Poll
What will be Rybka 3 elo point increase over 2.3.2a on single 32-bit processor? (Closed)
| <15 elo points | 8 | 6% | |
| 15-25 | 11 | 9% | |
| 25-35 | 19 | 15% | |
| 35-45 | 13 | 10% | |
| 45-55 | 27 | 22% | |
| 55-65 | 14 | 11% | |
| 65-75 | 12 | 10% | |
| 75-85 | 4 | 3% | |
| 85-95 | 2 | 2% | |
| >95 elo points | 15 | 12% |
I have heard this -3 figure from other sources also.
I think Uri also did tests to prove this from positions - perhaps he can repeat his analysis here?
I think Uri also did tests to prove this from positions - perhaps he can repeat his analysis here?
![[de]](/mwf/v12/flags/de.png "Germany")
I think I'll try this later. Although some people seem to claim that 1 ply in Rybka is more like 4 ply, and one node is a lot more than a node (is that correct?), but that should still be okay. My rating is quite close to those 2380. I'm still concerned about a) one-move blunders, which are inevitable for me in bullet (but that's my problem, I guess)
b) me losing on time. Maybe 1+1 in Fischerrandom would also be okay? (Even though it increases my chances to blunder dramatically before move 10.)
b) me losing on time. Maybe 1+1 in Fischerrandom would also be okay? (Even though it increases my chances to blunder dramatically before move 10.)
I think that the node count of Rybka may be accurate enough on a 1 ply search, but at normal levels I can believe the claim that adding an extra zero gives a number that is roughly comparable to the way other engines report nodes. I think 1'+1" is fair enough, as Rybka will also miss some very simple tactics at 1 ply, and will play endings very poorly. I think you will be handicapping yourself if you play Fischerrandom at this speed; I cannot imagine being able to play this version at this speed without making blunders, as you are just unfamiliar with the start positions. If you want to minimize the effect of book, just take white every game and open 1a3 and 1h3 alternately.
When I tried this, I realized that I cannot set the Fritz family interfaces for fixed depth for the engine and time limit for me (at least I don't see how to do this). I tried playing two games against depth 1 moving almost instantly, although I exceeded the 1'+1" limit by a bit in one game, and scored 1/2 out of 2.
Same problem for me. But I have a second problem: In the middle of the game, the engine starts to search much deeper, approximately 12 plies or so. Not in every move, but regularly. Could anyone tell me what I did wrong? My GUI is Shredder 9, and I just set 'fixed depth' to '1'.
I'm not sure, but turning off thinking on opponent's time (ponder) should solve this problem, if you haven't already done this.
Of course that was the problem, shame on me! It nevertheless cost me a considerable amount of time to find the 'ponder off' button (I usually use Rybka exclusively for analysis). I played 20 games against depth = 1 and scored a mere +1 (I even have to admit that I had planned 10 games, but as the result was -1, I had to play ten more :))
I don't post the games because I don't want to produce too much spam - the games are really lousy, unsurprisingly - but if anyone is nevertheless interested, I can make this up. Time control: I tried to spend less than 2 minutes for the first 60 moves and then allowed myself an (imaginary) increment of 1 second per move. I was usually a bit too fast, and lost only one game on time (which I realized only after the game which I had 'won'). It was amazingly tough and I must say that I only won after I changed the openings (to 1. f4 with white and 'offbeat' French with black). I won virtually every endgame; but what surprised me most was that in general, I was tactically worse. Partly for that reason, I played a match Rybka 2.32a vs. Toga II 3.12 SE at depth = 1 over 40 games. Toga scored only two draws. In one of the games, Rybka announced a mate in 6! I don't think this result can be explained solely by Rybka's superior evaluation!?
I don't post the games because I don't want to produce too much spam - the games are really lousy, unsurprisingly - but if anyone is nevertheless interested, I can make this up. Time control: I tried to spend less than 2 minutes for the first 60 moves and then allowed myself an (imaginary) increment of 1 second per move. I was usually a bit too fast, and lost only one game on time (which I realized only after the game which I had 'won'). It was amazingly tough and I must say that I only won after I changed the openings (to 1. f4 with white and 'offbeat' French with black). I won virtually every endgame; but what surprised me most was that in general, I was tactically worse. Partly for that reason, I played a match Rybka 2.32a vs. Toga II 3.12 SE at depth = 1 over 40 games. Toga scored only two draws. In one of the games, Rybka announced a mate in 6! I don't think this result can be explained solely by Rybka's superior evaluation!?
![[il]](/mwf/v12/flags/il.png "Israel")
Of course the result is not by rybka's superior's evaluation and Osipov found that you need to substract 3 from the depth of rybka so rybka practically get depth 4 and not depth 1 when she claims depth 1.
I tried some games at fixed depth between Rybka and Toga1.4 beta5 c(Rybka depth 15 against Toga depth 17) and the result was +5 =4 -4 for rybka
I had to adjudicate the last game as win for rybka because the sides needed many hours for playing in mate positions
[Event "URI-PC, 15ply/17ply"]
[Site "URI-PC"]
[Date "2008.02.09"]
[Round "13"]
[White "Rybka 2.3.2a 32-bit"]
[Black "Toga II 1.4 beta5c"]
[Result "1-0"]
[ECO "B22"]
[PlyCount "156"]
{W=15.6 ply; 102kN/s
B=17.6 ply; 312kN/s
} 1. e4 c5 2. c3 d5 3. exd5 Qxd5 4.
d4 Nf6 5. Nf3 Bg4 6. Be2 e6 7. O-O Nc6 8. h3 Bh5 9. Be3 cxd4 10. cxd4 Be7 11.
Nc3 Qd6 12. Qb3 {Both last book move [%eval 10,15] [%emt 0:04:58]} O-O {
[%eval 10,17] [%emt 0:02:38]} 13. Rac1 {(Rfd1) [%eval 12,15] [%emt 0:01:57]} a6
{(Rfd8) [%eval 9,17] [%emt 0:04:10]} 14. Rfd1 {[%eval 13,15] [%emt 0:02:27]} b5
{[%eval 4,17] [%emt 0:01:51]} 15. a3 {(d5) [%eval 13,15] [%emt 0:01:31]} Na5 {
[%eval 5,17] [%emt 0:02:15]} 16. Qa2 {[%eval 14,15] [%emt 0:00:24]} Nc4 {
(Nd5) [%eval -6,17] [%emt 0:01:35]} 17. Bxc4 {
(Bg5) [%eval 13,15] [%emt 0:00:36]} Bxf3 {[%eval -3,17] [%emt 0:01:23]} 18.
gxf3 {[%eval 18,15] [%emt 0:00:15]} bxc4 {[%eval -3,17] [%emt 0:02:16]} 19.
