Look at the game below, Romi came out of the book with a rather comfy position, but managed to play many blunders..
[Event "?"]
[Site "chessclub.com"]
[Date "2008.6.13"]
[Time "2:33:32"]
[Round "?"]
[White "apocalypse1112"]
[Black "Tinker"]
[Result "0-1"]
[WhiteElo "2307"]
[BlackElo "2437"]
[Opening "B90 Sicilian Najdorf, Byrne (English) attack"]
{Time control = 2 10 Engine book(s): }
1. e4 {0.00/1 0.3 } 1... c5 2. Nf3 {0.00/1 0.4 } 2... d6 3. d4 {0.00/1 0.7 }
3... cxd4 4. Nxd4 {0.00/1 0.7 } 4... Nf6 5. Nc3 {0.00/1 0.3 } 5... a6 6. Be3
{0.00/1 0.8 } 6... e5 7. Nb3 {0.00/1 0.3 } 7... Be6 8. f3 {0.00/1 0.3 } 8... Nbd7
9. Qd2 {0.00/1 0.3 } 9... b5 10. a4 {0.00/1 0.3 } 10... b4 11. Nd5 {0.00/1 0.3 }
11... Bxd5 12. exd5 {0.00/1 0.3 } 12... Nb6 13. Bxb6 {0.00/1 0.3 } 13... Qxb6
14. a5 {0.00/1 0.3 } 14... Qb7 15. Bc4 {0.00/1 0.3 } 15... g6 16. Ra4 {0.00/1 0.3
} 16... Rb8 17. Nc1 {0.00/1 0.3 } 17... h5 18. Na2 {0.00/1 0.3 } 18... Bh6 19. Qe2
{0.00/1 0.3 } 19... O-O 20. Nxb4 {0.00/1 0.3 } 20... e4 21. Bxa6 {0.00/1 10.4 }
21... Qc7 22. c3 {0.00/1 10.1 } 22... Qc5 23. Qd1 {1.03/12 9.2 } 23... exf3 24. gxf3
{0.28/13 9.6 } 24... Rfe8+ 25. Kf1 {0.92/13 7.3 } 25... Qe3 26. Be2 {1.84/12 9.2
} 26... Ne4 27. fxe4 {-0.67/14 7.6 } 27... Qh3+ 28. Ke1 {-0.70/15 8.2 } 28... Rxe4
29. Ra2 {-0.70/13 11.9 } 29... Rbe8 30. b3 {-6.37/12 7.5 } 30... Qg2 31. Rf1
{-7.37/15 9.5 } 31... Bg7 32. Qd3 {apocalypse1112 resigns} 0-1
Something is seriously wrong (with romi)
Moderators: hgm, Rebel, chrisw
-
- Posts: 6640
- Joined: Thu Mar 09, 2006 4:21 am
Re: Something is seriously wrong (with romi)
I'd guess Romi has
Slow Search algorithm + moderate evaluation + excellent Learning
Slow Search helps in longer time controls. Extremely longer like 2 hours with increment. Romi can play a lot of game with Rybka and learn from it and will be useful for correspondence games. Slow search algorithm makes the engines think deeper on few lines, as opposed to fast search - evaluating several different lines and choosing the best and continuing the search on it further. Ask Chris about Slow Search vs Fast search, I'm no expert
Slow Search algorithm + moderate evaluation + excellent Learning
Slow Search helps in longer time controls. Extremely longer like 2 hours with increment. Romi can play a lot of game with Rybka and learn from it and will be useful for correspondence games. Slow search algorithm makes the engines think deeper on few lines, as opposed to fast search - evaluating several different lines and choosing the best and continuing the search on it further. Ask Chris about Slow Search vs Fast search, I'm no expert
Re: Something is seriously wrong (with romi)
I think that Romi is a moderately fast searcher, look at the NPS..Romi is rated much higher than Tinker on my 1+1 list, so there is gotta be another reason why Romi is missing the obvious attacks..
-
- Posts: 6640
- Joined: Thu Mar 09, 2006 4:21 am
Re: Something is seriously wrong (with romi)
Could be a bug in that particular version you were using. Try comparing the analysis lines of the two versions (Ng4 and current one you have) against the move where Romi went horribly wrong and see if they match.Tony Thomas wrote:I think that Romi is a moderately fast searcher, look at the NPS..Romi is rated much higher than Tinker on my 1+1 list, so there is gotta be another reason why Romi is missing the obvious attacks..
