So Stock is right on track, especially with simplifications, I am a bit stuck, maybe it is time for a stake after all?
Stuck
Moderators: hgm, Rebel, chrisw
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Stuck
As I am a bit stuck in my eval suggestions regarding SF, and besides interest is very low, I hesitated very much if to post another thread, but in the end curiosity prevailed. Maybe I will just post one or 2 more very short subthreads, I know no one will answer, but still, I know, just as with me passively reading info posted by other members, a bit of additional info will probably not hurt, someone might use this, consciously or subconsciously, maybe even in the distant future.
So Stock is right on track, especially with simplifications, I am a bit stuck, maybe it is time for a stake after all?
So Stock is right on track, especially with simplifications, I am a bit stuck, maybe it is time for a stake after all?
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
An imbalance suggestion that might work in SF
Well, obviously most of the reasonable imbalance suggestions that might be implemented in SF do not work precisely because of the ubiquitous and unclear to many SF imbalance/quadratic table. But here is one that could probably do just fine in spite of this table, as it relates only to the more or less simplistic imbalance of minor vs pawns. The quadratic magic might still strike with vengeance, but still, you tune just smaller values and against pawns at that. Who knows, maybe the quadratic will not have such an impact this time?
Below a game giving some insight into the particular imbalance.
[pgn][PlyCount "90"]
[MLNrOfMoves "45"]
[MLFlags "000100"]
[Date "2014.04.21"]
[Round "427"]
[White "Gull"]
[Black "SF"]
[Result "1-0"]
[TimeControl "64+0.32"]
1. d4 {book} 1... Nf6 {book} 2. c4 {book} 2... g6 {book} 3. Nc3 {book} 3... Bg7
{book} 4. e4 {book} 4... d6 {book} 5. Be2 {book} 5... O-O {book} 6. Bg5 {book}
6... Nbd7 {book} 7. Qd2 {book} 7... e5 {book} 8. Nf3 {book} 8... Qe8 {book} 9.
dxe5 {+0.52/17 1.9s} 9... dxe5 {-0.23/21 2.3s} 10. Qc2 {+0.45/17 0.95s} 10...
Nc5 {-0.26/22 1.8s} 11. Be3 {+0.31/17 2.4s} 11... Ne6 {-0.17/23 1.9s} 12. Rd1
{+0.39/16 0.92s} 12... Ng4 {+0.22/19 1.7s} 13. Bc1 {+0.12/16 2.2s} 13... Nd4
{+0.07/21 2.5s} 14. Nxd4 {+0.03/17 2.8s} 14... exd4 {+0.10/22 2.2s} 15. Nb5
{+0.03/19 3.6s} 15... Qe5 {+0.06/22 2.3s} 16. f4 {+0.11/18 5.2s} 16... Qc5
{+0.03/19 1.6s} 17. h3 {+0.24/14 0.73s} 17... Nh6 {+0.05/21 2.4s} 18. Qd3
{+0.22/16 2.2s} 18... Rd8 {+0.05/23 4.5s} 19. O-O {+0.26/17 3.1s} 19... a6
{+0.08/22 2.6s} 20. Na3 {+0.26/16 0.000s} 20... b6 {-0.11/21 2.2s} 21. Bf3
{+0.17/16 0.94s} 21... Qe7 {-0.11/23 5.4s} 22. g4 {+0.28/15 1.1s} 22... Rb8
{-0.06/19 1.4s} 23. Nc2 {+0.36/16 1.4s} 23... Kh8 {-0.21/19 3.3s} 24. b4
{+0.32/17 3.1s} 24... c6 {-0.39/21 4.2s} 25. e5 {+0.61/15 0.67s} 25... c5
{-0.49/21 1.5s} 26. bxc5 {+0.62/17 1.3s} 26... bxc5 {-0.62/24 1.9s} 27. Ne3
{+0.62/16 0.000s} 27... Qh4 {-0.41/24 0.84s} 28. Bg2 {+0.69/14 0.80s} 28... f6
{-0.68/23 1.8s} 29. exf6 {+0.92/14 0.52s} 29... Bxf6 {-0.50/21 1.5s} 30. Nd5
{+0.92/16 0.69s} 30... Bg7 {-0.60/21 2.3s} 31. Bd2 {+0.93/17 2.9s} 31... Nxg4
{-0.45/21 1.1s} 32. hxg4 {+1.26/15 0.84s} 32... Bxg4 {-0.75/22 2.5s} 33. Rc1
{+1.25/14 0.83s} 33... Bf5 {-0.88/22 2.9s} 34. Qa3 {+1.24/16 0.55s} 34... Re8
{-0.75/22 0.67s} 35. Rf2 {+1.24/17 1.1s} 35... Bf8 {-1.07/23 5.6s} 36. Ba5
{+1.33/17 5.4s} 36... Rbc8 {-0.64/18 0.98s} 37. Bc7 {+1.33/16 0.000s} 37... Bg7
{-0.54/20 1.3s} 38. Bd6 {+1.33/16 0.000s} 38... d3 {-0.89/20 1.2s} 39. Bxc5
{+1.41/17 2.2s} 39... Bg4 {-0.48/20 1.2s} 40. Be7 {+1.82/14 0.36s} 40... Qg3
{-1.12/19 0.69s} 41. Qd6 {+1.82/13 0.000s} 41... Bh3 {-1.55/21 0.72s} 42. Bf6
{+1.64/15 0.41s} 42... Rxc4 {-1.36/24 0.95s} 43. Rcf1 {+1.69/16 0.31s} 43... Rc2
{-1.49/25 0.67s} 44. Rxc2 {+1.52/18 1.1s} 44... dxc2 {-1.55/25 0.88s} 45. Bxg7+
{+1.52/18 0.22s} 45... Kxg7 {White wins by adjudication} 1-0
[/pgn]
[d]1rbr3k/6bp/p5p1/2pN4/2Pp1Pnq/3Q3P/P2B2B1/3R1RK1 w - - 0 32
In an objectively worse position, SF sacrifices a pice for 2 pawns, probably even a worse decision. This is actually a pattern in SF play, it likes frequently sacrificing a minor piece for a couple of pawns, with the end result very often being bad.
[d]1r1r3k/6bp/p5p1/2pN1b2/2Pp1P1q/Q7/P2B2B1/2R2RK1 b - - 0 34
Gull shows 120cps white edge, SF just 70cps. That is too big a difference in a relatively quiet position. Obviously, something is wrong with the minor-pawn imbalance in SF.
[d]2r1rb1k/7p/p5p1/B1pN1b2/2Pp1P1q/Q7/P4RB1/2R3K1 w - - 0 37
130cps white edge Gull, 60cps SF.
In light of the above, my particular suggestion would be: just for the middlegame (endgame values are OK, for me), increase value for minor (B or N) over pawns, maybe by some 5-10%. I.e., that is precisely the opposite of what Mr. Jonathan Mysseno did in trying to weigh pawns higher. But again, this should be valid only for the middlegame.
Comments of course welcome, but I know my confidence credit is almost exhausted.
Below a game giving some insight into the particular imbalance.
[pgn][PlyCount "90"]
[MLNrOfMoves "45"]
[MLFlags "000100"]
[Date "2014.04.21"]
[Round "427"]
[White "Gull"]
[Black "SF"]
[Result "1-0"]
[TimeControl "64+0.32"]
1. d4 {book} 1... Nf6 {book} 2. c4 {book} 2... g6 {book} 3. Nc3 {book} 3... Bg7
{book} 4. e4 {book} 4... d6 {book} 5. Be2 {book} 5... O-O {book} 6. Bg5 {book}
6... Nbd7 {book} 7. Qd2 {book} 7... e5 {book} 8. Nf3 {book} 8... Qe8 {book} 9.
dxe5 {+0.52/17 1.9s} 9... dxe5 {-0.23/21 2.3s} 10. Qc2 {+0.45/17 0.95s} 10...
Nc5 {-0.26/22 1.8s} 11. Be3 {+0.31/17 2.4s} 11... Ne6 {-0.17/23 1.9s} 12. Rd1
{+0.39/16 0.92s} 12... Ng4 {+0.22/19 1.7s} 13. Bc1 {+0.12/16 2.2s} 13... Nd4
{+0.07/21 2.5s} 14. Nxd4 {+0.03/17 2.8s} 14... exd4 {+0.10/22 2.2s} 15. Nb5
{+0.03/19 3.6s} 15... Qe5 {+0.06/22 2.3s} 16. f4 {+0.11/18 5.2s} 16... Qc5
{+0.03/19 1.6s} 17. h3 {+0.24/14 0.73s} 17... Nh6 {+0.05/21 2.4s} 18. Qd3
{+0.22/16 2.2s} 18... Rd8 {+0.05/23 4.5s} 19. O-O {+0.26/17 3.1s} 19... a6
{+0.08/22 2.6s} 20. Na3 {+0.26/16 0.000s} 20... b6 {-0.11/21 2.2s} 21. Bf3
{+0.17/16 0.94s} 21... Qe7 {-0.11/23 5.4s} 22. g4 {+0.28/15 1.1s} 22... Rb8
{-0.06/19 1.4s} 23. Nc2 {+0.36/16 1.4s} 23... Kh8 {-0.21/19 3.3s} 24. b4
{+0.32/17 3.1s} 24... c6 {-0.39/21 4.2s} 25. e5 {+0.61/15 0.67s} 25... c5
{-0.49/21 1.5s} 26. bxc5 {+0.62/17 1.3s} 26... bxc5 {-0.62/24 1.9s} 27. Ne3
{+0.62/16 0.000s} 27... Qh4 {-0.41/24 0.84s} 28. Bg2 {+0.69/14 0.80s} 28... f6
{-0.68/23 1.8s} 29. exf6 {+0.92/14 0.52s} 29... Bxf6 {-0.50/21 1.5s} 30. Nd5
{+0.92/16 0.69s} 30... Bg7 {-0.60/21 2.3s} 31. Bd2 {+0.93/17 2.9s} 31... Nxg4
{-0.45/21 1.1s} 32. hxg4 {+1.26/15 0.84s} 32... Bxg4 {-0.75/22 2.5s} 33. Rc1
{+1.25/14 0.83s} 33... Bf5 {-0.88/22 2.9s} 34. Qa3 {+1.24/16 0.55s} 34... Re8
{-0.75/22 0.67s} 35. Rf2 {+1.24/17 1.1s} 35... Bf8 {-1.07/23 5.6s} 36. Ba5
{+1.33/17 5.4s} 36... Rbc8 {-0.64/18 0.98s} 37. Bc7 {+1.33/16 0.000s} 37... Bg7
{-0.54/20 1.3s} 38. Bd6 {+1.33/16 0.000s} 38... d3 {-0.89/20 1.2s} 39. Bxc5
{+1.41/17 2.2s} 39... Bg4 {-0.48/20 1.2s} 40. Be7 {+1.82/14 0.36s} 40... Qg3
{-1.12/19 0.69s} 41. Qd6 {+1.82/13 0.000s} 41... Bh3 {-1.55/21 0.72s} 42. Bf6
{+1.64/15 0.41s} 42... Rxc4 {-1.36/24 0.95s} 43. Rcf1 {+1.69/16 0.31s} 43... Rc2
{-1.49/25 0.67s} 44. Rxc2 {+1.52/18 1.1s} 44... dxc2 {-1.55/25 0.88s} 45. Bxg7+
{+1.52/18 0.22s} 45... Kxg7 {White wins by adjudication} 1-0
[/pgn]
[d]1rbr3k/6bp/p5p1/2pN4/2Pp1Pnq/3Q3P/P2B2B1/3R1RK1 w - - 0 32
In an objectively worse position, SF sacrifices a pice for 2 pawns, probably even a worse decision. This is actually a pattern in SF play, it likes frequently sacrificing a minor piece for a couple of pawns, with the end result very often being bad.
[d]1r1r3k/6bp/p5p1/2pN1b2/2Pp1P1q/Q7/P2B2B1/2R2RK1 b - - 0 34
Gull shows 120cps white edge, SF just 70cps. That is too big a difference in a relatively quiet position. Obviously, something is wrong with the minor-pawn imbalance in SF.
[d]2r1rb1k/7p/p5p1/B1pN1b2/2Pp1P1q/Q7/P4RB1/2R3K1 w - - 0 37
130cps white edge Gull, 60cps SF.
In light of the above, my particular suggestion would be: just for the middlegame (endgame values are OK, for me), increase value for minor (B or N) over pawns, maybe by some 5-10%. I.e., that is precisely the opposite of what Mr. Jonathan Mysseno did in trying to weigh pawns higher. But again, this should be valid only for the middlegame.
Comments of course welcome, but I know my confidence credit is almost exhausted.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Powerful h file
SF too frequently underestimates the importance of the open h file, when the enemy king has castled on a file adjacent to it (i.e., h file for white with black Kg8). I have seen this much too often in my private games against SF, but it is frequent also in other SF games against other engines. Below one such game.
[pgn][PlyCount "106"]
[MLNrOfMoves "53"]
[MLFlags "010100"]
[Date "2014.04.20"]
[Round "345"]
[White "Gull"]
[Black "SF"]
[Result "1-0"]
[TimeControl "64+0.32"]
1. e4 {book} 1... c6 {book} 2. Nc3 {book} 2... d5 {book} 3. Nf3 {book} 3... Nf6
{book} 4. e5 {book} 4... Ne4 {book} 5. Be2 {book} 5... e6 {book} 6. O-O {book}
6... Be7 {book} 7. Nb1 {book} 7... O-O {book} 8. d3 {book} 8... Nc5 {book} 9. c4
{+0.32/15 1.3s} 9... d4 {-0.18/21 3.0s} 10. Bf4 {+0.34/18 2.0s} 10... b6
{-0.17/24 2.7s} 11. Na3 {+0.35/19 2.5s} 11... Nca6 {-0.05/23 2.1s} 12. h3
{+0.28/18 2.2s} 12... Nb4 {+0.15/20 1.1s} 13. Kh1 {+0.14/16 3.1s} 13... Bb7
{+0.27/22 2.0s} 14. Re1 {+0.08/17 1.9s} 14... c5 {+0.48/23 3.6s} 15. Nh2
{-0.11/16 3.1s} 15... a6 {+0.49/22 3.0s} 16. Nc2 {-0.04/16 0.91s} 16... N4c6
{+0.47/20 3.1s} 17. Bf3 {0.00/18 2.2s} 17... b5 {+0.47/21 1.0s} 18. Be4
{-0.13/17 3.2s} 18... bxc4 {+0.60/19 2.0s} 19. dxc4 {-0.08/17 4.3s} 19... Nd7
{+0.57/21 3.6s} 20. Nf3 {-0.08/16 0.000s} 20... Qb6 {+0.58/19 2.5s} 21. b3
{0.00/15 0.047s} 21... Rfb8 {+0.64/20 1.4s} 22. Rb1 {+0.07/16 1.6s} 22... a5
{+0.56/18 2.0s} 23. a3 {0.00/17 2.4s} 23... Qc7 {+0.53/19 2.1s} 24. Qe2
{-0.05/15 1.5s} 24... Qc8 {+0.53/19 3.2s} 25. Bd2 {+0.02/15 1.9s} 25... Qd8
{+0.46/19 3.5s} 26. a4 {+0.14/17 2.4s} 26... Nf8 {+0.34/21 2.9s} 27. Na3
{+0.32/16 1.1s} 27... Ra6 {+0.34/21 3.4s} 28. Nb5 {+0.36/18 1.8s} 28... Ba8
{+0.24/21 4.7s} 29. Rbd1 {+0.37/17 1.2s} 29... Rab6 {+0.23/21 2.4s} 30. Qd3
{+0.41/16 1.1s} 30... Ng6 {+0.23/22 1.4s} 31. Qb1 {+0.39/17 1.9s} 31... Bb7
{+0.25/20 2.8s} 32. Re2 {+0.39/16 0.98s} 32... Ra8 {+0.25/21 1.1s} 33. Kg1
{+0.41/17 3.1s} 33... Raa6 {+0.03/21 5.1s} 34. g3 {+0.45/15 0.45s} 34... Qb8
{+0.09/18 0.50s} 35. Kf1 {+0.35/16 1.5s} 35... Nf8 {+0.27/19 0.81s} 36. Rde1
{+0.31/17 1.7s} 36... h6 {+0.15/19 3.0s} 37. Kg1 {+0.31/17 1.1s} 37... Bd8
{+0.12/21 0.91s} 38. Nh2 {+0.23/16 6.9s} 38... Ne7 {+0.08/20 0.39s} 39. Ng4
{+0.25/17 1.0s} 39... h5 {+0.16/19 0.64s} 40. Nh2 {+0.26/15 0.34s} 40... Bxe4
{+0.02/19 1.6s} 41. Rxe4 {+0.44/16 0.81s} 41... Ra8 {+0.08/19 0.34s} 42. Qd1
{+0.49/16 0.77s} 42... g6 {+0.04/21 0.34s} 43. Nf3 {+0.47/17 0.64s} 43... Nc6
{0.00/21 1.7s} 44. Kg2 {+0.44/16 0.69s} 44... Nd7 {0.00/21 0.83s} 45. Qe2
{+0.46/15 0.58s} 45... Be7 {0.00/19 0.45s} 46. g4 {+0.58/15 0.73s} 46... hxg4
{-0.12/21 0.81s} 47. hxg4 {+0.58/13 0.000s} 47... Qb7 {-0.48/16 0.52s} 48. Rh1
{+0.58/16 0.53s} 48... Bf8 {-0.62/15 0.28s} 49. Bh6 {+0.66/16 0.84s} 49... Be7
{-1.03/16 0.28s} 50. g5 {+0.86/13 0.22s} 50... d3 {-0.36/15 0.38s} 51. Qxd3
{+0.86/14 0.45s} 51... Nd4 {-0.58/19 0.33s} 52. Reh4 {+2.09/14 0.36s} 52... Nxf3
{-1.63/18 0.31s} 53. Qxf3 {+2.33/15 0.30s} 53... Qxf3+
{White wins by adjudication} 1-0
[/pgn]
[d]rq1b2k1/3n1p2/1rn1p1p1/pNp1P2p/P1PpR3/1P3NPP/3B1PK1/3QR3 w - - 0 45
+50cps white edge Gull, 0.0 score SF. SF sees it as perfectly equal, the possibility for the h file to open after white plays g3-g4 does not bother SF in the least.
[d]rq4k1/3nbp2/1rn1p1p1/pNp1P2p/P1PpR1P1/1P3N1P/3BQPK1/4R3 b - - 0 46
Now already SF sees black is losing by half a pawn, but still not a decisive advantage.
[d]r5k1/1q1nbp2/1rn1p1p1/pNp1P3/P1PpR1P1/1P3N2/3BQPK1/7R b - - 0 48
A rook on an open h or a edge file with the enemy king on an adjacent file, like Rh1 and Kg8, certainly deserves a very nice bonus over the regular rook attacking value. Such a rook on an end file is much more powerful than otherwise, simply because in most cases the remoteness of the file makes difficult or impossible the support of own defenders. So that a and h edge attacking files for the rooks are special cases.
[d]r5k1/1q1nbp2/1r2p1pB/pNp1P1P1/P1Pn3R/1P1Q1N2/5PK1/7R b - - 0 52
It is only now that SF sees the danger, after the white rooks are doubled along the h file. Look carefully again at the position: the additional bonus for rook on the h file is precisely because of the very difficult support for the black pieces to the own king; as the file is remote, those pieces have less access there.
In light of the above, and because this is a pattern in SF games, my suggestion would be the following: give further nice bonus, maybe at least some 30cps, for a rook on an open h or a file, when the enemy king is on an adjacent file, i.e. bonus for Rh1 with enemy Kg8, and Ra1 with enemy Kb8.
Have you observed frequent SF losses in similar lines?
[pgn][PlyCount "106"]
[MLNrOfMoves "53"]
[MLFlags "010100"]
[Date "2014.04.20"]
[Round "345"]
[White "Gull"]
[Black "SF"]
[Result "1-0"]
[TimeControl "64+0.32"]
1. e4 {book} 1... c6 {book} 2. Nc3 {book} 2... d5 {book} 3. Nf3 {book} 3... Nf6
{book} 4. e5 {book} 4... Ne4 {book} 5. Be2 {book} 5... e6 {book} 6. O-O {book}
6... Be7 {book} 7. Nb1 {book} 7... O-O {book} 8. d3 {book} 8... Nc5 {book} 9. c4
{+0.32/15 1.3s} 9... d4 {-0.18/21 3.0s} 10. Bf4 {+0.34/18 2.0s} 10... b6
{-0.17/24 2.7s} 11. Na3 {+0.35/19 2.5s} 11... Nca6 {-0.05/23 2.1s} 12. h3
{+0.28/18 2.2s} 12... Nb4 {+0.15/20 1.1s} 13. Kh1 {+0.14/16 3.1s} 13... Bb7
{+0.27/22 2.0s} 14. Re1 {+0.08/17 1.9s} 14... c5 {+0.48/23 3.6s} 15. Nh2
{-0.11/16 3.1s} 15... a6 {+0.49/22 3.0s} 16. Nc2 {-0.04/16 0.91s} 16... N4c6
{+0.47/20 3.1s} 17. Bf3 {0.00/18 2.2s} 17... b5 {+0.47/21 1.0s} 18. Be4
{-0.13/17 3.2s} 18... bxc4 {+0.60/19 2.0s} 19. dxc4 {-0.08/17 4.3s} 19... Nd7
{+0.57/21 3.6s} 20. Nf3 {-0.08/16 0.000s} 20... Qb6 {+0.58/19 2.5s} 21. b3
{0.00/15 0.047s} 21... Rfb8 {+0.64/20 1.4s} 22. Rb1 {+0.07/16 1.6s} 22... a5
{+0.56/18 2.0s} 23. a3 {0.00/17 2.4s} 23... Qc7 {+0.53/19 2.1s} 24. Qe2
{-0.05/15 1.5s} 24... Qc8 {+0.53/19 3.2s} 25. Bd2 {+0.02/15 1.9s} 25... Qd8
{+0.46/19 3.5s} 26. a4 {+0.14/17 2.4s} 26... Nf8 {+0.34/21 2.9s} 27. Na3
{+0.32/16 1.1s} 27... Ra6 {+0.34/21 3.4s} 28. Nb5 {+0.36/18 1.8s} 28... Ba8
{+0.24/21 4.7s} 29. Rbd1 {+0.37/17 1.2s} 29... Rab6 {+0.23/21 2.4s} 30. Qd3
{+0.41/16 1.1s} 30... Ng6 {+0.23/22 1.4s} 31. Qb1 {+0.39/17 1.9s} 31... Bb7
{+0.25/20 2.8s} 32. Re2 {+0.39/16 0.98s} 32... Ra8 {+0.25/21 1.1s} 33. Kg1
{+0.41/17 3.1s} 33... Raa6 {+0.03/21 5.1s} 34. g3 {+0.45/15 0.45s} 34... Qb8
{+0.09/18 0.50s} 35. Kf1 {+0.35/16 1.5s} 35... Nf8 {+0.27/19 0.81s} 36. Rde1
{+0.31/17 1.7s} 36... h6 {+0.15/19 3.0s} 37. Kg1 {+0.31/17 1.1s} 37... Bd8
{+0.12/21 0.91s} 38. Nh2 {+0.23/16 6.9s} 38... Ne7 {+0.08/20 0.39s} 39. Ng4
{+0.25/17 1.0s} 39... h5 {+0.16/19 0.64s} 40. Nh2 {+0.26/15 0.34s} 40... Bxe4
{+0.02/19 1.6s} 41. Rxe4 {+0.44/16 0.81s} 41... Ra8 {+0.08/19 0.34s} 42. Qd1
{+0.49/16 0.77s} 42... g6 {+0.04/21 0.34s} 43. Nf3 {+0.47/17 0.64s} 43... Nc6
{0.00/21 1.7s} 44. Kg2 {+0.44/16 0.69s} 44... Nd7 {0.00/21 0.83s} 45. Qe2
{+0.46/15 0.58s} 45... Be7 {0.00/19 0.45s} 46. g4 {+0.58/15 0.73s} 46... hxg4
{-0.12/21 0.81s} 47. hxg4 {+0.58/13 0.000s} 47... Qb7 {-0.48/16 0.52s} 48. Rh1
{+0.58/16 0.53s} 48... Bf8 {-0.62/15 0.28s} 49. Bh6 {+0.66/16 0.84s} 49... Be7
{-1.03/16 0.28s} 50. g5 {+0.86/13 0.22s} 50... d3 {-0.36/15 0.38s} 51. Qxd3
{+0.86/14 0.45s} 51... Nd4 {-0.58/19 0.33s} 52. Reh4 {+2.09/14 0.36s} 52... Nxf3
{-1.63/18 0.31s} 53. Qxf3 {+2.33/15 0.30s} 53... Qxf3+
{White wins by adjudication} 1-0
[/pgn]
[d]rq1b2k1/3n1p2/1rn1p1p1/pNp1P2p/P1PpR3/1P3NPP/3B1PK1/3QR3 w - - 0 45
+50cps white edge Gull, 0.0 score SF. SF sees it as perfectly equal, the possibility for the h file to open after white plays g3-g4 does not bother SF in the least.
[d]rq4k1/3nbp2/1rn1p1p1/pNp1P2p/P1PpR1P1/1P3N1P/3BQPK1/4R3 b - - 0 46
Now already SF sees black is losing by half a pawn, but still not a decisive advantage.
[d]r5k1/1q1nbp2/1rn1p1p1/pNp1P3/P1PpR1P1/1P3N2/3BQPK1/7R b - - 0 48
A rook on an open h or a edge file with the enemy king on an adjacent file, like Rh1 and Kg8, certainly deserves a very nice bonus over the regular rook attacking value. Such a rook on an end file is much more powerful than otherwise, simply because in most cases the remoteness of the file makes difficult or impossible the support of own defenders. So that a and h edge attacking files for the rooks are special cases.
[d]r5k1/1q1nbp2/1r2p1pB/pNp1P1P1/P1Pn3R/1P1Q1N2/5PK1/7R b - - 0 52
It is only now that SF sees the danger, after the white rooks are doubled along the h file. Look carefully again at the position: the additional bonus for rook on the h file is precisely because of the very difficult support for the black pieces to the own king; as the file is remote, those pieces have less access there.
In light of the above, and because this is a pattern in SF games, my suggestion would be the following: give further nice bonus, maybe at least some 30cps, for a rook on an open h or a file, when the enemy king is on an adjacent file, i.e. bonus for Rh1 with enemy Kg8, and Ra1 with enemy Kb8.
Have you observed frequent SF losses in similar lines?
-
jhellis3
- Posts: 546
- Joined: Sat Aug 17, 2013 12:36 am
Re: An imbalance suggestion that might work in SF
I would say you are mistaken in this particular position. AFAICS, the position appears to be lost regardless of what black plays.
As for the assertion that SF needs to value minor pieces more in the middle game, knights and bishops are already worth 4.1 and 4.2 pawns, which (I may be mistaken here) is more than most engines.
For this particular position, it appears black is already lost and thus if there is a clear mistake in that game it must have come before the knight sac. That is unless you have a clearly drawing line for black?
As for the assertion that SF needs to value minor pieces more in the middle game, knights and bishops are already worth 4.1 and 4.2 pawns, which (I may be mistaken here) is more than most engines.
For this particular position, it appears black is already lost and thus if there is a clear mistake in that game it must have come before the knight sac. That is unless you have a clearly drawing line for black?
-
carldaman
- Posts: 2283
- Joined: Sat Jun 02, 2012 2:13 am
Re: Powerful h file
Hi Lyudmil,Lyudmil Tsvetkov wrote:SF too frequently underestimates the importance of the open h file, when the enemy king has castled on a file adjacent to it (i.e., h file for white with black Kg8). I have seen this much too often in my private games against SF, but it is frequent also in other SF games against other engines. Below one such game.
[pgn][PlyCount "106"]
[MLNrOfMoves "53"]
[MLFlags "010100"]
[Date "2014.04.20"]
[Round "345"]
[White "Gull"]
[Black "SF"]
[Result "1-0"]
[TimeControl "64+0.32"]
1. e4 {book} 1... c6 {book} 2. Nc3 {book} 2... d5 {book} 3. Nf3 {book} 3... Nf6
{book} 4. e5 {book} 4... Ne4 {book} 5. Be2 {book} 5... e6 {book} 6. O-O {book}
6... Be7 {book} 7. Nb1 {book} 7... O-O {book} 8. d3 {book} 8... Nc5 {book} 9. c4
{+0.32/15 1.3s} 9... d4 {-0.18/21 3.0s} 10. Bf4 {+0.34/18 2.0s} 10... b6
{-0.17/24 2.7s} 11. Na3 {+0.35/19 2.5s} 11... Nca6 {-0.05/23 2.1s} 12. h3
{+0.28/18 2.2s} 12... Nb4 {+0.15/20 1.1s} 13. Kh1 {+0.14/16 3.1s} 13... Bb7
{+0.27/22 2.0s} 14. Re1 {+0.08/17 1.9s} 14... c5 {+0.48/23 3.6s} 15. Nh2
{-0.11/16 3.1s} 15... a6 {+0.49/22 3.0s} 16. Nc2 {-0.04/16 0.91s} 16... N4c6
{+0.47/20 3.1s} 17. Bf3 {0.00/18 2.2s} 17... b5 {+0.47/21 1.0s} 18. Be4
{-0.13/17 3.2s} 18... bxc4 {+0.60/19 2.0s} 19. dxc4 {-0.08/17 4.3s} 19... Nd7
{+0.57/21 3.6s} 20. Nf3 {-0.08/16 0.000s} 20... Qb6 {+0.58/19 2.5s} 21. b3
{0.00/15 0.047s} 21... Rfb8 {+0.64/20 1.4s} 22. Rb1 {+0.07/16 1.6s} 22... a5
{+0.56/18 2.0s} 23. a3 {0.00/17 2.4s} 23... Qc7 {+0.53/19 2.1s} 24. Qe2
{-0.05/15 1.5s} 24... Qc8 {+0.53/19 3.2s} 25. Bd2 {+0.02/15 1.9s} 25... Qd8
{+0.46/19 3.5s} 26. a4 {+0.14/17 2.4s} 26... Nf8 {+0.34/21 2.9s} 27. Na3
{+0.32/16 1.1s} 27... Ra6 {+0.34/21 3.4s} 28. Nb5 {+0.36/18 1.8s} 28... Ba8
{+0.24/21 4.7s} 29. Rbd1 {+0.37/17 1.2s} 29... Rab6 {+0.23/21 2.4s} 30. Qd3
{+0.41/16 1.1s} 30... Ng6 {+0.23/22 1.4s} 31. Qb1 {+0.39/17 1.9s} 31... Bb7
{+0.25/20 2.8s} 32. Re2 {+0.39/16 0.98s} 32... Ra8 {+0.25/21 1.1s} 33. Kg1
{+0.41/17 3.1s} 33... Raa6 {+0.03/21 5.1s} 34. g3 {+0.45/15 0.45s} 34... Qb8
{+0.09/18 0.50s} 35. Kf1 {+0.35/16 1.5s} 35... Nf8 {+0.27/19 0.81s} 36. Rde1
{+0.31/17 1.7s} 36... h6 {+0.15/19 3.0s} 37. Kg1 {+0.31/17 1.1s} 37... Bd8
{+0.12/21 0.91s} 38. Nh2 {+0.23/16 6.9s} 38... Ne7 {+0.08/20 0.39s} 39. Ng4
{+0.25/17 1.0s} 39... h5 {+0.16/19 0.64s} 40. Nh2 {+0.26/15 0.34s} 40... Bxe4
{+0.02/19 1.6s} 41. Rxe4 {+0.44/16 0.81s} 41... Ra8 {+0.08/19 0.34s} 42. Qd1
{+0.49/16 0.77s} 42... g6 {+0.04/21 0.34s} 43. Nf3 {+0.47/17 0.64s} 43... Nc6
{0.00/21 1.7s} 44. Kg2 {+0.44/16 0.69s} 44... Nd7 {0.00/21 0.83s} 45. Qe2
{+0.46/15 0.58s} 45... Be7 {0.00/19 0.45s} 46. g4 {+0.58/15 0.73s} 46... hxg4
{-0.12/21 0.81s} 47. hxg4 {+0.58/13 0.000s} 47... Qb7 {-0.48/16 0.52s} 48. Rh1
{+0.58/16 0.53s} 48... Bf8 {-0.62/15 0.28s} 49. Bh6 {+0.66/16 0.84s} 49... Be7
{-1.03/16 0.28s} 50. g5 {+0.86/13 0.22s} 50... d3 {-0.36/15 0.38s} 51. Qxd3
{+0.86/14 0.45s} 51... Nd4 {-0.58/19 0.33s} 52. Reh4 {+2.09/14 0.36s} 52... Nxf3
{-1.63/18 0.31s} 53. Qxf3 {+2.33/15 0.30s} 53... Qxf3+
{White wins by adjudication} 1-0
[/pgn]
[d]rq1b2k1/3n1p2/1rn1p1p1/pNp1P2p/P1PpR3/1P3NPP/3B1PK1/3QR3 w - - 0 45
+50cps white edge Gull, 0.0 score SF. SF sees it as perfectly equal, the possibility for the h file to open after white plays g3-g4 does not bother SF in the least.
[d]rq4k1/3nbp2/1rn1p1p1/pNp1P2p/P1PpR1P1/1P3N1P/3BQPK1/4R3 b - - 0 46
Now already SF sees black is losing by half a pawn, but still not a decisive advantage.
[d]r5k1/1q1nbp2/1rn1p1p1/pNp1P3/P1PpR1P1/1P3N2/3BQPK1/7R b - - 0 48
A rook on an open h or a edge file with the enemy king on an adjacent file, like Rh1 and Kg8, certainly deserves a very nice bonus over the regular rook attacking value. Such a rook on an end file is much more powerful than otherwise, simply because in most cases the remoteness of the file makes difficult or impossible the support of own defenders. So that a and h edge attacking files for the rooks are special cases.
[d]r5k1/1q1nbp2/1r2p1pB/pNp1P1P1/P1Pn3R/1P1Q1N2/5PK1/7R b - - 0 52
It is only now that SF sees the danger, after the white rooks are doubled along the h file. Look carefully again at the position: the additional bonus for rook on the h file is precisely because of the very difficult support for the black pieces to the own king; as the file is remote, those pieces have less access there.
In light of the above, and because this is a pattern in SF games, my suggestion would be the following: give further nice bonus, maybe at least some 30cps, for a rook on an open h or a file, when the enemy king is on an adjacent file, i.e. bonus for Rh1 with enemy Kg8, and Ra1 with enemy Kb8.
Have you observed frequent SF losses in similar lines?
(Don't feel discouraged, the interest IS there for others, but time is limited)
The "rook on h-file" may be more of an issue when SF is defending, rather than attacking. I like the idea of a penalty (or bonus, depending on which side) as you suggest, but only if the side with the rook on the h-file has another rook or a queen on the board as well.
Doubling on the h-file would have to be a possibility within the position, even if remote. Otherwise the penalty might be counterproductive.
Regards,
CL
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Closed refinements
Many thanks go specially to Marco Costalba for attempting to create a closed position detector in SF. SF certainly needs this, as it could help the engine avoid drawish continuations that it would otherwise regard as advantageous, for example when playing against me or other humans.
I know people will not pay much attention to what I write, but still below is my particularised vision of employing scaling down of score with closed positions:
- a position is closed when at least 6 pawns overall are blocked; of those, at least 3 must be central or semicentral c,d,e, or f pawns
- in the detector, central d and e files might get double value and semicentral c and f files 1.5 value than other files
- very small scaling down of score with just 6 blocked pawns, just some 10%
- 20-30% scaling down of score with 7 blocked pawns
- 50% scaling of score with 8 blocked pawns in all cases
- outright 0.0 draw score with 8 blocked pawns, but only when the side leading in score leads by less than 2 full pawns. I think this makes perfect sense, as, if the score is bigger than 2 pawns, different sacrifices are possible, for example minor piece for 2 pawns, 2 minor pieces for 3 pawns, etc.
- when we define that at least 6 pawns must be blocked for a closed eval to exist, and that at least 3 of those must be on c-f files, it is not necessary to specify that the pawns must belong to a single chain; they might very well belong to more, say 2 chains for example. And actually, this would be quite a frequent case.
I do not ask Marco anything, as I know he is quite busy, but probably it would still be worth to try to finetune a bit more the already written closed detector. Maybe it is just this finetuning that narrows the possibility for a closed patch to pass LTC.
Comments of course, appreciated. I would be very happy if SF sometime implements this fine positional feature.
I know people will not pay much attention to what I write, but still below is my particularised vision of employing scaling down of score with closed positions:
- a position is closed when at least 6 pawns overall are blocked; of those, at least 3 must be central or semicentral c,d,e, or f pawns
- in the detector, central d and e files might get double value and semicentral c and f files 1.5 value than other files
- very small scaling down of score with just 6 blocked pawns, just some 10%
- 20-30% scaling down of score with 7 blocked pawns
- 50% scaling of score with 8 blocked pawns in all cases
- outright 0.0 draw score with 8 blocked pawns, but only when the side leading in score leads by less than 2 full pawns. I think this makes perfect sense, as, if the score is bigger than 2 pawns, different sacrifices are possible, for example minor piece for 2 pawns, 2 minor pieces for 3 pawns, etc.
- when we define that at least 6 pawns must be blocked for a closed eval to exist, and that at least 3 of those must be on c-f files, it is not necessary to specify that the pawns must belong to a single chain; they might very well belong to more, say 2 chains for example. And actually, this would be quite a frequent case.
I do not ask Marco anything, as I know he is quite busy, but probably it would still be worth to try to finetune a bit more the already written closed detector. Maybe it is just this finetuning that narrows the possibility for a closed patch to pass LTC.
Comments of course, appreciated. I would be very happy if SF sometime implements this fine positional feature.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: Closed refinements
[d]6k1/5p2/1p1p1Pp1/pPpPp1Pp/P1P1P2P/8/8/6K1 w - - 0 1
Closed position with blocked pawns belonging to 2 chains of opposite colours.
[d]6k1/2nn1p2/1p1p1Pp1/pPpPp1Pp/P1P1P2P/2BN1N2/8/6K1 w - - 0 1
When one of the sides leads by more than 2 pawns, it would be wrong to assess the position as drawish, as sacrifices are possible, for example on e5 above.
Closed position with blocked pawns belonging to 2 chains of opposite colours.
[d]6k1/2nn1p2/1p1p1Pp1/pPpPp1Pp/P1P1P2P/2BN1N2/8/6K1 w - - 0 1
When one of the sides leads by more than 2 pawns, it would be wrong to assess the position as drawish, as sacrifices are possible, for example on e5 above.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: An imbalance suggestion that might work in SF
Hi Joseph. Thanks for replying.jhellis3 wrote:I would say you are mistaken in this particular position. AFAICS, the position appears to be lost regardless of what black plays.
As for the assertion that SF needs to value minor pieces more in the middle game, knights and bishops are already worth 4.1 and 4.2 pawns, which (I may be mistaken here) is more than most engines.
For this particular position, it appears black is already lost and thus if there is a clear mistake in that game it must have come before the knight sac. That is unless you have a clearly drawing line for black?
You are perfectly right that SF is already lost in the particular position, but that still does not change its score, which is still to low, and the salient feature is the minor-pawn imbalance.
Please take a look also at the first game against Komodo here http://talkchess.com/forum/viewtopic.ph ... 30&t=51888, where it is B vs 3 pawns in the middlegame. Same score favouring the pawns, when it should favour the minor piece. There are many more examples, but most are stored only in my memory. There is no doubt that SF underestimates minors vs pawns in the middlegame, but that would be true only of the middlegame.
-
Lyudmil Tsvetkov
- Posts: 6052
- Joined: Tue Jun 12, 2012 12:41 pm
Re: Powerful h file
Hi Carl.carldaman wrote:Hi Lyudmil,
(Don't feel discouraged, the interest IS there for others, but time is limited)
The "rook on h-file" may be more of an issue when SF is defending, rather than attacking. I like the idea of a penalty (or bonus, depending on which side) as you suggest, but only if the side with the rook on the h-file has another rook or a queen on the board as well.
Doubling on the h-file would have to be a possibility within the position, even if remote. Otherwise the penalty might be counterproductive.
Regards,
CL
Thanks for the support, nice to have a chess player's support amongst all those programmers...
Perfect observations. The bonus should be valid only in middlegame with queens present. Not in the endgame. Even better to specify that doubling of rooks or Q+R along that file gives even bigger bonus.
But, you know, I am afraid to specify, as programmers want it as slim as possible. SF does not even have doubled rooks on an open file, not to mention doubled rooks or R+Q along h/a attacking file. If someone of the SF team reads this, they are going to kill me. That is why I suggested simply rook on an edge h/a attacking file bonus.
Another perfect observation is that SF possibly understands this rule better when attacking than when defending, but why this is so is difficult to say, maybe has to do with search. Or maybe just we concentrate a bit too much on SF lost games, it is also possible that SF could get a boost also in attacking, who knows?
-
jhellis3
- Posts: 546
- Joined: Sat Aug 17, 2013 12:36 am
Re: An imbalance suggestion that might work in SF
So you are saying a minor piece should be worth more than 4.2 pawns?Lyudmil Tsvetkov wrote:There is no doubt that SF underestimates minors vs pawns in the middlegame, but that would be true only of the middlegame.