The limits of "Just-mobility-evaluation"

[OliverBr]

This fine loss against Rybka shows where are the problems of the current evaluation (just mobility) of OliThink 5.0.8:

In move 12 OliThink created a nice isolated triple pawn:

Code: Select all

rn1qk2r/pp3ppp/2p1p3/3p3n/3P1P1P/1PNB1P2/P1PQ1P2/R3K2R b KQkq - 0 12

It is evident that it will be nearly impossible to keep those pawns. Rybka will eventually shoot all its (her?) pieces on those pawns and eat them all three.

Unfortunately the first pawn falls not sooner than 23 plies later, beyond any search horizon.

So I guess, pawn structure will the first important add-on to the mobility evaltion to make OliThink significant stronger...
Are there other opinions about this?

PS: How can I display here a board with fen string?

Code: Select all

[Event "ICS Rated Blitz match"]
[Site "chessclub.com"]
[Date "2008.01.29"]
[Round "-"]
[White "OliPow"]
[Black "BountyHunter"]
[Result "*"]
[WhiteElo "2700"]
[BlackElo "3276"]
[TimeControl "180"]

1. Nf3 d6 2. Nc3 d5 3. d4 Nf6 4. Bf4 Bf5 5. e3 e6 6. h4 Bd6 7. Bb5+ c6 8.
Bd3 Bxf4 9. exf4 Bg4 10. b3 Nh5 11. Qd2 Bxf3 12. gxf3 Qf6 13. Ne2 Nd7 14.
c3 Nf8 15. c4 Ng6 16. Bxg6 fxg6 17. cxd5 exd5 18. Qe3+ Kd7 19. O-O-O Rhf8
20. Kb1 Rae8 21. Qd3 Qf5 22. Qxf5+ Rxf5 23. Rhe1 Nxf4 24. Ng3 Rxe1 25. Rxe1
Rf8 26. Rd1 Ne6 27. Rd3 Rf4 28. h5 gxh5 29. Nxh5 Rh4 30. Ng3 Nxd4 31. Ne4
Ne6 32. a4 h5 33. Kc2 Rf4 34. Ng3 h4 35. Ne2 Rf6 36. Kd1 h3 37. Ng3 h2 38.
Kc2 b6 39. Re3 Nd4+ 40. Kb2 Rxf3 41. Rxf3 Nxf3 42. b4 c5 43. bxc5 bxc5 44.
Kc1 d4 45. Kc2 c4 46. Kb2 d3 47. Nh1 d2 48. Kc2 c3 49. a5 Ke6 50. Ng3 Nd4+
51. Kxc3 d1=Q 52. Kc4 Qc2+ 53. Kxd4 Qxf2+ 54. Kc4 Qxg3 55. Kc5 h1=Q 56. Kb4
Qd5 57. Ka4 Qgb3#
*

[Uri Blass]

You can display a diagram by writing D before the fen when you write [ before the D and ] after the D

[D]rn1qk2r/pp3ppp/2p1p3/3p3n/3P1P1P/1PNB1P2/P1PQ1P2/R3K2R b KQkq - 0 12

I think that pawn structure was not the only problem in the game.

14.0-0-0 is only slightly inferior for white based on rybka

OliPow - BountyHunter
[D]r3k2r/pp1n1ppp/2p1pq2/3p3n/3P1P1P/1P1B1P2/P1PQNP2/R3K2R w KQkq - 0 1

Analysis by Rybka 2.3.2a 32-bit :

14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 0-0 17.Kb2 a5
= (-0.09) Depth: 7 00:00:00 5kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 0-0 17.Kb2 a5
= (-0.16) Depth: 8 00:00:00 10kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 0-0 17.Kb2 a5
= (-0.16) Depth: 9 00:00:00 21kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 a5 17.Kb2 a4 18.Qe5
= (-0.13) Depth: 10 00:00:00 41kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 a5 17.Qe5 Qxe5 18.fxe5 a4 19.Rg1
= (-0.15) Depth: 11 00:00:01 94kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 a5 17.Qe5 Qxe5 18.fxe5 f6 19.Kb2 fxe5 20.Re3
= (-0.15) Depth: 12 00:00:03 182kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 a5 17.Qe5 Qxe5 18.fxe5 a4 19.Rg1 axb3 20.Rxb3 b5
= (-0.16) Depth: 13 00:00:05 321kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 a5 17.Qe5 Qxe5 18.fxe5 a4 19.Rg1 axb3 20.Rxb3 b5
= (-0.14) Depth: 14 00:00:12 704kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 a5 17.Qe5 Qxe5 18.fxe5 a4 19.Rg1 axb3 20.Rxb3 b5
= (-0.17) Depth: 15 00:00:18 1126kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 a5 17.Qe5 Qxe5 18.fxe5 Rf8 19.f4 f6 20.Rh2 Ke7
= (-0.17) Depth: 16 00:00:39 2479kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 a5 17.Qe5 Qxe5 18.fxe5 Rf8 19.f4 f6 20.Rh2 Ke7
= (-0.21) Depth: 17 00:01:09 4388kN
14.0-0-0 Nc5 15.Qe3 Nxd3+ 16.Rxd3 a5 17.Qe5 Qxe5 18.fxe5 Rf8 19.Rg1 f6 20.Kd2 fxe5
= (-0.19) Depth: 18 00:02:15 8456kN

(Uri, MyTown 29.01.2008)

[Dirt]

You can display a diagram by writing D before the fen when you write [ before the D and ] after the D

[D]rn1qk2r/pp3ppp/2p1p3/3p3n/3P1P1P/1PNB1P2/P1PQ1P2/R3K2R b KQkq - 0 12

Frustratingly, it doesn't work during preview. You actually have to submit your post before you can check that it worked.

[OliverBr]

You can display a diagram by writing D before the fen when you write [ before the D and ] after the D

[D]rn1qk2r/pp3ppp/2p1p3/3p3n/3P1P1P/1PNB1P2/P1PQ1P2/R3K2R b KQkq - 0 12

I think that pawn structure was not the only problem in the game.

That's suprising, because as far as my chess knowledge goes, white's pawn structure is quite bad.
But perhaps it's less worse than I thought until white opens the black's f-file. Since that moment the game is lost.

[Uri Blass]

You can display a diagram by writing D before the fen when you write [ before the D and ] after the D

[D]rn1qk2r/pp3ppp/2p1p3/3p3n/3P1P1P/1PNB1P2/P1PQ1P2/R3K2R b KQkq - 0 12

I think that pawn structure was not the only problem in the game.

That's suprising, because as far as my chess knowledge goes, white's pawn structure is quite bad.
But perhaps it's less worse than I thought until white opens the black's f-file. Since that moment the game is lost.

I believe that it is a common mistake of humans to overevaluate pawn structure and one of the reasons that rybka is better is the fact that rybka has smaller scores for bad pawn structure relative to most programs.

Uri

[Gerd Isenberg]

This fine loss against Rybka shows where are the problems of the current evaluation (just mobility) of OliThink 5.0.8:

In move 12 OliThink created a nice isolated triple pawn.

Unfortunately the first pawn falls not sooner than 23 plies later, beyond any search horizon.

So I guess, pawn structure will the first important add-on to the mobility evaltion to make OliThink significant stronger...
Are there other opinions about this?

Triple pawns are positional monsters, but of course you have some half-open lines as "compensation" and probably too high mobility score because of that. You may consider triple pawns by intersection of pawns with their front-and back-spans for a penalty or to scale down mobility. For a rough measure you may even try the difference in number of pawn islands - the dispersion as mentioned in:
http://chessprogramming.wikispaces.com/ ... Distortion

[OliverBr]

I believe that it is a common mistake of humans to overevaluate pawn structure and one of the reasons that rybka is better is the fact that rybka has smaller scores for bad pawn structure relative to most programs.

Uri

This sounds reasonable. OliThink 5.0.8 has no pawn structure at all and has easily a rating of 2700 on ICC.
But on the other hand, most games OliThink loses against other engines (It has not lost against any human, GM, IM or whatever yet, looks like a "human killer") were lost because in endgames the other engines eat all the pawns that have been completely unstructured. So did Rybka in that very game.

[BubbaTough]

OliThink 5.0.8 has no pawn structure at all and has easily a rating of 2700 on ICC.

You have definitely done a great job on your engine, but from what I observed in our games I think you may not yet be at a reliable 2700 on ICC yet (unless you were to play only vs. humans, which seems to give all engines who do that a several hundred point pop). I would suggest playing a bunch of games with the non-rybka engines that regularly show up (Tinker, Arasan, Amateur, Now, LearningLemming, Crafty) to get a good feel for where your engine stacks up.

I have tried to limit Lemming's # of games with you because some people get very annoyed at playing the same engine over and over again (Lemming plays with auto-rematch) but would be happy to log in and play bunches of games if you would like.

-Sam

[bob]

You can display a diagram by writing D before the fen when you write [ before the D and ] after the D

[D]rn1qk2r/pp3ppp/2p1p3/3p3n/3P1P1P/1PNB1P2/P1PQ1P2/R3K2R b KQkq - 0 12

I think that pawn structure was not the only problem in the game.

That's suprising, because as far as my chess knowledge goes, white's pawn structure is quite bad.
But perhaps it's less worse than I thought until white opens the black's f-file. Since that moment the game is lost.

I believe that it is a common mistake of humans to overevaluate pawn structure and one of the reasons that rybka is better is the fact that rybka has smaller scores for bad pawn structure relative to most programs.

Uri

One thing is clear. In this position if white doesn't have some marvelous tactical idea to press on with, he is going to lose any endgame reached with that pawn structure...

[Uri Blass]

You can display a diagram by writing D before the fen when you write [ before the D and ] after the D

[D]rn1qk2r/pp3ppp/2p1p3/3p3n/3P1P1P/1PNB1P2/P1PQ1P2/R3K2R b KQkq - 0 12

I think that pawn structure was not the only problem in the game.

That's suprising, because as far as my chess knowledge goes, white's pawn structure is quite bad.
But perhaps it's less worse than I thought until white opens the black's f-file. Since that moment the game is lost.

I believe that it is a common mistake of humans to overevaluate pawn structure and one of the reasons that rybka is better is the fact that rybka has smaller scores for bad pawn structure relative to most programs.

Uri

One thing is clear. In this position if white doesn't have some marvelous tactical idea to press on with, he is going to lose any endgame reached with that pawn structure...

I agree that I expect white to lose endgame with that pawn structure but
I tried it in comp-comp game and rybka as white can win in part of the games because white does not have to trade pieces and get an endgame with that pawn structure.

Here is an example.

[Event "URI-AMD, Blitz:3'+3""]
[Site "URI-AMD"]
[Date "2008.01.30"]
[Round "2"]
[White "Rybka 2.3.2a 32-bit"]
[Black "Toga II 1.4 beta5c"]
[Result "1-0"]
[Annotator "-0.17;-0.46"]
[SetUp "1"]
[FEN "rn1qk2r/pp3ppp/2p1p3/3p3n/3P1P1P/1PNB1P2/P1PQ1P2/R3K2R b KQkq - 0 1"]
[PlyCount "168"]
[TimeControl "180+3"]

1... Qb6 {Both last book move -0.17/12 9} 2. Bf1 {-0.17/12 8} Nd7 {-0.46/14 11}
3. O-O-O {(f4-f5) -0.16/12 2} O-O-O {-0.76/15 10} 4. Kb2 {(Bf1-h3) -0.15/13 9}
g6 {-0.85/15 15} 5. Qe3 {(Bf1-h3) -0.13/13 6} Qc7 {-0.90/16 9} 6. Ne2 {
-0.17/12 2} Kb8 {-0.89/16 8} 7. Bh3 {-0.19/12 3} Nhf6 {(Rh8-e8) -0.93/15 8} 8.
Rhg1 {-0.19/12 11} Rc8 {-0.98/14 8} 9. Nc3 {(c2-c3) -0.21/12 25} Rhe8 {
(Rh8-g8) -1.07/14 7} 10. Ne2 {-0.26/11 10} Ka8 {(Qc7-d6) -1.12/13 5} 11. Nc3 {
(c2-c3) -0.21/12 11} Nh5 {(Qc7-d6) -1.25/14 5} 12. Ne2 {-0.21/11 2} c5 {
(Ka8-b8) -1.09/14 8} 13. dxc5 {-0.25/11 8} Nxc5 {(Qc7xc5) -1.10/13 7} 14. Bg4 {
(Kb2-a1) -0.13/11 2} Nf6 {(Nh5-g7) -1.07/14 7} 15. Bh3 {(Ne2-d4) -0.13/11 9}
Qa5 {(Qc7-b6) -1.10/12 8} 16. Nd4 {-0.11/12 7} h6 {(Qa5-b4) -1.08/13 6} 17. Bf1
{(c2-c3) -0.03/11 8} a6 {-1.10/13 5} 18. Bd3 {(Bf1-h3) -0.11/11 8} Nh5 {
(Qa5-b4) -1.31/12 6} 19. Ne2 {(Kb2-b1) -0.28/11 6} b5 {(Qa5-d8) -1.25/12 8} 20.
c3 {-0.22/11 16} Qd8 {(b5-b4) -1.09/12 5} 21. Bc2 {(Rg1-h1) 0.00/10 4} Qc7 {
(Ka8-b7) -0.98/12 10} 22. Rd2 {(Rg1-e1) -0.03/10 6} Nf6 {(Ka8-b7) -0.99/13 5}
23. Nd4 {(Kb2-a1) 0.00/11 4} e5 {(Nf6-h5) -2.06/13 5} 24. f5 {
(Nd4-e2) -0.51/11 6} exd4 {-1.59/12 5} 25. Qxd4 {-0.47/9 0} Qd8 {
(Qc7-e5) -1.80/12 4} 26. fxg6 {-0.58/11 2} Ne6 {-1.70/14 7} 27. Qe3 {
(Qd4-b4) -0.46/13 6} Ng5 {-1.52/13 7} 28. Qd3 {-0.38/12 1} Nh3 {-1.24/13 6} 29.
gxf7 {-0.28/12 3} Nxg1 {-1.07/13 4} 30. fxe8=Q {(f7xe8B) -0.41/13 2} Qxe8 {
-0.88/15 5} 31. a4 {(Rd2-d1) -0.40/13 6} Rc7 {-1.26/12 4} 32. axb5 {-0.48/13 3}
Qxb5 {-0.99/13 5} 33. Qe3 {-0.46/13 3} Qc5 {-0.81/13 4} 34. Bd1 {
(Rd2-d1) -0.54/12 7} Qc6 {(a6-a5) -1.11/12 5} 35. Rd4 {-0.17/11 3} Nh3 {
(a6-a5) -0.91/12 5} 36. Be2 {(Bd1-c2) 0.00/12 6} Rb7 {(Nh3xf2) -0.84/13 6} 37.
Qxh6 {0.08/9 1} Rb8 {(a6-a5) -0.85/13 6} 38. Bf1 {(Qh6-c1) 0.00/11 5} Nxf2 {
(Nh3-g1) -0.47/13 5} 39. Qd2 {0.09/12 4} Nh1 {0.23/14 5} 40. Qh2 {
(Qd2-g2) 0.00/12 4} Rc8 {-0.33/13 4} 41. Rd3 {0.00/13 4} Nd7 {-1.27/12 4} 42.
Qd2 {(c3-c4) -0.03/12 3} Nc5 {-0.48/12 9} 43. Rxd5 {0.03/12 2} Nxb3 {-0.56/13 4
} 44. Kxb3 {0.11/13 2} Rb8+ {(Nh1-g3) -0.36/12 4} 45. Ka3 {(Kb3-c2) 1.27/12 4}
Ng3 {1.18/12 3} 46. Bd3 {1.32/11 1} Qb7 {1.09/12 4} 47. Qa2 {1.19/9 0} Rc8 {
1.53/11 4} 48. Bc4 {(Ka3-a4) 1.58/9 1} Qb6 {1.76/11 4} 49. Ka4 {1.92/11 9} Rh8
{2.00/11 7} 50. Bxa6 {(Rd5-d4) 1.95/11 1} Rxh4+ {2.11/11 4} 51. Bc4 {1.76/10 0}
Rh7 {2.18/12 2} 52. Rb5 {(Rd5-a5+) 2.09/12 1} Ra7+ {2.15/14 3} 53. Kb3 {
1.89/11 0} Qxb5+ {(Ra7xa2) 2.36/15 3} 54. Bxb5 {1.95/9 0} Rxa2 {2.46/18 1} 55.
Kxa2 {2.41/20 1} Kb7 {2.62/21 4} 56. Kb3 {2.81/21 5} Nf5 {(Ng3-h5) 2.62/21 4}
57. Kb4 {3.03/20 3} Kc7 {3.09/19 4} 58. Kc5 {3.04/19 1} Kd8 {3.20/21 4} 59. Bd3
{3.21/19 2} Ne7 {3.51/21 4} 60. Kd6 {(f3-f4) 3.73/20 2} Nc8+ {3.51/18 4} 61.
Ke6 {(Kd6-e5) 4.36/21 1} Nb6 {3.53/18 3} 62. Bb5 {(c3-c4) 5.16/20 2} Na8 {
(Kd8-c7) 10.38/18 9} 63. c4 {9.66/18 7} Nc7+ {10.34/19 4} 64. Kd6 {9.64/17 3}
Na8 {10.53/19 4} 65. f4 {9.64/15 3} Nb6 {10.72/17 3} 66. Ke6 {10.97/15 2} Nc8 {
13.70/17 6} 67. f5 {10.97/13 1} Ne7 {13.76/18 4} 68. c5 {(f5-f6) 14.75/15 2}
Ng8 {14.31/17 4} 69. f6 {14.75/13 1} Nh6 {14.31/17 4} 70. f7 {14.79/12 1} Nxf7
{#13/19 4} 71. Kxf7 {14.79/13 1} Kc8 {#10/22 2} 72. Ke6 {14.87/12 3} Kc7 {
#9/25 2} 73. c6 {(Ke6-e5) 14.91/10 2} Kb6 {14.09/12 7} 74. Kd6 {
(Ke6-d5) 15.17/9 2} Kxb5 {(Kb6-a5) 10.53/10 3} 75. c7 {35.35/8 7} Kc4 {
(Kb5-b6) 10.60/10 7} 76. Kc6 {(Kd6-e5) 11.33/8 8} Kd4 {10.43/9 2} 77. c8=Q {
11.45/8 10} Ke3 {10.53/9 2} 78. Kd5 {11.46/6 3} Kf4 {10.80/10 3} 79. Qh3 {
(Qc8-f8+) 12.05/6 3} Kg5 {10.39/4 0} 80. Ke5 {#6/4 0} Kg6 {10.41/4 0} 81. Qe6+
{(Qh3-f5+) #5/3 0} Kg7 {#4/63 0} 82. Qe8 {(Qe6-e7+) #4/3 0} Kh6 {#3/63 0} 83.
Kf6 {#3/3 0} Kh7 {#2/4 0} 84. Qg6+ {(Qe8-d7+) #2/3 0} Kh8 {#1/4 0} 85. Qg7# {
#1/3 0} 1-0