Not logged inRybka Chess Community Forum
- - By cma6 (****) Date 2008-01-11 23:07
I was interested in the mention of search depth in the discussion of the latest Rybka-Benjamin match. Here is a question for users of Rybka.

    What kind of search depth, for a typical middlegame position,  do you customarily get (on a fast O/C system) for both slow chess and correspondence chess?
                                                                
Parent - By Vasik Rajlich (Silver) Date 2008-01-12 09:52
Maybe 20 or so. It does vary quite a bit from position to position.

Vas
Parent - - By cma6 (****) Date 2008-01-13 01:28
Thanks, Vas, that is what I was looking for, a ballpark number. I get 20-26 depending on the complexity of the middlegame position.
Parent - - By Gaмßito (****) Date 2008-01-13 07:13
20-26? Really?

That is too much. What kind of machine do you have and how much time do you use?

Regards,
Gaмßito.
Parent - - By cma6 (****) Date 2008-01-14 00:07 Edited 2008-01-14 00:22
I have an intel Q6600 O/C with Phil Harris' guidance to 395 Mhz FSB (semi-stable) X 9 multiplier = 3.555 Ghz clock. Typical time for a move is 30 minutes, though it varies quite a bit from that average.  Typical kn/s is 675 in the DS11 GUI.
Parent - - By Permanent Brain (*****) Date 2008-01-14 00:51
I wonder what effect such enormous depths of 20+, or even 26(!!) plies in the middlegame have, on the style and on the role of combinations (or tactical play). I think such depths are fairly new except in corr.analysis, and have been achieved with the new engines and fast dual and quad cpus. Basically I do not doubt that 'tactical things' still happen, but a 20+ plies opponent should spot and defend threats up to 10 full moves deep. For a human mind, such things cannot be explained or analysed in a traditional way. Or are computer games more positional now? I am not sure if I am a player strong enough to answer that question myself.

I remember a long article on the so called 'diminishing returns' in computer chess from many years ago, where the conclusion was something like: After depths like 20 or 22, there will be no significant gain from even bigger depths anymore. Maybe that's somewhat simplified but that is what I remember as the main hypothesis, or prediction.

I am aware that due to more forward pruning or less, and/or due to the risks involved, calculations from different engines up to the same depth are not necessarlity worth the same (let alone differences of evaluation quality). Being no programmer, I hope I understood it correctly that it means something like a full alpha/beta search without forward pruning to depth x would be stronger - or at least much more reliable - than the one of a modern engine which prunes a lot. On the other hand, I have been told that in comparison to (very) old chess programs, the extensions of modern engines are much better and more effective, so that the new ones are actually stronger even at smaller depths than x, than the old ones at x.

That was after I had matched 13 old chess computers, set to 45...60 seconds per move, against some engines set to 10 plies of depth, ponder off. Result 1-51 Although, most of these comps of course didn't achieve 10 plies of depth on that level, except the Mephisto Risc which lost 0-4 against Rybka 1.0 beta :-D It was amazing how easily the engines beat these good chess comps, rarely using more than 3 seconds for a move.

A rare exception:

[Event "Nostalgie"]
[Site "Schrotty"]
[Date "2007.02.15"]
[Round "?"]
[White "Sphinx Dominator A6"]
[Black "Naum 2.0 T10"]
[Result "1/2-1/2"]
[ECO "D48"]
[WhiteElo "1880"]
[PlyCount "118"]

1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. e3 e6 5. Nf3 Nbd7 6. Bd3 dxc4 7. Bxc4 b5 8. Bd3
a6 9. e4 c5 10. d5 c4 11. Bc2 Bb4 12. a3 Bxc3+ 13. bxc3 Bb7 14. dxe6 fxe6 15.
Qd6 Bxe4 16. Qxe6+ Qe7 17. Qxe7+ Kxe7 18. Bxe4 Nxe4 19. Bd2 Ndc5 20. O-O Nxd2
21. Rfe1+ Kf6 22. Nxd2 Rhe8 23. Kf1 Rad8 24. Rxe8 Rxe8 25. Re1 Rd8 26. Ne4+
Nxe4 27. Rxe4 Rd1+ 28. Re1 Rd3 29. Re3 Rxe3 30. fxe3 Ke5 31. Ke2 Ke4 32. Kd2 g5
33. Ke2 g4 34. Kd2 h5 35. Ke2 a5 36. Kd2 h4 37. Ke2 h3 38. g3 a4 39. Kd2 Kf3
40. e4 Kxe4 41. Ke2 Kd5 42. Ke3 Ke5 43. Kd2 Kd6 44. Ke3 Kd5 45. Kd2 Ke5 46. Ke3
Kd6 47. Kd4 Kc6 48. Ke3 Kb7 49. Ke4 Ka6 50. Ke3 Ka7 51. Ke4 Kb7 52. Ke3 Kc6 53.
Kd4 Kb6 54. Ke4 Ka7 55. Kd4 Ka6 56. Ke3 Ka7 57. Kd4 Kb7 58. Kc5 Ka6 59. Kb4 Kb6
1/2-1/2