Re: Something is seriously wrong (with romi)
Look at the next game, almost 0 problems..
Code: Select all
[Event "?"]
[Site "chessclub.com"]
[Date "2008.6.13"]
[Time "4:04:46"]
[Round "?"]
[White "Timea"]
[Black "apocalypse1112"]
[Result "0-1"]
[WhiteElo "2368"]
[BlackElo "2294"]
[Opening "A00 Saragossa opening"]
{Time control = 2 12 Engine book(s): }
1. c3 e5 {0.00/1 8.3 } 2. d4 e4 {0.28/13 12.4 } 3. e3 Nf6 {0.49/12 5.2 } 4. d5
b6 {0.00/1 12.6 } 5. Nh3 Bb7 {0.00/12 6.4 } 6. Bc4 Na6 {0.00/1 10.6 } 7. Na3 Bxa3
{0.00/13 8.2 } 8. bxa3 O-O {0.07/13 11.1 } 9. O-O d6 {0.13/13 10.1 } 10. Rb1 Nc5
{0.46/13 11.1 } 11. Nf4 Nfd7 {0.00/1 18.4 } 12. Bb2 g5 {0.00/1 9.0 } 13. Nh5 Ne5
{0.00/1 10.2 } 14. Be2 f5 {0.00/1 13.6 } 15. c4 Qe8 {0.39/12 17.2 } 16. Bxe5 dxe5
{0.00/1 13.6 } 17. g4 f4 {0.52/13 14.1 } 18. Qd2 Qe7 {0.00/1 15.5 } 19. Rfd1 Rad8
{0.87/13 9.2 } 20. Kh1 Bc6 {0.91/11 11.1 } 21. Rb2 Ba4 {1.18/12 11.5 } 22. Rdb1
Be8 {1.15/13 12.5 } 23. Qc3 Na4 {1.45/13 13.4 } 24. Qb4 Qd7 {1.93/14 11.2 } 25. Rd2
f3 {0.00/1 12.5 } 26. Bf1 Qxg4 {0.00/1 13.9 } 27. Ng3 Rd6 {0.00/1 15.9 } 28. Qb3
Bd7 {2.13/14 11.7 } 29. c5 Nxc5 {2.72/12 11.6 } 30. Qc3 Rh6 {3.11/14 19.4 } 31. Qxe5
Qh4 {4.74/14 11.8 } 32. Nh5 Qxh5 {5.55/13 8.1 } 33. Qg3 Rd6 {0.00/1 15.2 } 34. Rc1
Rd8 {0.00/1 13.6 } 35. Rd4 a5 {5.72/12 11.1 } 36. Bc4 h6 {0.00/1 14.1 } 37. Bf1
Kh7 {5.81/12 9.3 } 38. Rb1 a4 {5.83/13 11.2 } 39. Bc4 Qg6 {0.00/1 14.4 } 40. Bf1
Qf6 {5.93/13 11.5 } 41. h3 h5 {0.00/1 19.7 } 42. Qh2 Qf5 {6.08/12 11.2 } 43. Rbd1
h4 {6.30/14 20.2 } 44. Qg1 Be8 {0.00/1 13.4 } 45. Qh2 Kh6 {0.00/1 12.1 } 46. Kg1
Bd7 {0.00/1 16.0 } 47. Kh1 g4 {0.00/1 13.1 } 48. hxg4 Qxg4 {6.68/14 13.5 } 49. Bc4
Rg8 {7.12/14 12.1 } 50. Rg1 Qxg1+ {7.42/17 12.6 } 51. Qxg1 Rxg1+ {8.09/18 12.2 }
52. Kxg1 Kg6 {0.00/1 15.4 } 53. Rd1 Kf5 {0.00/1 15.4 } 54. Rd2 Ke5 {0.00/1 13.8 }
55. Rd1 b5 {9.97/19 13.6 } 56. Bf1 Rxd5 {11.22/21 14.9 } 57. Rb1 c6 {11.23/20 11.0
} 58. Rc1 Kd6 {11.65/20 14.5 } 59. Re1 Rd2 {11.96/19 14.4 } 60. Ra1 Nd3 {13.34/18
11.7 } 61. Rb1 {Timea resigns} 0-1
-
- Posts: 536
- Joined: Thu Mar 09, 2006 3:01 pm
Re: Something is seriously wrong (with romi)
Romi version p3k is one of the engines I use in Tinker gauntlet tests.
It typically searches about 3x faster than Tinker. Perhaps this is due to the "minimal" eval
(but it does extensive eval pcsq and mobility initialization at the root and ply 1 nodes I think).
Anyway, the interesting thing is that Tinker version 6.50 is about 80 elo below Romi,
yet Tinker version 6.54 (current experimental version) is only about 25 elo less.
I run 50 test postions (20 Nunn II and 30 Noomen 2006) with both black and white,
so 100 games at level 0 1 2. There are 16 engines in the gauntlet.
Moreover, I have recently noticed that different Tinker versions tend to score about the same overall (1,600 games),
but the individual results against specific opponents vary quite a lot.
I think this is due to small Tinker changes finding different strengths/weaknesses vs specific engines.
In addition, when playing on ICC the 64 bit version of Tinker does not support its own book,
so Tinker runs under BookThinker. Currently Tinker is using the d-corbit-v02__superbook.abk book,
converted from Arena .abk format to .pgn with the tool by Osipov Jury
modified for Windows by Dann Corbit and promotions corrected by Richard Pijl, see
http://64.68.157.89/forum/viewtopic.php?t=20661
Then it is cleaned up with PGNExtract by David Barnes, and finally loaded with makebook.
The point is that Tinker's own book has always been very poor, so this may also be a factor.
I plan to use a different book for the ACCA WCRCC, FWIW
It typically searches about 3x faster than Tinker. Perhaps this is due to the "minimal" eval
(but it does extensive eval pcsq and mobility initialization at the root and ply 1 nodes I think).
Anyway, the interesting thing is that Tinker version 6.50 is about 80 elo below Romi,
yet Tinker version 6.54 (current experimental version) is only about 25 elo less.
I run 50 test postions (20 Nunn II and 30 Noomen 2006) with both black and white,
so 100 games at level 0 1 2. There are 16 engines in the gauntlet.
Moreover, I have recently noticed that different Tinker versions tend to score about the same overall (1,600 games),
but the individual results against specific opponents vary quite a lot.
I think this is due to small Tinker changes finding different strengths/weaknesses vs specific engines.
In addition, when playing on ICC the 64 bit version of Tinker does not support its own book,
so Tinker runs under BookThinker. Currently Tinker is using the d-corbit-v02__superbook.abk book,
converted from Arena .abk format to .pgn with the tool by Osipov Jury
modified for Windows by Dann Corbit and promotions corrected by Richard Pijl, see
http://64.68.157.89/forum/viewtopic.php?t=20661
Then it is cleaned up with PGNExtract by David Barnes, and finally loaded with makebook.
The point is that Tinker's own book has always been very poor, so this may also be a factor.
I plan to use a different book for the ACCA WCRCC, FWIW
Re: Something is seriously wrong (with romi)
As you already know I have tinker 6.18 and it is rated much lower than Romi in my rating list. If I had to make a guess, you have been working your behind of lately, my first impression based on the few ICC games is that Tinker has indeed improved a lot were as Romi is at a standstill due to Mike having personal reasons. I will be trying to play against you tonight using winboard were romi isnt handicapped..
-
- Posts: 6640
- Joined: Thu Mar 09, 2006 4:21 am
Re: Something is seriously wrong (with romi)
TwistedLogic-wc At ICC would like to challenge Apocalypse. So Please look out for me and seek challenge.Tony Thomas wrote:As you already know I have tinker 6.18 and it is rated much lower than Romi in my rating list. If I had to make a guess, you have been working your behind of lately, my first impression based on the few ICC games is that Tinker has indeed improved a lot were as Romi is at a standstill due to Mike having personal reasons. I will be trying to play against you tonight using winboard were romi isnt handicapped..
Re: Something is seriously wrong (with romi)
I hate you Swami, you defeated Romi three times..swami wrote:TwistedLogic-wc At ICC would like to challenge Apocalypse. So Please look out for me and seek challenge.Tony Thomas wrote:As you already know I have tinker 6.18 and it is rated much lower than Romi in my rating list. If I had to make a guess, you have been working your behind of lately, my first impression based on the few ICC games is that Tinker has indeed improved a lot were as Romi is at a standstill due to Mike having personal reasons. I will be trying to play against you tonight using winboard were romi isnt handicapped..
you aint getting no goat loving anytime soon..
Tony