Qxc4 {[%eval 20,15] [%emt 0:00:16]} Rfb8 {(Rfc8) [%eval 0,17] [%emt 0:03:30]}
20. Rd2 {(Rc2) [%eval 36,15] [%emt 0:00:50]} Rc8 {[%eval 12,17] [%emt 0:02:38]}
21. Qd3 {[%eval 32,15] [%emt 0:00:38]} Qd7 {(Qb8) [%eval 12,17] [%emt 0:02:43]}
22. Rdc2 {[%eval 33,15] [%emt 0:01:10]} Rd8 {[%eval 11,17] [%emt 0:03:12]} 23.
Kg2 {(Kh2) [%eval 29,15] [%emt 0:02:48]} Qb7 {(h6) [%eval 14,17] [%emt 0:05:11]
} 24. b4 {[%eval 30,15] [%emt 0:01:47]} a5 {(h6) [%eval -5,17] [%emt 0:02:50]}
25. Ne4 {[%eval 30,15] [%emt 0:00:46]} Nd5 {[%eval 0,17] [%emt 0:03:05]} 26. b5
{[%eval 27,15] [%emt 0:00:39]} e5 {[%eval 0,17] [%emt 0:02:57]} 27. Rc6 {
(Bd2) [%eval 27,15] [%emt 0:01:34]} exd4 {[%eval 10,17] [%emt 0:05:23]} 28. Bd2
{[%eval 30,15] [%emt 0:01:44]} a4 {(h6) [%eval 4,17] [%emt 0:05:28]} 29. Nc5 {
[%eval 36,15] [%emt 0:00:56]} Qb8 {[%eval 0,17] [%emt 0:01:24]} 30. Re1 {
[%eval 34,15] [%emt 0:01:52]} Nf4+ {[%eval 1,17] [%emt 0:03:38]} 31. Bxf4 {
[%eval 39,15] [%emt 0:00:13]} Qxf4 {[%eval 1,17] [%emt 0:01:55]} 32. Ne4 {
[%eval 39,15] [%emt 0:00:28]} Bf8 {(Rab8) [%eval 0,17] [%emt 0:04:20]} 33. b6 {
[%eval 56,15] [%emt 0:01:41]} Rab8 {[%eval 30,17] [%emt 0:03:45]} 34. Rb1 {
[%eval 51,15] [%emt 0:00:35]} Bd6 {[%eval 31,17] [%emt 0:03:57]} 35. Nxd6 {
[%eval 60,15] [%emt 0:00:48]} Rxd6 {[%eval 34,17] [%emt 0:01:05]} 36. Rxd6 {
[%eval 62,15] [%emt 0:00:02]} Qxd6 {[%eval 35,17] [%emt 0:00:33]} 37. b7 {
(Rb4) [%eval 64,15] [%emt 0:00:33]} h5 {(g6) [%eval 37,17] [%emt 0:01:43]} 38.
Rb5 {(Rb4) [%eval 102,15] [%emt 0:00:24]} Qa6 {
(g6) [%eval 65,17] [%emt 0:00:40]} 39. Qc4 {[%eval 133,15] [%emt 0:00:14]} Qg6+
{[%eval 55,17] [%emt 0:00:32]} 40. Kf1 {[%eval 122,15] [%emt 0:00:02]} Qa6 {
[%eval 90,17] [%emt 0:00:51]} 41. Ke2 {(Qc5) [%eval 131,15] [%emt 0:00:27]} g6
{[%eval 72,17] [%emt 0:04:15]} 42. Kd3 {(f4) [%eval 101,15] [%emt 0:00:12]} Qd6
{[%eval 66,17] [%emt 0:02:21]} 43. Qc8+ {[%eval 93,15] [%emt 0:00:42]} Kh7 {
[%eval 57,17] [%emt 0:02:21]} 44. Rb6 {[%eval 132,15] [%emt 0:00:57]} Qxa3+ {
(Qf4) [%eval 87,17] [%emt 0:10:30]} 45. Kxd4 {[%eval 137,15] [%emt 0:00:06]}
Qxf3 {[%eval 87,17] [%emt 0:01:11]} 46. Qxb8 {[%eval 137,15] [%emt 0:00:03]}
Qxf2+ {[%eval 88,17] [%emt 0:04:07]} 47. Kc4 {[%eval 137,15] [%emt 0:00:01]}
Qxb6 {[%eval 87,17] [%emt 0:02:31]} 48. Qc8 {[%eval 137,15] [%emt 0:00:01]}
Qb3+ {(Qa6+) [%eval 87,17] [%emt 0:01:44]} 49. Kc5 {
[%eval 170,15] [%emt 0:00:02]} Qe3+ {(Qc3+) [%eval 160,17] [%emt 2:50:00]} 50.
Kb5 {[%eval 197,15] [%emt 0:00:01]} Qb3+ {[%eval 160,17] [%emt 0:03:32]} 51.
Ka5 {[%eval 197,15] [%emt 0:00:00]} Qd5+ {[%eval 160,17] [%emt 0:02:08]} 52.
Kxa4 {[%eval 197,15] [%emt 0:00:00]} Qd4+ {[%eval 160,17] [%emt 0:00:16]} 53.
Kb5 {[%eval 205,15] [%emt 0:00:01]} Qd5+ {[%eval 160,17] [%emt 0:00:49]} 54.
Kb6 {[%eval 216,15] [%emt 0:00:01]} Qb3+ {(Qd6+) [%eval 168,17] [%emt 0:01:54]}
55. Kc7 {[%eval 216,15] [%emt 0:00:01]} Qg3+ {[%eval 171,17] [%emt 0:00:32]}
56. Kd8 {[%eval 216,15] [%emt 0:00:00]} Qh4+ {
(Qg5+) [%eval 169,17] [%emt 0:04:23]} 57. Ke8 {[%eval 216,15] [%emt 0:00:01]}
Qe4+ {(Qe1+) [%eval 183,17] [%emt 0:00:42]} 58. Kxf7 {
[%eval 216,15] [%emt 0:00:01]} Qd5+ {(Qf3+) [%eval 182,17] [%emt 0:00:01]} 59.
Kf8 {(Ke7) [%eval 216,15] [%emt 0:00:00]} Qd6+ {[%eval 183,17] [%emt 0:00:07]}
60. Ke8 {[%eval 216,15] [%emt 0:00:03]} Qe5+ {[%eval 184,17] [%emt 0:00:30]}
61. Kd7 {(Kd8) [%eval 247,15] [%emt 0:00:02]} Qd5+ {
(Qd4+) [%eval 189,17] [%emt 0:00:59]} 62. Kc7 {[%eval 247,15] [%emt 0:00:01]}
Qe5+ {[%eval 201,17] [%emt 0:01:57]} 63. Kb6 {[%eval 246,15] [%emt 0:00:01]}
Qb2+ {[%eval 201,17] [%emt 0:00:23]} 64. Ka7 {[%eval 247,15] [%emt 0:00:05]}
Qd4+ {(Qa3+) [%eval 356,17] [%emt 0:03:39]} 65. Ka8 {
[%eval 258,15] [%emt 0:00:01]} Qa4+ {[%eval 686,17] [%emt 0:46:38]} 66. Kb8 {
[%eval 258,15] [%emt 0:00:00]} Qb4 {[%eval 686,17] [%emt 0:01:34]} 67. Qd8 {
[%eval 258,15] [%emt 0:00:01]} Kh6 {[%eval 835,17] [%emt 0:04:03]} 68. Qh8+ {
[%eval 988,15] [%emt 0:07:44]} Kg5 {[%eval 206,4] [%emt 0:00:00]} 69. Qe5+ {
[%eval 988,15] [%emt 0:05:22]} Kh6 {[%eval 856,17] [%emt 0:01:00]} 70. Kc7 {
(h4) [%eval 988,15] [%emt 0:04:18]} Qc4+ {[%eval 1001,17] [%emt 0:04:11]} 71.
Kd7 {[%eval 1042,15] [%emt 0:07:47]} Qf7+ {[%eval 1023,17] [%emt 0:04:54]} 72.
Qe7 {[%eval 1097,15] [%emt 0:06:29]} Qb3 {(Qf4) [%eval 1138,17] [%emt 0:24:05]}
73. Qf8+ {[%eval 1142,15] [%emt 0:10:09]} Kg5 {[%eval 1118,17] [%emt 0:17:36]}
74. Qd8+ {[%eval 1163,15] [%emt 0:23:37]} Kf5 {[%eval 2004,17] [%emt 0:25:30]}
75. b8=Q {[%eval 1210,15] [%emt 1:19:01]} Qe6+ {[%eval 1991,17] [%emt 0:28:30]}
76. Kc7 {[%eval 1032,8] [%emt 0:00:04]} Qc4+ {[%eval 2002,17] [%emt 0:19:36]}
77. Kb6 {(Kb7) [%eval 1232,15] [%emt 1:34:18]} Qb3+ {
[%eval 32748,17] [%emt 9:10:22]} 78. Ka6 {[%eval 1234,15] [%emt 6:37:36]} Qe6+
{adjud. [%eval 32750,17] [%emt 1:40:33]} 1-0
I tried some games at fixed depth between Rybka and Toga1.4 beta5 c(Rybka depth 15 against Toga depth 17) and the result was +5 =4 -4 for rybka
I had to adjudicate the last game as win for rybka because the sides needed many hours for playing in mate positions
[Event "URI-PC, 15ply/17ply"]
[Site "URI-PC"]
[Date "2008.02.09"]
[Round "13"]
[White "Rybka 2.3.2a 32-bit"]
[Black "Toga II 1.4 beta5c"]
[Result "1-0"]
[ECO "B22"]
[PlyCount "156"]
{W=15.6 ply; 102kN/s
B=17.6 ply; 312kN/s
} 1. e4 c5 2. c3 d5 3. exd5 Qxd5 4.
d4 Nf6 5. Nf3 Bg4 6. Be2 e6 7. O-O Nc6 8. h3 Bh5 9. Be3 cxd4 10. cxd4 Be7 11.
Nc3 Qd6 12. Qb3 {Both last book move [%eval 10,15] [%emt 0:04:58]} O-O {
[%eval 10,17] [%emt 0:02:38]} 13. Rac1 {(Rfd1) [%eval 12,15] [%emt 0:01:57]} a6
{(Rfd8) [%eval 9,17] [%emt 0:04:10]} 14. Rfd1 {[%eval 13,15] [%emt 0:02:27]} b5
{[%eval 4,17] [%emt 0:01:51]} 15. a3 {(d5) [%eval 13,15] [%emt 0:01:31]} Na5 {
[%eval 5,17] [%emt 0:02:15]} 16. Qa2 {[%eval 14,15] [%emt 0:00:24]} Nc4 {
(Nd5) [%eval -6,17] [%emt 0:01:35]} 17. Bxc4 {
(Bg5) [%eval 13,15] [%emt 0:00:36]} Bxf3 {[%eval -3,17] [%emt 0:01:23]} 18.
gxf3 {[%eval 18,15] [%emt 0:00:15]} bxc4 {[%eval -3,17] [%emt 0:02:16]} 19.
Qxc4 {[%eval 20,15] [%emt 0:00:16]} Rfb8 {(Rfc8) [%eval 0,17] [%emt 0:03:30]}
20. Rd2 {(Rc2) [%eval 36,15] [%emt 0:00:50]} Rc8 {[%eval 12,17] [%emt 0:02:38]}
21. Qd3 {[%eval 32,15] [%emt 0:00:38]} Qd7 {(Qb8) [%eval 12,17] [%emt 0:02:43]}
22. Rdc2 {[%eval 33,15] [%emt 0:01:10]} Rd8 {[%eval 11,17] [%emt 0:03:12]} 23.
Kg2 {(Kh2) [%eval 29,15] [%emt 0:02:48]} Qb7 {(h6) [%eval 14,17] [%emt 0:05:11]
} 24. b4 {[%eval 30,15] [%emt 0:01:47]} a5 {(h6) [%eval -5,17] [%emt 0:02:50]}
25. Ne4 {[%eval 30,15] [%emt 0:00:46]} Nd5 {[%eval 0,17] [%emt 0:03:05]} 26. b5
{[%eval 27,15] [%emt 0:00:39]} e5 {[%eval 0,17] [%emt 0:02:57]} 27. Rc6 {
(Bd2) [%eval 27,15] [%emt 0:01:34]} exd4 {[%eval 10,17] [%emt 0:05:23]} 28. Bd2
{[%eval 30,15] [%emt 0:01:44]} a4 {(h6) [%eval 4,17] [%emt 0:05:28]} 29. Nc5 {
[%eval 36,15] [%emt 0:00:56]} Qb8 {[%eval 0,17] [%emt 0:01:24]} 30. Re1 {
[%eval 34,15] [%emt 0:01:52]} Nf4+ {[%eval 1,17] [%emt 0:03:38]} 31. Bxf4 {
[%eval 39,15] [%emt 0:00:13]} Qxf4 {[%eval 1,17] [%emt 0:01:55]} 32. Ne4 {
[%eval 39,15] [%emt 0:00:28]} Bf8 {(Rab8) [%eval 0,17] [%emt 0:04:20]} 33. b6 {
[%eval 56,15] [%emt 0:01:41]} Rab8 {[%eval 30,17] [%emt 0:03:45]} 34. Rb1 {
[%eval 51,15] [%emt 0:00:35]} Bd6 {[%eval 31,17] [%emt 0:03:57]} 35. Nxd6 {
[%eval 60,15] [%emt 0:00:48]} Rxd6 {[%eval 34,17] [%emt 0:01:05]} 36. Rxd6 {
[%eval 62,15] [%emt 0:00:02]} Qxd6 {[%eval 35,17] [%emt 0:00:33]} 37. b7 {
(Rb4) [%eval 64,15] [%emt 0:00:33]} h5 {(g6) [%eval 37,17] [%emt 0:01:43]} 38.
Rb5 {(Rb4) [%eval 102,15] [%emt 0:00:24]} Qa6 {
(g6) [%eval 65,17] [%emt 0:00:40]} 39. Qc4 {[%eval 133,15] [%emt 0:00:14]} Qg6+
{[%eval 55,17] [%emt 0:00:32]} 40. Kf1 {[%eval 122,15] [%emt 0:00:02]} Qa6 {
[%eval 90,17] [%emt 0:00:51]} 41. Ke2 {(Qc5) [%eval 131,15] [%emt 0:00:27]} g6
{[%eval 72,17] [%emt 0:04:15]} 42. Kd3 {(f4) [%eval 101,15] [%emt 0:00:12]} Qd6
{[%eval 66,17] [%emt 0:02:21]} 43. Qc8+ {[%eval 93,15] [%emt 0:00:42]} Kh7 {
[%eval 57,17] [%emt 0:02:21]} 44. Rb6 {[%eval 132,15] [%emt 0:00:57]} Qxa3+ {
(Qf4) [%eval 87,17] [%emt 0:10:30]} 45. Kxd4 {[%eval 137,15] [%emt 0:00:06]}
Qxf3 {[%eval 87,17] [%emt 0:01:11]} 46. Qxb8 {[%eval 137,15] [%emt 0:00:03]}
Qxf2+ {[%eval 88,17] [%emt 0:04:07]} 47. Kc4 {[%eval 137,15] [%emt 0:00:01]}
Qxb6 {[%eval 87,17] [%emt 0:02:31]} 48. Qc8 {[%eval 137,15] [%emt 0:00:01]}
Qb3+ {(Qa6+) [%eval 87,17] [%emt 0:01:44]} 49. Kc5 {
[%eval 170,15] [%emt 0:00:02]} Qe3+ {(Qc3+) [%eval 160,17] [%emt 2:50:00]} 50.
Kb5 {[%eval 197,15] [%emt 0:00:01]} Qb3+ {[%eval 160,17] [%emt 0:03:32]} 51.
Ka5 {[%eval 197,15] [%emt 0:00:00]} Qd5+ {[%eval 160,17] [%emt 0:02:08]} 52.
Kxa4 {[%eval 197,15] [%emt 0:00:00]} Qd4+ {[%eval 160,17] [%emt 0:00:16]} 53.
Kb5 {[%eval 205,15] [%emt 0:00:01]} Qd5+ {[%eval 160,17] [%emt 0:00:49]} 54.
Kb6 {[%eval 216,15] [%emt 0:00:01]} Qb3+ {(Qd6+) [%eval 168,17] [%emt 0:01:54]}
55. Kc7 {[%eval 216,15] [%emt 0:00:01]} Qg3+ {[%eval 171,17] [%emt 0:00:32]}
56. Kd8 {[%eval 216,15] [%emt 0:00:00]} Qh4+ {
(Qg5+) [%eval 169,17] [%emt 0:04:23]} 57. Ke8 {[%eval 216,15] [%emt 0:00:01]}
Qe4+ {(Qe1+) [%eval 183,17] [%emt 0:00:42]} 58. Kxf7 {
[%eval 216,15] [%emt 0:00:01]} Qd5+ {(Qf3+) [%eval 182,17] [%emt 0:00:01]} 59.
Kf8 {(Ke7) [%eval 216,15] [%emt 0:00:00]} Qd6+ {[%eval 183,17] [%emt 0:00:07]}
60. Ke8 {[%eval 216,15] [%emt 0:00:03]} Qe5+ {[%eval 184,17] [%emt 0:00:30]}
61. Kd7 {(Kd8) [%eval 247,15] [%emt 0:00:02]} Qd5+ {
(Qd4+) [%eval 189,17] [%emt 0:00:59]} 62. Kc7 {[%eval 247,15] [%emt 0:00:01]}
Qe5+ {[%eval 201,17] [%emt 0:01:57]} 63. Kb6 {[%eval 246,15] [%emt 0:00:01]}
Qb2+ {[%eval 201,17] [%emt 0:00:23]} 64. Ka7 {[%eval 247,15] [%emt 0:00:05]}
Qd4+ {(Qa3+) [%eval 356,17] [%emt 0:03:39]} 65. Ka8 {
[%eval 258,15] [%emt 0:00:01]} Qa4+ {[%eval 686,17] [%emt 0:46:38]} 66. Kb8 {
[%eval 258,15] [%emt 0:00:00]} Qb4 {[%eval 686,17] [%emt 0:01:34]} 67. Qd8 {
[%eval 258,15] [%emt 0:00:01]} Kh6 {[%eval 835,17] [%emt 0:04:03]} 68. Qh8+ {
[%eval 988,15] [%emt 0:07:44]} Kg5 {[%eval 206,4] [%emt 0:00:00]} 69. Qe5+ {
[%eval 988,15] [%emt 0:05:22]} Kh6 {[%eval 856,17] [%emt 0:01:00]} 70. Kc7 {
(h4) [%eval 988,15] [%emt 0:04:18]} Qc4+ {[%eval 1001,17] [%emt 0:04:11]} 71.
Kd7 {[%eval 1042,15] [%emt 0:07:47]} Qf7+ {[%eval 1023,17] [%emt 0:04:54]} 72.
Qe7 {[%eval 1097,15] [%emt 0:06:29]} Qb3 {(Qf4) [%eval 1138,17] [%emt 0:24:05]}
73. Qf8+ {[%eval 1142,15] [%emt 0:10:09]} Kg5 {[%eval 1118,17] [%emt 0:17:36]}
74. Qd8+ {[%eval 1163,15] [%emt 0:23:37]} Kf5 {[%eval 2004,17] [%emt 0:25:30]}
75. b8=Q {[%eval 1210,15] [%emt 1:19:01]} Qe6+ {[%eval 1991,17] [%emt 0:28:30]}
76. Kc7 {[%eval 1032,8] [%emt 0:00:04]} Qc4+ {[%eval 2002,17] [%emt 0:19:36]}
77. Kb6 {(Kb7) [%eval 1232,15] [%emt 1:34:18]} Qb3+ {
[%eval 32748,17] [%emt 9:10:22]} 78. Ka6 {[%eval 1234,15] [%emt 6:37:36]} Qe6+
{adjud. [%eval 32750,17] [%emt 1:40:33]} 1-0
Rybka's 1 ply search is comparable to a 2 or 3 ply search on a standard program like Fritz, but since you were allowed about 2" per move this is not too unfair. Since the current program is now about 90 Elo stronger than yours on short fixed depth, you would probably need about twice the time to win a match from it at "1 ply", so perhaps your search at this longer time limit might be fairly close to Rybka at 1 ply. Since you say you are around 2380, I take this result as confirmation that the eval on Rybka now evaluates at a level not too far away from a 2380 player with similar search, though of course this has no precise meaning.
I must admit that I did not understand why this is supposed to be a reasonable comparison of evaluation from the beginning. 2 seconds are not enough for me to even identify every legal move in a given position (I don't claim that I do this in games played with tournament time control :)). Rybka, on the other hand, most likely calculated several hundred possibilities on each move. So why do you assume that these are 'equal conditions'?
(In fact, this is not a critical question. I don't know anything about this technical stuff, so I'm just curious.)
NB: 90 points is really impressing. But the difference between (me) playing 2 minutes per game or 4 minutes per game should be much bigger!?
(In fact, this is not a critical question. I don't know anything about this technical stuff, so I'm just curious.)
NB: 90 points is really impressing. But the difference between (me) playing 2 minutes per game or 4 minutes per game should be much bigger!?
Computers' search and humans' are very different. At 1 ply, Rybka looks at every legal move, rejects those that are blunders or clearly inferior, and looks at 1 or more of the moves a bit further than that. You probably do pretty much the same thing in 4 seconds, although you don't actually look at each legal move, only at those which seem promising. You both pick the move that gives the best eval (in your respective opinions) after allowing for simple tactics. I don't claim that the conditions are fully equal, maybe Rybka needs to be set to zero ply (you can't do this I know) to make it fully fair, but this would only change the result by around a hundred elo. As for 2 seconds vs 4 seconds, you are probably right that this is worth more than 90 Elo for you, perhaps 3.5 seconds or so would be more accurate for 90 Elo. Anyway, I think this shows at least that Rybka eval (on the new version) is in the master range.
What is the meaning of "at eval"? I guess at least a one ply search is required (to play), because without that, there are no moves at all :-) Or maybe I totally misunderstand something. Maybe, if quiescence and other extensions are not disabled, the effective depth may be much bigger than a small nominal depth? But with one ply only, 2000 seem impossible to me, and even 1500 hardly imaginable.
Do you mean the quality of the evaluation is on 2380 Elo level? Has that been tested, somehow?
Do you mean the quality of the evaluation is on 2380 Elo level? Has that been tested, somehow?
I meant that eval ability is like human eval ability (without search) at 2380 level. But as Uri points out, this is a pretty meaningless statement unless I define my terms clearly. As I explain above, I mean that a 1 ply search might play as well as a human 2380 player moving in one second, which might indeed be around 1500 level in tournament chess. I think that the 1 ply search in the current version might well play at 1500 level or above with the 1500 playing tournament conditions. It's true that a 1 ply search can see a lot of simple tactics, but so can a human master in a second.
I've tried Fritz 11 - Rybka 2.3.2a with 1 ply for both and Rybka wins virtually every game. Since already Fritz 8 on 1 ply wins against Mephisto III on 10s/move (which is almost 1500) I'm not sure the claim that Rybka-1ply is 1500 level in turnament chess would hold...
http://www.xs4all.nl/~tluif/chescom/EngMephIII.html
Martin
http://www.xs4all.nl/~tluif/chescom/EngMephIII.html
Martin
The Mephisto III was so slow, that I think it's blitz wasn't at the level of a 1500's blitz. It calculated ~2 nodes(!) per second. :-D Considering that, it played astonishingly well. The CPU had 8 MHz in the sensor board version. The following game was at tournament time controls:
[Event "Turnierstufe"]
[Site "Wien"]
[Date "1996.??.??"]
[Round "?"]
[White "me"]
[Black "Mephisto III/Lev.6"]
[Result "1-0"]
[ECO "D30"]
[PlyCount "55"]
1. c4 e6 2. d4 d5 3. e3 Bb4+ 4. Bd2 Nc6 5. Nf3 Nf6 6. Qa4 Bxd2+ 7. Nbxd2 Bd7 8.
Qc2 O-O 9. a3 Ng4 10. Bd3 f5 11. cxd5 exd5 12. b4 a5 13. b5 Na7 14. a4 c6 15.
Qb3 f4 16. e4 Qe8 17. O-O Qe6 18. Rfe1 Nc8 19. exd5 Qxe1+ 20. Rxe1 cxb5 21. d6+
Kh8 22. Ng5 bxa4 23. Nf7+ Rxf7 24. Qxf7 Nb6 25. Re7 Rg8 26. Rxd7 Nxd7 27. Qxd7
Nf6 28. Qxb7 1-0
I was lucky that I could start with chess in an era when some comps were so weak, still (although, some were already much stronger in the late 1980s). Today I would probably play poker, instead... so maybe it was actually bad luck! :-D
There was a tuned 12 MHz version which missed a #2 at tournament level (8...Ne8?? looks like it didn't manage 4 plies here):
[Event "4/00-48 CSS 1/85-22"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "Bäurle, Stefan"]
[Black "Mephisto III 12 MHz (Stufe 6)"]
[Result "1-0"]
[ECO "C50"]
[PlyCount "19"]
1. e4 e5 2. Nf3 Nc6 3. Bc4 Be7 4. d3 Nf6 5. Nc3 O-O 6. Ng5 h6 7. h4 hxg5 8.
hxg5 Ne8 9. Qh5 Bb4 10. Qh8# 1-0
[Event "Turnierstufe"]
[Site "Wien"]
[Date "1996.??.??"]
[Round "?"]
[White "me"]
[Black "Mephisto III/Lev.6"]
[Result "1-0"]
[ECO "D30"]
[PlyCount "55"]
1. c4 e6 2. d4 d5 3. e3 Bb4+ 4. Bd2 Nc6 5. Nf3 Nf6 6. Qa4 Bxd2+ 7. Nbxd2 Bd7 8.
Qc2 O-O 9. a3 Ng4 10. Bd3 f5 11. cxd5 exd5 12. b4 a5 13. b5 Na7 14. a4 c6 15.
Qb3 f4 16. e4 Qe8 17. O-O Qe6 18. Rfe1 Nc8 19. exd5 Qxe1+ 20. Rxe1 cxb5 21. d6+
Kh8 22. Ng5 bxa4 23. Nf7+ Rxf7 24. Qxf7 Nb6 25. Re7 Rg8 26. Rxd7 Nxd7 27. Qxd7
Nf6 28. Qxb7 1-0
I was lucky that I could start with chess in an era when some comps were so weak, still (although, some were already much stronger in the late 1980s). Today I would probably play poker, instead... so maybe it was actually bad luck! :-D
There was a tuned 12 MHz version which missed a #2 at tournament level (8...Ne8?? looks like it didn't manage 4 plies here):
[Event "4/00-48 CSS 1/85-22"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "Bäurle, Stefan"]
[Black "Mephisto III 12 MHz (Stufe 6)"]
[Result "1-0"]
[ECO "C50"]
[PlyCount "19"]
1. e4 e5 2. Nf3 Nc6 3. Bc4 Be7 4. d3 Nf6 5. Nc3 O-O 6. Ng5 h6 7. h4 hxg5 8.
hxg5 Ne8 9. Qh5 Bb4 10. Qh8# 1-0
Your argument, if valid, means that Rybka 1 ply plays better than 1500, which agrees with what I said. My feeling is that it is quite a bit better than 1500, perhaps more like 1700.
![[ar]](/mwf/v12/flags/ar.png "Argentina")
Hi Larry,
I think I understand what you are saying.
So my question is, what do you think the max elo level that Rybka (or any other chess engine) can reach (with the proper chess knowledge modeled inside it) in 1 ply. I mean without any search.
Do you consider that there is some practical limit o restriction to reach the same elo level of a top five GM?
Best Regards,
Patricio.
I think I understand what you are saying.
So my question is, what do you think the max elo level that Rybka (or any other chess engine) can reach (with the proper chess knowledge modeled inside it) in 1 ply. I mean without any search.
Do you consider that there is some practical limit o restriction to reach the same elo level of a top five GM?
Best Regards,
Patricio.
1 ply of rybka is clearly deeper search than 1 ply of other programs.
Without any search is simply wrong here.
Rybka has no limit that it can reach in 1 ply because Vas can easily tell it to print depth 1 when she get depth 12.
Today he probably tells her to print depth 1 when she gets depth 3 or 4.
Uri
Without any search is simply wrong here.
Rybka has no limit that it can reach in 1 ply because Vas can easily tell it to print depth 1 when she get depth 12.
Today he probably tells her to print depth 1 when she gets depth 3 or 4.
Uri
Thank you Uri.
I might not have explain myself well.
So far I understand, a chess engine to assign a concrete value to a position must evaluate it. In programming terms it might be done through variables that take different values in relation to the position.
If I not wrong that particular value that each variable takes will give you the actual value of one position. That, in some way, represent the knowledge that the engine has to evaluate a specific position. If I'm still not wrong... How accurate this value is in relation to the position will tell you the chess level that the engine has. My question is, if a chess engine could be so strong as a top five GM to analyze a position, but with the chess engine not doing the loop that goes in depth.
Sorry If I'm completely wrong in this question.
Best regards.
Patricio.
I might not have explain myself well.
So far I understand, a chess engine to assign a concrete value to a position must evaluate it. In programming terms it might be done through variables that take different values in relation to the position.
If I not wrong that particular value that each variable takes will give you the actual value of one position. That, in some way, represent the knowledge that the engine has to evaluate a specific position. If I'm still not wrong... How accurate this value is in relation to the position will tell you the chess level that the engine has. My question is, if a chess engine could be so strong as a top five GM to analyze a position, but with the chess engine not doing the loop that goes in depth.
Sorry If I'm completely wrong in this question.
Best regards.
Patricio.
My point is that rybka does some loop on depth even at depth 1.
Chess is a search based game so I believe that it is impossible for a program to be as strong as GM without some significant search.
Uri
Chess is a search based game so I believe that it is impossible for a program to be as strong as GM without some significant search.
Uri
I got it.
Thank you Uri for your explanation.
Best Regards,
Patricio.
Thank you Uri for your explanation.
Best Regards,
Patricio.
I believe that the stated depth in Rybka corresponds to the minimum PV depth at that level. So Rybka 1 ply is really 1 ply in the sense that sometimes the PV is just 1 ply. But often it will be longer than 1 ply, so the strength of Rybka 1 ply is more like 2 or 3 plies in other programs. Please correct me if you know of examples where the PV is shorter than the displayed depth; maybe this can happen due to hash table effects or display issues. My point is that the nominal depth is not arbitrary, there is a rationale for it, even if one could also argue for a higher displayed depth.
The strength of rybka 1 ply is more than the strength of depth 3 of most of the other programs
It seems to me that the strength is between depth 3 and 4 of most programs
I play 3 matchs against depth 3 of other programs
results are:
Rybka2.3.2a depth 1-Toga1.4.1 depth 3 +22 =9 -9(the match was with all the legal position from the initial position with reversed colors meaning 1.a3 1.a4,...)
Rybka 2.3.2a depth1-Shredder9 depth 3 +22 =4 -14
Rybka2.3.2a depth 1-Glaurung2.0.1 depth 3 +35 =2 -3
It seems to me that the strength is between depth 3 and 4 of most programs
I play 3 matchs against depth 3 of other programs
results are:
Rybka2.3.2a depth 1-Toga1.4.1 depth 3 +22 =9 -9(the match was with all the legal position from the initial position with reversed colors meaning 1.a3 1.a4,...)
Rybka 2.3.2a depth1-Shredder9 depth 3 +22 =4 -14
Rybka2.3.2a depth 1-Glaurung2.0.1 depth 3 +35 =2 -3
Thanks, I'm not familiar with those programs except Shredder, so I have no idea how they count plies. Anyway, what I should have said was that the tactical ability of Rybka may be like a standard Fritz-like program with 2 extra plies; the actual strength may be higher due to better eval.
I think that this is often 3 extra plies
Here is a typical tactics of 4 plies that rybka can see at depth 1 based on the score that is +1.35(it is not based on evaluation because rybka can see positive score for black at depth 1 if it is black to move)
Here is a typical tactics of 4 plies that rybka can see at depth 1 based on the score that is +1.35(it is not based on evaluation because rybka can see positive score for black at depth 1 if it is black to move)
k4b2/4p1p1/4P1P1/5p2/8/1N6/8/5K2 w - - 0 1
![[hu]](/mwf/v12/flags/hu.png "Hungary")
Actually, it really is quite arbitrary.
The 'true PV' (ie. the source of the backed-up score) can be shorter than the nominal depth or it can be longer. Usually, it's significantly longer, and the gap (between PV length and depth) typically grows with depth.
There is also no firm relationship between depth and the ability to stand-pat.
If we added 1 to Rybka's depth, or subtracted 1, I can't see any reason why this would be more or less correct. I'm quite flexible about this issue for Rybka 3.
Vas
The 'true PV' (ie. the source of the backed-up score) can be shorter than the nominal depth or it can be longer. Usually, it's significantly longer, and the gap (between PV length and depth) typically grows with depth.
There is also no firm relationship between depth and the ability to stand-pat.
If we added 1 to Rybka's depth, or subtracted 1, I can't see any reason why this would be more or less correct. I'm quite flexible about this issue for Rybka 3.
Vas
![[au]](/mwf/v12/flags/au.png "Australia")
>Actually, it really is quite arbitrary.
Assuming that a search uses an iterative deepening technique, I would think it to be rather natural to denote by "depth" the number of times the root is considered (with obvious fudges for aspiration search), though this need not be a perfect metric.
If a program skips even plies (as I think some in the past did), your rule would report a 21 ply search as only an 11 ply search, which would be quite misleading. Personally I prefer to see the depth reported being one that will generally see simple tactics at the same depth as a traditional program of the 1980s (i.e full width search with check extensions and capture search).
Your example isn't very convincing. If you skip even plies, you would count 1, 3, 5, ... instead of 1, 2, 3, ... as is done in Crafty, Toga, Glaurung, etc. The potential for differences in the extent of the QS and the effects of extensions and reductions are not taken into account, but at least everybody is speaking the same language.
It seems that you and Vas prefer to use artificial metrics whenever possible, either because you feel it really makes more sense, or more likely because it obscures the reality of what you are doing. If the goal is to protect IP, this is probably counterproductive as it provides a powerful incentive to reverse engineer the code.
Regards,
Alan
It seems that you and Vas prefer to use artificial metrics whenever possible, either because you feel it really makes more sense, or more likely because it obscures the reality of what you are doing. If the goal is to protect IP, this is probably counterproductive as it provides a powerful incentive to reverse engineer the code.
Regards,
Alan
I don't know about what reasons Vas had for defining the depth the way he did starting with Rybka 1 ( I wasn't involved then), but at this point (for Rybka 3) the only issue is to define the depth in a way that is most consistent with what people would expect. Due to Rybka's peculiarities, counting every iteration from the first possible one would produce a 1 ply search that would play ridiculously badly, something like a 1 ply search with no quiescence, and we would be criticized for trying to claim unreasonably large search depths. My recommendation to Vas is to increase the reported count by 2, as I feel that the search depths would then be comparable to most normal programs.
... My recommendation to Vas is to increase the reported count by 2, as I feel that the search depths would then be comparable to most normal programs.
Please Larry, don´t hurry. Shown depth is a sensitive thing, I think, after reading all the posts here in this forum. My remarks:
1. Rybka is #1 and she makes the rules what is normal (all the other new engines (updates) will be like Rybka).
2. If I remember right, Zappa Mexico (#2) has lower depth (normally) than Rybka in the middlegame in the Mexican tournament.
3. Rybka has very often incredible high depth in the endgame (much higher than Zappa); but I don´t trust in this high depths. Zappa is much more careful here.
So my vote: Rybka is almost rigth in showing depth, but she has to open her window in endgames much more :-).
Please Larry, don´t hurry. Shown depth is a sensitive thing, I think, after reading all the posts here in this forum. My remarks:
1. Rybka is #1 and she makes the rules what is normal (all the other new engines (updates) will be like Rybka).
2. If I remember right, Zappa Mexico (#2) has lower depth (normally) than Rybka in the middlegame in the Mexican tournament.
3. Rybka has very often incredible high depth in the endgame (much higher than Zappa); but I don´t trust in this high depths. Zappa is much more careful here.
So my vote: Rybka is almost rigth in showing depth, but she has to open her window in endgames much more :-).
Actually there is a second reason for my recommendation to increase nominal depth by 2. As things are now, Rybka 3 set to 1 ply search will simply be too strong for the majority of amateur players, and most (or all) interfaces don't allow the user to set search depth to zero (or negative). So an increase of 2 will provide two substantially weaker levels that ordinary amateurs can beat. Imagine your frustration if you found that there is no level weak enough for you to beat (I exclude artificial levels that deliberately blunder). Of course even with this change the lowest level will still be too strong for many people, but at least the majority of Rybka purchasers should be able to beat a 1 ply search if it is redefined as I propose. Otherwise that will probably not be true.
![[us]](/mwf/v12/flags/us.png "United States")
I'll probably have to wait for the pocket pc version at 1 ply to be able to beat it. :)
... but at least the majority of Rybka purchasers should be able to beat a 1 ply search ...
He Larry, the opposite is true! I´m waiting for the Rybka engine, which is always rigth at ply 1 (after 0 seconds on my watch). Okay, I can wait :-)! But who cares about Rybka purchasers and their play against Rybka? They are all believers and are lucky in domination of the engine. Ok, only 99.99% after I have read the Benjamin commentary to his last tournament against Rybka. There was a time (good old times), where only the the winner give interviews and comments the games. Okay, Rybka can´t speech, but if she can hear all the reproaches against her playing strength by the the loser, she will be very angry.
He Larry, the opposite is true! I´m waiting for the Rybka engine, which is always rigth at ply 1 (after 0 seconds on my watch). Okay, I can wait :-)! But who cares about Rybka purchasers and their play against Rybka? They are all believers and are lucky in domination of the engine. Ok, only 99.99% after I have read the Benjamin commentary to his last tournament against Rybka. There was a time (good old times), where only the the winner give interviews and comments the games. Okay, Rybka can´t speech, but if she can hear all the reproaches against her playing strength by the the loser, she will be very angry.
Hm, that would mean that a ply 17 search in Rybka 2.3.2a would be the same as a ply 19 search in rybka 3?
Yes, at least until the endgame.
> If a program skips even plies
Perhaps I mis-explained and/or over-simplified my description - I meant for the calculation of "depth" to be essentially the same as counting. To wit:
a) Start at depth 0
b) When iteratively deepening by X ply, add X to the depth [fractional/negative increments allowed]
In this way, search depth is well-defined, and in many cases, concords with what might be intuitively expected. As you say, its utility to the user might be less clear when q-search is considered.
Suppose that the first iteration in a hypothetical program was 1 ply with no quiescence search, the second allowed 1 ply of capture, the third allowed 2, the fourth 3, the fifth 4, the sixth 5, and the seventh unlimited (or whatever maximum might be used). Then the eighth iteration would be 2 ply with full quiescence etc. This is a crude approximation to the situation with Rybka. Currently we would call the seventh iteration "1 ply" in this hypothetical program. You would call the very first (or perhaps second) one "1 ply". Your solution might be technically "correct", but it would hardly be consistent with standard industry practice. I say that the first iteration to count is the first one that plays at some relatively minimal level, maybe 1200 or 1300 Elo. That means for Rybka 3 probably starting two levels of quiescence earlier than we do now.
> Currently we would call the seventh iteration "1 ply" in this hypothetical program.
I would probably use fractional ply to describe the situation between iteration 1 and iteration 7 in this case (it is not clear to me if the term "iterative deepening" should apply for these first iterations, as it could be argued that they use substantively dissimilar methodologies). Perhaps 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 1.0, 2.0, though the decimals would be place-holders rather than connoting any arithmetic meaning. Alternatively, the UCI-variable
seldepth could give some notion of this. Or, if you use various qsearch-es throughout the search, two "depths" could be returned at any time: the horizon-distance (from the root), and the qsearch-thoroughness. In any event, with your example, I would hope that the 7th iteration (1 ply + full q) and the 8th (2 ply + full q) would have depths that differ by 1. As I indicate above, without the rigidity of the iterative deepening framework (which I realise is perhaps too strong an assumption to have broad validity in modern chess engines), the particularities of the labelling of iterations vis-a-vis depth becomes much less clear. On the other hand, I can't say that I have great sentiment toward re-construing "depth" merely to conform to the numerology of previous times. [I am reminded of the "rated speed" of various processors, which was supposed to ease comparisons for consumers, but didn't always deliver such a result].
Please give me a hint. If Rybka has no reported depth (only new PV every 10 seconds), is there any problem? Okay, some GUI (like Fritz) appoint to depth. But who cares? Ask a GM after his game, which depth he arrived after 40 minutes of thinking. He will think, you are imbicile.
> Please give me a hint. If Rybka has no reported depth (only new PV every 10 seconds), is there any problem
For that matter, why even return a PV? For instance, if all you care about is Rybka playing the best move, the PV can be considered essentially useless.
Conversely, if (say) you want a chess analysis tool, having more robust output can be beneficial. For instance, if the Rybka output reads "depth 30" in some endgame, you might want to be able to re-interpret this (when combined with positional considerations and engine particularities) in terms of a confidence level regarding the validity of the search result.
These suggests are certainly reasonably, as are Larry's.
In fact, we're going to do two things with the depths for Rybka 3:
1) The general depths will be higher (probably by two) for the reasons Larry mentions.
2) The depths will be adjusted by adding or subtracting a constant derived from some property of the root so that there is a close correlation between the depth and the time taken to search that depths. In practice, this will mean subtracting 1 or 2 from the output depth as the position simplifies.
#2 will be useful for people who like to invoke fixed-depth searches and to play fixed-depth games, for example with the randomizer. Currently, the quality of randomizer game degrades as the position gets simplified.
We're not planning to report any 'seldepth's.
And yes, probably some people will complain that we don't follow the standards. My position on this is that when it's appropriate, we'll set the standards.
Vas
In fact, we're going to do two things with the depths for Rybka 3:
1) The general depths will be higher (probably by two) for the reasons Larry mentions.
2) The depths will be adjusted by adding or subtracting a constant derived from some property of the root so that there is a close correlation between the depth and the time taken to search that depths. In practice, this will mean subtracting 1 or 2 from the output depth as the position simplifies.
#2 will be useful for people who like to invoke fixed-depth searches and to play fixed-depth games, for example with the randomizer. Currently, the quality of randomizer game degrades as the position gets simplified.
We're not planning to report any 'seldepth's.
And yes, probably some people will complain that we don't follow the standards. My position on this is that when it's appropriate, we'll set the standards.
Vas
> And yes, probably some people will complain that we don't follow the standards. My position on this is that when it's appropriate, we'll set the standards.
I don't exactly care if what you use is considered standard, but tather that there is some documentation. [I recently saw a talk by Paul Zimmermann on semantics and such - the thrust of the first half was: "No specification => No bug ... but useless." :)]
Attachment: BUG.pdf (31k)
A chess program can evaluate a position without search only if it is quiescent, meaning that there are no tactics left unresolved. Only for such positions does it make sense to talk about how well a program evaluates compared to a human (and even then it's only very approximate). I estimate that Rybka 3 will evaluate such positions as well as a human master who evaluates without any analysis of variations (this is hard to prove of course). We have made very steady progress in this area, with no sign of reaching the end, but whether it is possible for a program like Rybka to ever evaluate (without search) as well as a top five GM does is hard to say. I believe it might be possible using radically different methods than we do now, but more likely this will never happen because other search techniques will make this unnecessary. As Uri points out, the distinction between search and eval is rather arbitrary anyway.
Thank you Larry!!
Best Regards,
Patricio.
Best Regards,
Patricio.
Is it possible that it could be decreased ? ;-)
I wonder that in some conditions it is real danger i.e. short reflection time, inferior hardware etc. ?
Rgds
Hetman
I wonder that in some conditions it is real danger i.e. short reflection time, inferior hardware etc. ?
Rgds
Hetman
![[fi]](/mwf/v12/flags/fi.png "Finland")
Well, the more expensive eval probably results in heavier use of 64-bit operations so I imagine the 32-bit slowdown could get slightly bigger.
Not if the estimates are remotely close to the truth.
> I wonder that in some conditions it is real danger i.e. short reflection time, inferior hardware etc. ?
Not if the estimates are remotely close to the truth.
I am not sure.
I have been analysing my OTB game with Rybka 2.2 and 2.3.2.a with time per move 30 sec.
R2.2 has found winning shot in that interval 2.3.2.a not. :-(
My opponnent has not found it either :-) .
Rgds
Hetman
I have been analysing my OTB game with Rybka 2.2 and 2.3.2.a with time per move 30 sec.
R2.2 has found winning shot in that interval 2.3.2.a not. :-(
My opponnent has not found it either :-) .
Rgds
Hetman
I do not see a reason to assume that more expensive evaluation means bigger slow down for 32 bits.
Maybe it is the opposite and more expensive evaluation means that you have to use some 32 bit operations and cannot use 64 bits.
Uri
Maybe it is the opposite and more expensive evaluation means that you have to use some 32 bit operations and cannot use 64 bits.
Uri
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill