Moderator: Ras
Thanks Dann.Dann Corbit wrote:Here is a 32 bit Windows compile:nepossiver wrote:Is there a w32 compile? And I don't want to start a flame war, but if this is a modified SF, shouldn't the sources be made available as well?
http://cap.connx.com/chess-engines/new- ... 32.exe.bz2
Note that if you want to use the changes in:
http://cap.connx.com/chess-engines/new-approach/sfc.zip
to build your own version, you will need to define macro SMOOTH_SCALING
Usually, null move is just a big giant drop-off (or two or three drop-offs). This version scales null move more smoothly (and can almost certainly be improved).SzG wrote:What does it mean? Could you explain please?Dann Corbit wrote:This version of Stockfish:
http://cap.connx.com/chess-engines/new- ... x64_ja.rar
Scales null move smoothly.
Stockfish without smooth scaling. I also added a change to hash look up verses approximate eval around the same point in the code.Relative to what?Dann Corbit wrote: I get about +150 Elo so far in my testing.
Hi Dan,Dann Corbit wrote:I sent the sources to Tord and Jim Ablett. They may want to experiment a bit before it is released. The curve I used was just an eyeball crude approximation and so it probably could be made a lot better.nepossiver wrote:Is there a w32 compile? And I don't want to start a flame war, but if this is a modified SF, shouldn't the sources be made available as well?
Here is my MS 64 bit compile (the other one is Jim's and I have actually been testing my build and so I do not know if they behave differently):
http://cap.connx.com/chess-engines/new- ... sh.exe.bz2
I will make a 32 bit compile also.
Dann's for sure.Werner wrote:Hi Dan,Dann Corbit wrote:I sent the sources to Tord and Jim Ablett. They may want to experiment a bit before it is released. The curve I used was just an eyeball crude approximation and so it probably could be made a lot better.nepossiver wrote:Is there a w32 compile? And I don't want to start a flame war, but if this is a modified SF, shouldn't the sources be made available as well?
Here is my MS 64 bit compile (the other one is Jim's and I have actually been testing my build and so I do not know if they behave differently):
http://cap.connx.com/chess-engines/new- ... sh.exe.bz2
I will make a 32 bit compile also.
there is a difference in your 64bit compile and in that of Jim. Did you use the same sources for your compiles?
And which version should we use?
Using your smooth Stockfish of 32 bit:Dann Corbit wrote:This version of Stockfish:
http://cap.connx.com/chess-engines/new- ... x64_ja.rar
Scales null move smoothly. I get about +150 Elo so far in my testing. How about yours?
Code: Select all
Stockfish 1.6s 32bit ½½½1111½½0 6.5/10
Stockfish 1.6 JA 32bit ½½½0000½½1 3.5/10
Code: Select all
Stockfish 1.6s 32bit ½½11½½½111010½111½½½½½½110½11½ 20.0/30
Stockfish 1.6 JA 32bit ½½00½½½000101½000½½½½½½001½00½ 10.0/30
Code: Select all
Stockfish 1.6s 32bit 26.5/40
Stockfish 1.6 JA 32bit 13.5/40
Doesn't that make the release illegal until the source is made available?Dann Corbit wrote:I sent the sources to Tord and Jim Ablett. They may want to experiment a bit before it is released. The curve I used was just an eyeball crude approximation and so it probably could be made a lot better.nepossiver wrote:Is there a w32 compile? And I don't want to start a flame war, but if this is a modified SF, shouldn't the sources be made available as well?
Here is my MS 64 bit compile (the other one is Jim's and I have actually been testing my build and so I do not know if they behave differently):
http://cap.connx.com/chess-engines/new- ... sh.exe.bz2
I will make a 32 bit compile also.
George Tsavdaris wrote:Dann's for sure.Werner wrote:Hi Dan,Dann Corbit wrote:I sent the sources to Tord and Jim Ablett. They may want to experiment a bit before it is released. The curve I used was just an eyeball crude approximation and so it probably could be made a lot better.nepossiver wrote:Is there a w32 compile? And I don't want to start a flame war, but if this is a modified SF, shouldn't the sources be made available as well?
Here is my MS 64 bit compile (the other one is Jim's and I have actually been testing my build and so I do not know if they behave differently):
http://cap.connx.com/chess-engines/new- ... sh.exe.bz2
I will make a 32 bit compile also.
there is a difference in your 64bit compile and in that of Jim. Did you use the same sources for your compiles?
And which version should we use?
I think he forgot to tell Jim about enabling some options to make the compile so Jim's compile is not how it was supposed to be.
In my computer i get in most positions +2 ply in the same time for Dann's S1.6 compared to normal S1.6 and with comparable evaluations(but not the same and this is that puzzles me).
There are too few games and of course self-play is not the best measure for elo measurements, but 54.44% is still almost 30 elo extra if the error margins are not taken into account, which would not be bad at allMartin T wrote:Am I doing something wrong here?
Using http://cap.connx.com/chess-engines/new- ... sh.exe.bz2
1 min + 1 sec time control
After 44 games:
20.5 Stockfish 1.6 64-bit JA 2CPU (the "official" version of Stockfish)
23.5 Stockfish above 64-bit 2CPU
The new one scored 54.44%, that is what, 20 elo?
Best Regards,
Martin
I don't think much more can be achieved from a more aggressive/better tuned nullmove, just based on general experiences but if this holds it would be very good. Gian-Carlo got negative numbers so it is too early to tell I think Martin 
The move numbers below are not correct because Shredder GUI does not pick up that the game should start at move nine, but this is what I get with Rainbow Serpent now, I had similar problem resolving that 33... Rc8 is bad and you simply can overstep the time limit resolving it. Maybe in case of "Problem" more time should be allocated, "Problem" is a variable which is set when the value of the first move drops a lot compared to the previous iteration. But if the move is not resolved yet the search may still be aborted before "Problem" is set in root-search. You can have one or two of these moves in a game but if you do not complete the calculation and just play the -probably bad- move, you probably are lost instantly! So resolving these fail lows is important but can take up valuable time.George Tsavdaris wrote:
In the first match in the last game(a very exciting game!) Dann's Stockfish would not lose had it found the easy for its brother move 33...Rxd7! , a move that would draw.
I wonder why it does not play 33...Rxd7 "easily" and needs 16 ply to find it while it's brother only 14.
[d]1k1r4/pp1R4/bb6/8/Q1np2PP/5P2/P5K1/8 b - - 0 33
The game:
[Event "S16vsDann16"]
[Site "?"]
[Date "2009.12.29"]
[Round "10.1"]
[White "Stockfish 1.6 JA"]
[Black "Stockfish 1.6s"]
[Result "1-0"]
[Annotator "0.84;0.68"]
[SetUp "1"]
[FEN "2kr3r/pp2bppp/2np2n1/8/4P1b1/2P2N2/PPQ2PPP/R3K2R w KQ - 0 9"]
[PlyCount "83"]
[EventDate "2009.12.29"]
9. Nd4 {0.84/15 8} Nxd4 {0.68/15 7} 10. cxd4+ {0.76/12 1} Kb8 {0.76/
16 10} 11. O-O {0.76/14 10} Rc8 {0.76/15 10} 12. Qd2 {0.88/15 10} Bd7 {0.92/15
8} 13. Rfe1 {(f2-f4) 1.01/15 8} f5 {(Rh8-d8) 0.92/14 10} 14. exf5 {0.92/14 9}
Bxf5 {0.72/12 1} 15. Qa5 {0.76/15 10} Rhf8 {0.52/15 6} 16. Rac1 {(g2-g3) 0.76/
15 10} Rxc1 {0.68/14 9} 17. Rxc1 {0.88/13 1} Bd8 {0.68/16 9} 18. Qd5 {(Qa5-b4)
0.80/15 13} Bc7 {0.92/17 14} 19. Re1 {0.96/14 5} Nf4 {0.72/15 7} 20. Qb3 {0.92/
14 8} Bd3 {(Bf5-c8) 0.76/15 23} 21. f3 {(Re1-e7) 1.21/15 16} Ba6 {(b7-b5) 1.17/
14 6} 22. Qa4 {1.45/15 7} Rd8 {1.45/15 5} 23. Re8 {1.29/15 6} Kc8 {1.37/17 8}
24. Re7 {1.17/14 5} Bb6 {1.53/17 10} 25. g3 {1.61/16 10} Nd3 {1.49/15 4} 26.
Rxg7 {1.49/15 5} Kb8 {1.45/16 11} 27. Rxh7 {(Kg1-h1) 1.81/13 9} Nxb2 {1.41/15 7
} 28. Qb3 {1.45/13 4} Bxd4+ {1.41/15 6} 29. Kg2 {1.37/14 6} Nc4 {1.57/15 27}
30. g4 {1.57/13 9} Bb6 {(Rd8-f8) 1.73/13 8} 31. h4 {1.33/13 15} d5 {(Nc4-e3+)
1.29/14 7} 32. Qa4 {0.52/14 14} d4 {(Kb8-c8) 1.17/12 8} 33. Rd7 {0.00/14 7} Rc8
{(Rd8xd7) 1.29/12 4} 34. h5 {1.21/12 8} Bc5 {(Nc4-e3+) 1.13/14 10} 35. Qb3 {1.
41/13 8} Nb6 {1.29/14 6} 36. Rh7 {1.21/12 4} Bc4 {1.41/12 8} 37. Qb1 {(Qb3-d1)
1.49/12 4} Bd5 {1.41/12 5} 38. Qf5 {(g4-g5) 1.53/12 10} Bd6 {1.93/12 4} 39. h6
{(g4-g5) 2.78/12 7} Rf8 {(Bd5xa2) 1.77/12 7} 40. Qxf8+ {(Qf5-d3) 5.33/12 3}
Bxf8 {-3.03/8 0} 41. Rh8 {(g4-g5) 5.73/14 4} Nd7 {5.45/14 3} 42. h7 {5.73/13 4}
Kc7 {5.69/15 5} 43. Rxf8 {5.25/9 0} Nxf8 {5.41/8 0} 44. h8=Q {5.73/8 0} Ne6 {
5.65/15 3} 45. g5 {5.77/13 3} Bxa2 {(Kc7-d7) 6.22/13 7} 46. g6 {6.34/13 5} b5 {
(a7-a5) 6.38/12 2} 47. Qe5+ {7.51/14 5} Kc6 {7.23/14 2} 48. g7 {7.43/15 3} Nxg7
{(Ne6-c7) 7.31/14 4} 49. Qxg7 {8.28/16 4} d3 {(a7-a5) 8.08/14 2} 50. Qh6+ {
(Qg7-g6+) 8.40/16 3} 1-0