Page 1 of 1
Fun Position
Posted: Sat Jun 12, 2010 1:58 pm
by Rebel
A few years back I composed the below position to test my engine. It can't find Kg2 with a winning score (say +3.xx). I am running SF 1.6 JA, it is currently at depth 59, score still 0.00.
Ed
[Event "?"]
[Site "?"]
[Date ""]
[Round "?"]
[White ""]
[Black ""]
[Result "1-0"]
[Setup "1"]
[FEN "4b1kn/3p1p1r/p2P1Ppr/p5Pp/p6P/p7/P7/7K w - - 0 1"]
1. Kg2 Rg7 2. Kf3 Kh7 3. Ke4 Rg8 4. Kd5 Rf8 5. Kc5 Kg8 6. Kb6 Rh7 7. Kxa6 Rh6 8. Kxa5 Kh7
9. Kxa4 Rg8 10. Kxa3 Rg7 11. Kb4 Kg8 12. a4 Kh7 13. a5 Rg8 14. a6 Rf8 15. a7 Rg8 16. a8Q
Rf8 17. Kb5 Kg8 18. Kb6 Rh7 19. Kc7 Rh6 20. Kd8 Rh7 21. Ke7 Rg7 22. Qa1 Rh7 23. Qd4 Rh6
24. gxh6 g5 25. h7+ Kxh7 26. Kxf8 Kg6 27. Kxe8 1-0
Re: Fun Position
Posted: Sat Jun 12, 2010 2:20 pm
by zullil
Rebel wrote:A few years back I composed the below position to test my engine. It can't find Kg2 with a winning score (say +3.xx). I am running SF 1.6 JA, it is currently at depth 59, score still 0.00.
Ed
[Event "?"]
[Site "?"]
[Date ""]
[Round "?"]
[White ""]
[Black ""]
[Result "1-0"]
[Setup "1"]
[FEN "4b1kn/3p1p1r/p2P1Ppr/p5Pp/p6P/p7/P7/7K w - - 0 1"]
1. Kg2 Rg7 2. Kf3 Kh7 3. Ke4 Rg8 4. Kd5 Rf8 5. Kc5 Kg8 6. Kb6 Rh7 7. Kxa6 Rh6 8. Kxa5 Kh7
9. Kxa4 Rg8 10. Kxa3 Rg7 11. Kb4 Kg8 12. a4 Kh7 13. a5 Rg8 14. a6 Rf8 15. a7 Rg8 16. a8Q
Rf8 17. Kb5 Kg8 18. Kb6 Rh7 19. Kc7 Rh6 20. Kd8 Rh7 21. Ke7 Rg7 22. Qa1 Rh7 23. Qd4 Rh6
24. gxh6 g5 25. h7+ Kxh7 26. Kxf8 Kg6 27. Kxe8 1-0
4b1kn/3p1p1r/p2P1Ppr/p5Pp/p6P/p7/P7/7K w - - 0 1
Re: Fun Position
Posted: Sat Jun 12, 2010 2:43 pm
by Chris Whittington
Rebel wrote:A few years back I composed the below position to test my engine. It can't find Kg2 with a winning score (say +3.xx). I am running SF 1.6 JA, it is currently at depth 59, score still 0.00.
Ed
[Event "?"]
[Site "?"]
[Date ""]
[Round "?"]
[White ""]
[Black ""]
[Result "1-0"]
[Setup "1"]
[FEN "4b1kn/3p1p1r/p2P1Ppr/p5Pp/p6P/p7/P7/7K w - - 0 1"]
1. Kg2 Rg7 2. Kf3 Kh7 3. Ke4 Rg8 4. Kd5 Rf8 5. Kc5 Kg8 6. Kb6 Rh7 7. Kxa6 Rh6 8. Kxa5 Kh7
9. Kxa4 Rg8 10. Kxa3 Rg7 11. Kb4 Kg8 12. a4 Kh7 13. a5 Rg8 14. a6 Rf8 15. a7 Rg8 16. a8Q
Rf8 17. Kb5 Kg8 18. Kb6 Rh7 19. Kc7 Rh6 20. Kd8 Rh7 21. Ke7 Rg7 22. Qa1 Rh7 23. Qd4 Rh6
24. gxh6 g5 25. h7+ Kxh7 26. Kxf8 Kg6 27. Kxe8 1-0
programmer attempts at solution ....
remove q-search
remove captures from the move sort
increase possession of passed pawn bonus to something ridiculously large
might speed up a conclusion, but, whoops - all lead to less 'ELO' after 1 second of testing 60,000 games. junk idea.
slightly more clever solution ..... (not available to bean-counters)
introduce a new algorithm that counts number of discrete available positions reached in search for one side (I think black has only 9 variants or so)
conclude something is wrong somehow, but then what?
a strong human player would say "what can black do" and is therefore lost for black
bean-counting hill-climbers of sub-optimal hills mistake ELO (win rate against similar machine opponents) for strength. their methodology allocates lots (way too much) of ELO for insignificantly unimportant win rate improvements. They can't measure certain strength increases at all and they can even end up giving negative ELO to actual, real improvements. The match between their ELO and reality is completely non-linear, can correlate negatively under some conditions and can represent changes in strength as zero ELO change and vice versa.
Re: Fun Position
Posted: Sat Jun 12, 2010 4:53 pm
by Robert Flesher
Chris Whittington wrote:Rebel wrote:A few years back I composed the below position to test my engine. It can't find Kg2 with a winning score (say +3.xx). I am running SF 1.6 JA, it is currently at depth 59, score still 0.00.
Ed
[Event "?"]
[Site "?"]
[Date ""]
[Round "?"]
[White ""]
[Black ""]
[Result "1-0"]
[Setup "1"]
[FEN "4b1kn/3p1p1r/p2P1Ppr/p5Pp/p6P/p7/P7/7K w - - 0 1"]
1. Kg2 Rg7 2. Kf3 Kh7 3. Ke4 Rg8 4. Kd5 Rf8 5. Kc5 Kg8 6. Kb6 Rh7 7. Kxa6 Rh6 8. Kxa5 Kh7
9. Kxa4 Rg8 10. Kxa3 Rg7 11. Kb4 Kg8 12. a4 Kh7 13. a5 Rg8 14. a6 Rf8 15. a7 Rg8 16. a8Q
Rf8 17. Kb5 Kg8 18. Kb6 Rh7 19. Kc7 Rh6 20. Kd8 Rh7 21. Ke7 Rg7 22. Qa1 Rh7 23. Qd4 Rh6
24. gxh6 g5 25. h7+ Kxh7 26. Kxf8 Kg6 27. Kxe8 1-0
programmer attempts at solution ....
remove q-search
remove captures from the move sort
increase possession of passed pawn bonus to something ridiculously large
might speed up a conclusion, but, whoops - all lead to less 'ELO' after 1 second of testing 60,000 games. junk idea.
slightly more clever solution ..... (not available to bean-counters)
introduce a new algorithm that counts number of discrete available positions reached in search for one side (I think black has only 9 variants or so)
conclude something is wrong somehow, but then what?
a strong human player would say "what can black do" and is therefore lost for black
bean-counting hill-climbers of sub-optimal hills mistake ELO (win rate against similar machine opponents) for strength. their methodology allocates lots (way too much) of ELO for insignificantly unimportant win rate improvements. They can't measure certain strength increases at all and they can even end up giving negative ELO to actual, real improvements. The match between their ELO and reality is completely non-linear, can correlate negatively under some conditions and can represent changes in strength as zero ELO change and vice versa.
Great position Ed ! I think I remember the last time you posted this, but maybe not. I am getting old!
Re: Fun Position
Posted: Sat Jun 12, 2010 5:15 pm
by Mincho Georgiev
Code: Select all
Pawny 0.1.9:
setboard 4b1kn/3p1p1r/p2P1Ppr/p5Pp/p6P/p7/P7/7K w - - 0 1
go
1 3 -13.67 g5h6 h7h6
2 6 -13.67 g5h6 h7h6
2 24 -13.63 h1h2 g8f8 g5h6 h7h6
2 36 -13.57 h1g2 g8f8 g5h6 h7h6
3 15 -13.57 h1g2 g8f8 g5h6 h7h6
4 110 -11.74 h1g2 g8f8 g2h1 h7g7 f6g7
5 79 -11.74 h1g2 g8f8 g2h1 h7g7 f6g7
6 123 -10.55 h1g2 g8f8 g2h1 h7g7 f6g7 f8g7 g5h6 g7h6
7 239 -10.38 h1g2 g8f8 g2h1 h7g7 f6g7 f8g7 g5h6 g7h6 h1g2
8 348 -10.58 h1g2 g8f8 g2h1 h7g7 f6g7 f8g7 g5h6 g7h6 h1g2 h6g7
9 1829 -10.44 h1g2 h7g7 f6g7 h6h7 g7h8Q g8h8 g2f3 h8g7 f3e4 f7f5 g5f6 g7f6
10 1466 -10.44 h1g2 h7g7 f6g7 h6h7 g7h8Q g8h8 g2f3 h8g7 f3e4 f7f5 g5f6 g7f6 e4d4
11 2094 -10.45 h1g2 h7g7 f6g7 g8g7 g5h6
14 9417 -11.56 h1g2 h7g7 f6g7 g8g7 g5h6 g7h6 g2f3 f7f5 f3e3 h8f7 e3d4 f7d6
15 8547 -11.75 h1g2 h7g7 f6g7 g8g7 g5h6 g7h6 g2f3 f7f5 f3e3 h8f7 e3f4 f7d6 f4e5
16 19488 -11.92 h1g2 h7g7 f6g7 g8g7 g5h6 g7h6 g2f3 f7f5 f3e3 h8f7 e3d4 f7d6 d4c5 d6e4 c5b6 g6g5 h4g5 h6g5 b6a6
17 324961 -12.64 h1g2 h7g7 g5h6 g7h7 g2f3 h7h6 f3e4 h6h7 e4d4 g6g5 h4g5
20 163299 -12.84 h1g2 h7g7
21 547563 -12.65 h1g2 h7g7 g2f3 h6h7 f3e3
22 112887 -12.45 h1g2 h7g7
23 75430 -12.25 h1g2 h7g7
24 63459 -12.05 h1g2 h7g7
25 74774 -11.85 h1g2 h7g7
26 97421 -11.65 h1g2 h7g7
27 196133 -11.45 h1g2 h7g7
28 176008 -11.25 h1g2 h7g7
29 411999 -11.05 h1g2 h7g7
30 291360 -10.85 h1g2 h7g7
31 467358 -10.65 h1g2 h7g7
32 7423395 -10.46 h1g2 h7g7 g2f3 h6h7 f3e3 g8f8 e3d4 h7h6 d4c5 f8g8 c5b6
33 5894858 -10.30 h1g2 h7g7 g2f3 g8h7 f3e3 g7g8 e3d4 g8f8 d4c5 h7g8 c5b6 h6h7 b6a6 h7g7 a6a5 g8h7 a5a4
34 17783475 -10.11 h1g2 h7g7 g2f3 h6h7 f3e3
690 Kn/s, 50.00 sec
move h1g2
It could be caused by a lot of thing and as Chris Whittington suggests, you probably should isolate search add-ons one at a time
until you find it.
Re: Fun Position
Posted: Sun Jun 13, 2010 12:56 pm
by Ignacio
Illegal but nice position. It´s mate in 34 and after 14... Rf8 Chest says mate in 20.
Search for 1. Special-Mate [C0/R0/K5/P3/X5] in 126 ... (Hash=600MB)
20 03:49 161.443.699 811.323 +M20 a6a7
Search completed ... (Time=229.10s)
Mate in 20 found ! (1. Solution in 03:49)
20/20 03:49 185.666.655 810.431 +M20 a6a7 Rf8g8 a7a8Q Rg8f8 Qa8b8 Rf8g8 Qb8b5 Rg8f8 Qb5e5 Rf8g8 Kb4b5 Rg8f8 Kb5b6 Rf8g8 Kb6c7 Rg8f8 Kc7d8 Rf8g8 Qe5xe8 Rg8xe8+ Kd8xe8 Kh7g8 Ke8xd7 Rh6h7 Kd7e7 Rh7g7 f6xg7 f7f6 g7xh8Q+ Kg8xh8 d6d7 f6xg5 d7d8Q+ Kh8h7 Qd8f8 g5g4 Ke7f7 g6g5 Qf8g7+
Best,
Ignacio.