(Has an interesting pawn ending, see 40.e4! and Black could only fight for a win with 40...Kg2 instead of 40...Kxe4.)
Parent - By Gaмßito (****) Date 2008-01-14 06:15
Yes, after 40. e4 the only possibility for black to play for a win is Kg2! It is an interesting endgame.

Note that Fritz 11 took 21-22 plies to change Kxe4 and see Kg2. After all, it is not so easy.
Rybka 2.3.2a took 23-24 plies to see it.

The topic about deep search is incredibly important in chess. Hope that Vas can really improve Rybka 3 in this aspect, and can achieve a much more efficient search.

8/8/8/1p6/p1p3p1/P1P1PkPp/3K3P/8 w - - 0 1


Analysis by Rybka 2.3.2a 32-bit :

40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5
  -+  (-2.21)   Depth: 11   00:00:00
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5
  -+  (-2.21)   Depth: 11   00:00:00  0kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5
  -+  (-2.21)   Depth: 11   00:00:00  0kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5
  -+  (-2.21)   Depth: 11   00:00:00  0kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5
  -+  (-2.21)   Depth: 11   00:00:00  0kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5
  -+  (-2.21)   Depth: 11   00:00:00  0kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5
  -+  (-2.21)   Depth: 11   00:00:00  0kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kd5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kd5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kd5
  -+  (-2.21)   Depth: 12   00:00:00  1kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kd5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kd5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kd5
  -+  (-2.21)   Depth: 13   00:00:00  1kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kf5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kf5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kf5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kf5
  -+  (-1.52)   Depth: 14   00:00:00  2kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kf5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kf5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kf5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kf5
  =/+  (-0.59)   Depth: 15   00:00:00  5kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kf5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kf5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kf5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kf5
  =/+  (-0.59)   Depth: 16   00:00:00  7kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kf5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kf5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kf5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kf5
  =/+  (-0.59)   Depth: 17   00:00:00  9kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kf5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kf5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kf5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kf5
  =/+  (-0.59)   Depth: 18   00:00:01  12kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kf5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kf5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kf5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kf5
  =/+  (-0.59)   Depth: 19   00:00:01  18kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kf5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kf5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kf5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kf5
  =/+  (-0.59)   Depth: 20   00:00:01  25kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kf5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kf5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kf5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kf5
  =/+  (-0.59)   Depth: 21   00:00:01  40kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kf5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kf5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kf5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kf5
  =/+  (-0.59)   Depth: 22   00:00:01  64kN
40.e4 Kxe4 41.Ke2 Kd5 42.Kf2 Kd6 43.Ke2 Ke6 44.Ke3 Ke5 45.Ke2 Ke4 46.Kd2 Kd5 47.Ke3 Ke5 48.Ke2 Ke4 49.Kd2 Kd5 50.Ke3 Ke5 51.Ke2 Ke4 52.Kd2 Kd5 53.Ke3 Ke5 54.Ke2 Ke4 55.Kd2 Kd5
  =/+  (-0.59)   Depth: 23   00:00:03  112kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Qxc3+ 52.Kb1 Qe1+ 53.Kc2 Qe4+ 54.Kc1 Kg2 55.Qd1 Qe3+
  -+  (-1.76)   Depth: 24   00:00:25  1469kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Qxc3+ 52.Kb1 Qe1+ 53.Kc2 Qe4+ 54.Kc1 Kg2 55.Qd1 Qe3+
  -+  (-1.80)   Depth: 25   00:00:33  2082kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Qb4 Kxg3 53.Qb8+ Kg2 54.Qe5 g3 55.a4 Kf3
  -+  (-2.58)   Depth: 26   00:01:12  4568kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Qb4 Kxg3 53.Qb8+ Kg2 54.Qe5 g3 55.a4 Kf3
  -+  (-2.84)   Depth: 27   00:01:57  7694kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Qb4 Kxg3 53.Qb8+ Kg2 54.Qe5 g3 55.a4 Kf3
  -+  (-3.24)   Depth: 28   00:03:17  13110kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Kb2 Kxg3 53.Qe8 Qd2+ 54.Kb1 Qd1+ 55.Kb2 Qb3+
  -+  (-3.24)   Depth: 29   00:04:08  16634kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Kb2 Kxg3 53.Qb4 Kf2 54.a4 g3 55.Qb6+ Ke2
  -+  (-3.25)   Depth: 30   00:06:11  26113kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Qb5 Kxg3 53.Qg5 Qxc3+ 54.Kb1 Qd3+ 55.Kb2 Qd4+
  -+  (-4.21)   Depth: 31   00:13:36  54851kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Qa5 Kxg3 53.Qg5 Qxc3+ 54.Kb1 Qd3+ 55.Kb2 Qd4+
  -+  (-4.28)   Depth: 32   00:21:57  89275kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Qe8 Qxc3+ 53.Kd1 Qd3+ 54.Kc1 Qxa3+ 55.Kb1 Qd3+
  -+  (-4.28)   Depth: 33   00:29:10  115mN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Qe8 Qxc3+ 53.Kb1 Qd3+ 54.Kc1 Qxa3+ 55.Kb1 Qd3+
  -+  (-4.28)   Depth: 34   00:38:45  152mN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qc5+ Kf1 49.Qxb5 Qe4 50.Qxa4 Qd3+ 51.Kc1 Kg2 52.Qe8 Qxc3+ 53.Kb1 Qd3+ 54.Kc1 Qxa3+ 55.Kb1 Qd3+
  -+  (-4.28)   Depth: 35   00:53:26  208mN

(, AMD 14.01.2008)

Analysis by Fritz 11:

40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Ke5 44.Ke3 Kd5 45.Kd2
  -+  (-1.63)   Depth: 9/12   00:00:00  2kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd5 44.Kd2 Ke5 45.Ke3 Kd5 46.Kd2
  -+  (-1.63)   Depth: 10/14   00:00:00  3kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd5 44.Kd2 Ke5 45.Ke3 Kd5 46.Kd2
  -+  (-1.63)   Depth: 11/14   00:00:00  4kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd5 44.Kd2 Ke5 45.Ke3 Kd5 46.Kd2
  -+  (-1.63)   Depth: 12/15   00:00:00  6kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd5 44.Kd2 Kd6 45.Ke3 Ke5 46.Kd2 Ke6
  -+  (-1.63)   Depth: 13/15   00:00:00  8kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd5 44.Kd2 Kd6 45.Ke3 Ke5 46.Kd2 Ke6
  -+  (-1.63)   Depth: 14/24   00:00:00  13kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd5 44.Kd2 Kd6 45.Ke3 Ke5 46.Kd2 Ke6
  -+  (-1.63)   Depth: 15/23   00:00:00  19kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Ke5 44.Ke3 Kd5 45.Kd2 Kd6 46.Ke3 Ke5 47.Kd2 Ke6
  -+  (-1.63)   Depth: 16/26   00:00:00  28kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Ke5 44.Ke3 Kd5 45.Kd2 Kd6 46.Ke3 Ke5 47.Kd2 Ke6
  -+  (-1.63)   Depth: 17/28   00:00:00  51kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd6 44.Kd2 Kd5 45.Ke3 Ke5 46.Kd2 Ke6
  -+  (-1.63)   Depth: 18/26   00:00:00  77kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd6 44.Kd2 Kd5 45.Ke3 Ke5 46.Kd2 Ke6
  -+  (-1.63)   Depth: 19/30   00:00:00  105kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd6 44.Kd2 Kd5 45.Ke3 Kc5 46.Ke2 Kc6 47.Kd2 Kd6 48.Ke3 Ke5 49.Kd2 Ke6
  -+  (-1.63)   Depth: 20/30   00:00:00  158kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd6 44.Kd2 Ke5 45.Ke3 Kd5 46.Kd2 Kd6 47.Ke3 Ke5 48.Kd2 Ke6
  -+  (-1.63)   Depth: 21/33   00:00:00  235kN
40.e4 Kxe4 41.Ke2 Kf5 42.Kd2 Ke6 43.Ke2 Kd6 44.Kd2 Ke5 45.Ke3 Kd5 46.Kd2 Kd6 47.Ke2 Kd5 48.Kd2
  -/+  (-0.99)   Depth: 22/34   00:00:01  826kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qb6+ Kf1 49.Qxb5 Qg2+ 50.Kc1 Qxg3 51.Qxc4+ Kg2 52.Qe2+ Kh1 53.Qe4+ Qf3 54.Qe1+ Kh2
  -+  (-2.47)   Depth: 23/45   00:00:08  6885kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qb6+ Kf1 49.Qxb5 Qg2+ 50.Kc1 Qxg3 51.Qxc4+ Kg2 52.Qe2+ Kh1 53.Qe4+ Qf3 54.Qxa4 Qxc3+
  -+  (-2.74)   Depth: 24/44   00:00:13  11484kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qb6+ Kf1 49.Qxb5 Qg2+ 50.Kc1 Qxg3 51.Qxc4+ Kg2 52.Qe2+ Kh1 53.Qe4+ Qf3 54.Qxa4 Qxc3+
  -+  (-2.85)   Depth: 25/44   00:00:21  18388kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qb6+ Kf1 49.Qxb5 Qg2+ 50.Kc1 Qxg3 51.Qxc4+ Kg2 52.Qd4 Qe1+ 53.Kc2 g3 54.Qxa4 Kf3
  -+  (-2.93)   Depth: 26/47   00:00:37  31461kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qb6+ Kf1 49.Qxb5 Qg2+ 50.Kc1 Qxg3 51.Qxc4+ Kg2 52.Qe2+ Kh1 53.Qc4 Qe1+ 54.Kc2 g3
  -+  (-3.14)   Depth: 27/51   00:01:09  60051kN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qb6+ Kf1 49.Qxb5 Qg2+ 50.Kc1 Qxg3 51.Qxc4+ Kg2 52.Qe2+ Kh1 53.Qc4 Qe1+ 54.Kc2 g3
  -+  (-3.20)   Depth: 28/50   00:01:55  103mN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qb6+ Kf1 49.Qxb5 Qg2+ 50.Kc1 Qxg3 51.Qxc4+ Kg2 52.Qe2+ Kh1 53.Qc4 Qe1+ 54.Kc2 g3
  -+  (-3.20)   Depth: 29/52   00:02:33  138mN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qb6+ Kf1 49.Qxb5 Qg2+ 50.Kc1 Qxg3 51.Qxc4+ Kg2 52.Qe2+ Kh1 53.Qc4 Qe1+ 54.Kc2 g3
  -+  (-3.31)   Depth: 30/55   00:05:24  298mN
40.e4 Kg2 41.e5 Kxh2 42.e6 Kg1 43.e7 h2 44.e8Q h1Q 45.Qe3+ Kh2 46.Qh6+ Kg2 47.Qc6+ Kg1 48.Qb6+ Kf1 49.Qxb5 Qg2+ 50.Kc1 Qxg3 51.Qxc4+ Kf2 52.Kb1 Qf3 53.Qxa4 g3 54.Qd4+ Ke2
  -+  (-3.42)   Depth: 31/57   00:11:09  611mN

(, AMD 13.01.2008)

Regards,
Gaмßito
Parent - - By Vasik Rajlich (Silver) Date 2008-01-14 21:44
The searches are so selective now that a nominal depth value really has no meaning. In a 20-ply search, you might get 60 move variations and 6-move variations.

Re. diminishing returns, people have been predicting them for a long time. At some point, the prediction will be right - maybe in 2020 or so.

Vas
Parent - - By Sesse (****) Date 2008-01-14 23:08
By the way, do you have any idea how strong an engine would be without search altogether? Ie., just try all possible moves and try the one with the largest static evaluation.

Actually, it annoys me a bit when I step through my own games with Rybka. Suddenly the evaluation jumps a lot due to some trap ten moves ahead in time which neither side would be anywhere near seeing in any case, so even though one move is probably objectively “better” I would never have been able to convert that advantage into anything real. Perhaps I should plot the graph of evaluation over time; if it's got a sudden rise or drop at depth 10 or 11 it's probably due to some subtlety which I can safely ignore at my level...

/* Steinar */
Parent - - By lkaufman (*****) Date 2008-01-14 23:20
     With no search at all, an engine would play at about the level of a typical 8 year old, something like 500 rating or less, because it would regularly put pieces where they would be captured. You have to give a program a capture search (or something equivalent) to have it play at all decently. With capture search, a typical program like Fritz doing a 1 ply search plays somewhere in the ballpark of 1200 level chess.
     As for the Rybka eval being too subtle for your level, just set Rybka for fixed depth searches, maybe 5 ply or so (depending on your own level), to get evals that are more relevant to what you could hope to see in your own games if at your best.
Parent - - By Sesse (****) Date 2008-01-14 23:33
Doesn't Rybka have a static exchange evaluator as part of its evaluation function?

/* Steinar */
Parent - - By lkaufman (*****) Date 2008-01-15 00:01
Without commenting on Rybka in this respect (I'll leave that to Vas if he wishes), most modern programs (as far as I know) use capture search (or capture +check) rather than static exchange evaluators. As Uri pointed out elsewhere, there is little real difference between a capture search or simulating one in the evaluation.
Parent - By Vasik Rajlich (Silver) Date 2008-01-18 13:59
Right - Rybka handles captures in the search.

How strong would Rybka play without any search if we built in an SEE into the eval? That would depend on how good this SEE is, much more than on what the eval understands positinally.

Vas
Parent - - By Roland Rösler (****) Date 2008-01-15 01:05
... maybe in 2020 or so.
Oh, I see, it´s the time, when you will retire. The scene gets boring, because you have shown chess is drawish?  No, the party will become crazy!
For me, dimishing returns are only the consequences of selectivity and / or bad (static!) evaluation. I see no other reasons. Okay, there are some programmers techniques, which are good for the engine strength but bad for a clear look at this issue. But why should be two consecutive 20 depth searches (at position 20 and 30) be better than one 40 depth searching (at position 20; we speak about 2020!)? Let us look to the possible reasons:
1. Selectivity
I call it selectivity (Uri and Vas don´t), when the programm don´t show the best moves at the given depth (give her the best moves, she will accept at given depth). And the chance, that the fault isn´t in the first or second ply in a 20 depth search is higher than the fault is in ply 19 or 20. What´s the reason for this? Shown depth is a blunder! Nobody knows the real depth for each ply.
2. Bad (static) eval
There are many position in chess, where search isn´t enough. (Okay, against humans, search is enough in 99,9%, because chess is a tactical game and not so much strategic as human think.) But there are so many endgames, where Rybka thinks it´s +3 or +5 and they are clear draws. And when she has this eval, you can´t expect, that she will avoid this endgame (today I see Radjabov vs. van Wely and she shows <-3 and it was a clear draw). And then we have the fortresses in the middlegames and no engine have a clue. Here we have a lot of dimishing returns (depth 20 is equal to depth 80).
Parent - - By Uri Blass (*****) Date 2008-01-15 20:00
1)I disagree that bad evaluation means diminishing returns.
It is the opposite
If you have a good evaluation then you usually find the best move at small depth so more depth is not going to help you.

The only reason that there is positive returns is the fact that the evaluation is not perfect.

2)My opinion is that diminishing returns is something that happens today.
the rating advantage that programs can get from 64 bit relative to 32 bit is in most cases bigger at 40/4 relative to 40/40

Here is an example about zappa:

ccrl blitz
 
Zappa Mexico 64-bit 2905 +15 −15 53.9% −25.3 34.8% 1580 +56 elo
Zappa Mexico 32-bit 2849 +20 −20 52.1% −14.4 37.2% 820

Zappa 1.1 64-bit 2713 +15 −15 46.8% +21.1 32.7% 1472 +75 elo
100.0%
Zappa 1.1 32-bit 2638 +25 −25 43.2% +48.4 29.4% 575

ccrl 40/40

Zappa Mexico 64-bit 2934 +21 −21 56.9% −44.1 44.3% 729 +46 elo<+56
Zappa Mexico 32-bit 2888 +20 −20 54.9% −29.9 42.7% 813

Zappa 1.1 64-bit 2734 +24 −24 45.3% +33.0 37.8% 545  +47 elo<+75
98.8%
Zappa 1.1 32-bit 2687 +32 −32 48.9% +8.6 32.6% 322

Uri
Parent - By Banned for Life (Gold) Date 2008-01-15 20:30
Your second point is consistent with playing strength (when comparing an engine with itself) being correlated with difference in number of plies performed (which will be larger at shorter time control games) rather than the ratio of the number of nodes calculated (which should be fairly constant for shorter and longer games). When looked at in this context, this is not a surprising result.

Regards,
Alan
Parent - By Vasik Rajlich (Silver) Date 2008-01-18 14:03
Even if you did a full minimax search, with no selectivity at all, you'd still get dimishing returns comparable to what we see today.

There are some evaluation shortcomings which can be overcome with a little bit of extra searching, and some shortcomings which would need a ton of extra searching.

Vas

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill