9.45/16 6} Na5 {-11.38/5 4 (h7-h6)} 12.
Nf3 {-9.42/15 18 (b2-b3)} Ne7 {-11.40/5 4 (h7-h6)} 13. Bg5 {-8.89/14 4} Qg6 {
-11.36/6 4 (Qf6-f5)} 14. Bxe7 {-8.67/13 2 (b2-b4)} Bxe7 {-12.05/6 4} 15. Ne5 {
-8.71/14 3 (Nf3-d2)} Qf6 {-12.01/6 4} 16. Ng4 {-8.71/12 3 (Bg2-f3)} Qf5 {-12.
01/6 3} 17. Ne3 {-8.75/15 4 (Rc1-e1)} Qg6 {-12.03/6 3 (Qf5-h5)} 18. Rfd1 {-9.
06/14 10 (b2-b3)} Nc4 {-12.04/5 3 (Kc8-b8)} 19. Nxc4 {-9.00/12 1} dxc4 {-12.04/
6 4} 20. a4 {-9.17/13 11} h5 {-12.17/5 4} 21. Rf1 {-9.73/14 10 (e2-e3)} Qf5 {
-12.58/5 4 (h5-h4)} 22. e3 {-9.34/15 3 (h3-h4)} Kb8 {-12.03/5 4 (c7-c6)} 23.
Ra1 {-9.33/16 3 (Kg1-h2)} Qc2 {-12.25/5 4 (h5-h4)} 24. Rfc1 {-9.45/16 5
(Rf1-b1)} Qb3 {-12.25/6 4} 25. Bf3 {-9.59/16 5 (Rc1-b1)} h4 {-13.01/5 3 (f7-f5)
} 26. g4 {-9.43/13 1} f5 {-12.84/5 3} 27. Rcb1 {-9.70/14 3 (g4xf5)} Qc2 {-13.
02/5 3 (f5xg4)} 28. Rc1 {-8.56/10 1} Qd3 {-12.95/6 3 (Qc2xb2)} 29. Rd1 {-5.59/
10 0} Qc2 {-12.95/6 3 (Rd8xd4)} 30. Rdc1 {0.00/22 0} Bg5 {-1.35/6 3 (Qc2xb2)}
31. Rxc2 {0.46/12 0} Rd6 {-1.36/6 3} 32. Be2 {0.51/15 1} fxg4 {-1.17/6 3
(Rd6-c6)} 33. hxg4 {0.68/15 2 (Be2xg4)} Rc6 {-1.01/6 3} 34. f4 {0.80/15 3} Be7
{-1.10/7 3} 35. g5 {1.11/13 1 (Be2-f3)} Re8 {-0.43/6 3 (Rh8-d8)} 36. Kg2 {1.33/
17 2 (a4-a5)} g6 {-0.78/6 3 (Re8-h8)} 37. Kh3 {1.78/16 3 (Rc2-c1)} Rh8 {0.10/7
3} 38. a5 {1.91/15 1 (Be2-f3)} Bd8 {-0.09/6 3 (b7-b5)} 39. Ra4 {2.11/15 1
(Be2-f3)} Be7 {1.22/6 2 (Rc6-d6)} 40. Bxc4 {2.37/13 2} Rd6 {0.84/6 2} 41. a6 {
2.53/13 1 (Ra4-a2)} bxa6 {1.21/6 2 (b7-b6)} 42. Bxa6 {2.76/13 1 (Rc2-g2)} Rdd8
{0.74/6 2 (Be7-d8)} 43. b4 {3.08/15 3 (Rc2-h2)} Rhg8 {2.07/6 2 (e6-e5)} 44.
Rca2 {4.15/14 2 (Kh3xh4)} e5 {2.26/6 2 (Rd8-d6)} 45. Bc4 {4.86/13 2} exf4 {3.
44/6 2 (Rg8-g7)} 46. Bxg8 {5.56/13 0} fxe3 {3.88/6 2 (Rd8xg8)} 47. Bc4 {6.62/
11 1 (Ra4xa7)} Bxg5 {4.02/6 2 (c7-c6)} 48. Ra5 {7.15/14 0 (Ra4xa7)} Bf6 {3.71/
6 2 (Bg5-f4)} 49. Rb5+ {10.25/9 0 (Ra5xa7)} Ka8 {5.03/6 2 (Kb8-c8)} 50. Ra6 {
11.27/10 0 (Bc4-d5+)} Bg7 {5.28/6 2 (Bf6xd4)} 51. Bf1 {#5/17 2 (Bc4-d5+)} c6 {
#4/7 2 (Rd8-b8)} 52. Bg2 {#4/21 2 (Ra6xc6)} Rd5 {#3/7 2} 53. Rxc6 {#3/20 1} Be5
{#2/7 2} 54. Bxd5 {#2/18 1 (Rc6-c8+)} g5 {#1/8 2 (e3-e2)} 55. Rc8# {#1/18 1}
1-0

[Event "DpR4x64 vs Pir0.5, Qs Odds for Black, B"]
[Site ""]
[Date "2010.09.20"]
[Round "8"]
[White "Deep Rybka 4 x64 p1"]
[Black "Piranha 0.5"]
[Result "1-0"]
[Annotator "-9.87;-10.93"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNB1KBNR w KQkq - 0 1"]
[PlyCount "139"]
[TimeControl "240"]

{AMD Phenom(tm) II X4 955 Processor 3214 MHz  W=14.1 plies; 338kN/s; 72 TBAs
B=8.8 plies; 92kN/s} 1. Nf3 {-9.87/17 18} d5 {-10.93/7 5} 2. d4 {-9.88/16 9}
Nc6 {-11.04/6 5 (Ng8-f6)} 3. c3 {-9.85/17 9 (e2-e3)} Bg4 {-11.51/6 4 (Ng8-f6)}
4. Bf4 {-9.71/17 12} e6 {-11.60/6 4 (Ng8-f6)} 5. Nbd2 {-9.78/17 18} Bxf3 {-11.
51/5 4 (Ng8-f6)} 6. exf3 {-9.55/18 12 (Nd2xf3)} g6 {-11.71/6 4 (Bf8-d6)} 7. Be3
{-9.44/15 5 (Bf1-d3)} Qh4 {-11.53/6 4 (a7-a5)} 8. Be2 {-9.29/15 12 (g2-g3)}
O-O-O {-11.59/5 4 (a7-a5)} 9. O-O {-9.28/14 5 (0-0-0)} Nf6 {-11.83/6 4 (g6-g5)}
10. a4 {-9.22/13 4 (g2-g3)} Bh6 {-11.90/5 4 (Kc8-b8)} 11. f4 {-9.22/15 7} Ng4 {
-11.89/6 4 (Nf6-e4)} 12. Nf3 {-8.89/13 2} Qf6 {-11.99/6 5 (Qh4-e7)} 13. g3 {-8.
95/14 4 (Nf3-e1)} Nxe3 {-12.33/5 4 (Nc6-a5)} 14. fxe3 {-8.55/11 1} Qf5 {-12.12/
6 4} 15. Rf2 {-8.70/12 2 (Be2-b5)} Qe4 {-12.08/5 3 (g6-g5)} 16. Re1 {-8.43/14
15} a6 {-12.04/5 3 (Bh6-g7)} 17. Bf1 {-8.30/15 4} f6 {-11.89/5 4 (Qe4-f5)} 18.
Bg2 {-6.36/13 1 (c3-c4)} Qd3 {-12.03/5 4 (Bh6xf4)} 19. Bf1 {-4.68/13 0} Qf5 {
-11.93/6 4} 20. Nh4 {-4.71/13 0 (c3-c4)} Qg4 {-11.79/6 4 (Qf5-e4)} 21. Be2 {-3.
00/11 0 (e3-e4)} Qh3 {-11.69/6 4} 22. Bf1 {-3.00/11 0 (b2-b4)} Qg4 {-11.79/6 4
(Qh3xf1+)} 23. Be2 {0.00/23 0} Qh3 {-11.64/6 4} 24. Bf1 {0.00/23 0 (c3-c4)}
Rdg8 {-0.95/6 4 (Qh3xf1+)} 25. Bxh3 {0.18/15 1} f5 {-0.95/6 3} 26. Nf3 {0.22/
16 3 (b2-b4)} g5 {-0.80/6 3 (Kc8-d7)} 27. fxg5 {0.30/16 3 (Nf3-e5)} Bxg5 {-0.
64/6 3} 28. c4 {0.30/16 2} Nb4 {-0.85/6 3} 29. cxd5 {0.33/16 1 (Bh3-f1)} Nd3 {
-1.13/6 3} 30. Ree2 {0.28/17 3 (Nf3xg5)} Nxf2 {-0.55/6 3} 31. Kxf2 {0.28/18 1}
exd5 {-0.13/7 3} 32. Bxf5+ {0.28/17 1} Kb8 {-0.22/7 2} 33. Be6 {0.28/18 1} Rg7
{0.10/6 2 (Rg8-g6)} 34. Bxd5 {0.31/15 1} Rd8 {0.07/6 2 (c7-c6)} 35. Be6 {0.26/
15 5 (Bd5-e4)} Rf8 {0.21/6 2 (Bg5-f6)} 36. e4 {0.56/16 5 (Kf2-g2)} Bh6 {0.10/5
2 (c7-c6)} 37. e5 {0.57/15 3 (Re2-c2)} Rg6 {0.02/6 3 (c7-c6)} 38. Bc4 {0.68/16
3 (d4-d5)} Bc1 {0.09/5 3 (c7-c6)} 39. Kg2 {0.96/14 3} Rb6 {0.25/5 2 (c7-c6)}
40. b3 {0.90/13 1} Rg6 {0.43/6 2 (Rb6-h6)} 41. e6 {1.02/14 3} Rg7 {0.73/5 2}
42. Ne5 {0.93/15 1 (e6-e7)} Ka8 {0.59/6 2 (Kb8-c8)} 43. d5 {1.20/14 3} h5 {0.
66/5 2 (Bc1-a3)} 44. Nd7 {1.45/15 3 (Re2-c2)} Rc8 {0.89/5 2 (Rf8-h8)} 45. d6 {
2.85/12 0 (Bc4-d3)} Rgg8 {0.92/5 2 (Bc1-g5)} 46. e7 {7.42/13 1} Rge8 {2.41/6 2
(Rg8-h8)} 47. Nf6 {7.86/12 0 (Bc4-f7)} cxd6 {6.18/6 2} 48. Nxe8 {7.86/14 0}
Rxe8 {7.06/7 2} 49. Bf7 {7.86/14 0} Rc8 {7.10/6 2 (Re8xe7)} 50. e8=Q {8.72/15
2 (Bf7xh5)} Rxe8 {7.32/7 2} 51. Rxe8+ {8.72/15 0} Ka7 {0.00/100 0} 52. Bxh5 {
8.72/16 0} b5 {7.23/7 2 (d6-d5)} 53. h4 {9.31/13 2 (a4xb5)} bxa4 {6.92/7 2
(Bc1-b2)} 54. bxa4 {9.80/14 1} Kb6 {7.13/7 2} 55. g4 {9.80/13 0} Ka5 {6.92/7 2
(Kb6-c7)} 56. g5 {11.33/12 1 (Re8-e1)} Kxa4 {6.83/7 2} 57. g6 {14.47/12 0
(Re8-e6)} Bh6 {7.63/6 2 (Bc1-b2)} 58. Rh8 {14.46/10 0 (Re8-e7)} Bg7 {8.04/7 2}
59. Rh7 {16.76/10 0} Bd4 {8.11/7 2 (Bg7-e5)} 60. g7 {18.31/11 1} Bxg7 {8.34/7 1
} 61. Rxg7 {#12/10 1} a5 {8.46/7 1} 62. Bd1+ {#12/9 0 (Rg7-a7)} Kb4 {8.55/6 1
(Ka4-a3)} 63. h5 {#11/9 0} Kc3 {11.79/7 1 (a5-a4)} 64. h6 {#7/12 3} Kd2 {14.63/
7 1} 65. Rc7 {#6/11 1 (Bd1-f3)} Kxd1 {13.36/8 1 (a5-a4)} 66. Kf3 {#5/12 1
(h6-h7)} Kd2 {#4/8 1} 67. h7 {#4/12 1} a4 {#3/8 1 (Kd2-e1)} 68. h8=Q {#3/10 3}
a3 {0.00/6 1} 69. Qd4+ {#2/10 4} Ke1 {0.00/100 0} 70. Rc1# {#1/10 1} 1-0

In both games DpRybka4x64 won at Qs Odds against, "her" opponent seems to have a contempt-for-draw problem. :twisted:
Parent - By BankShots (***) Date 2010-09-20 18:14
I had to vote >=3800. :grin:
- - By wem511 (**) Date 2010-09-21 04:57
what is this 'Protector' program he speaks of, being  stornger than rybka???!
Parent - By Uly (Gold) Date 2010-09-21 06:12
http://protector.svn.sourceforge.net/viewvc/protector/bin/

^There.

Source (of the link, not the code):

http://www.open-aurec.com/wbforum/viewtopic.php?f=2&t=50337

It's not stronger than Rybka, it's just another flavor of Fruit/Toga.
Parent - By mindbreaker (****) Date 2010-09-21 16:48
Stronger than Fritz 5.32 not Rybka.
Parent - - By mindbreaker (****) Date 2010-09-21 16:58 Edited 2010-09-21 17:06
There was a claim that Fritz 5.32 would get "draws" against Rybka 4 in a 100 game match.  I was skeptical showing that a four-thread 2950 engine did not get a draw in more than 100 games.  But I tested the claim and neither of us was right.  It was not perfect and it did not give up draws plural just one.

To be fair he was talking about the default Rybka 4 which may well have yielded more draws.
Parent - - By Uly (Gold) Date 2010-09-22 00:58
I still think Rybka 3 Dynamic at contempt 30 would win all games :smile:
Parent - - By mindbreaker (****) Date 2010-09-22 03:33
The ideal would probably be something with a lot of knowledge and not very aggressive pruning but reasonably fast.  Out processing it should be easy, as it is 4 threads to 1, you just don't want any of Fritz's special knowledge to trip up the generally stronger opponent.  Hiarcs might be a good choice.  Maybe Rybka 2.3.2a.  Still, without extreme strength Fritz could just get lucky.  Dynamic is best at not loosing rather than avoiding draws.  I think there would be some perpetuals.  StockFish might be a good try too, even though it is brute force and aggressive pruning, the speed, depth, and strength should negate most special knowledge.

If we were just trying to avoid a loss, Dynamic would likely be a great choice.

Exp 31 might be a good choice; it held Deep Fritz 11 to 13.5% in a 100 game match.  Exp 24 also held it to 13.5%.  Exp 6 holds the record at almost 93% but there were too few games, probably just an anomaly.  But I don't know whether this approach makes sense as Deep Fritz 11 and Fritz 5.32 may have little in common besides the author.
Parent - By Uly (Gold) Date 2010-09-22 08:26

> Hiarcs might be a good choice.  Maybe Rybka 2.3.2a.


I don't think the change in knowledge of such engines related to Rybka 4 is critical. For your purposes, perhaps Zappa Mexico II with Null Move pruning disabled would do.

I still claim that playing style will have more weight in this than strength or knowledge, and Dynamic has it. Maybe Thinker too, and I recall Bright 0.2c had a very low drawing % against a variety of engines around here.

> Dynamic is best at not loosing rather than avoiding draws.


Not with contempt 30, it'll focus on going for unclear positions with material imbalances and avoiding draws, it sounds perfect.

> StockFish might be a good try too, even though it is brute force and aggressive pruning, the speed, depth, and strength should negate most special knowledge.


Probably, I'd suggest the 1.6s version.
- By Eagleclaww (***) Date 2010-09-28 20:37
I think at some point, we're going to hit a wall on how much a chess engine can improve without a substantial increase in speed. For example, running say a 24 hour game (12 hours each side) on  a tricked out computer (Not cluster, but say 6 core Intel chip), clearly would have vastly different games than say 40/40 games and 40/4 games that are used to populate the engine ranking sites.

It certianly shows about a +200 point gain over the same engine give a very long time control -vs- a normal time control.
i[That is, say White gets 120 minutes for a game, but black gets 12 or 24 hours.]

Now, not enough games have been played at that unbalanced time control to warrent an exact rating, but certainly does bring to light that given enough time, a chess engine will invariable make better moves.
-Its also not practical to use longer time controls for engine ranking anyway, but ok for some analysis like theoretical openings, etc.

I think its possible with just current technology trends (transistors getting smaller, and more cores per CPU) and minor improvements to the engine AI to break 3600 by 2015, or sooner.
Up Topic Rybka Support & Discussion / Rybka Discussion / Perfect "God" Play ELO = 3800?
1 2 3 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill