Fat Fritz update

[Damir]

When there is new engine or network improvement… the settings that once worked will no longer work. You will have to find new settings that will work with latest engine improvement. dkappe you should stop beeing lazy and try and tune the engine yourself. Don't expect others to do your work for you….

[Albert Silver]

I’ve heard rumors that CB got the tuning wrong. Hopefully an update is coming soon.

For the OpenCL build yes, but the CUDA is as tuned. Otherwise anyone can enter the new tuned values manually.

[corres]

The latest update is Fat Fritz 1.1 with changes to Cpuct, Fpu, Policy temp, NN cache and couple of others.
Tests are required to see where it stands.

Question is whether Kiudee settings work with Fat Fritz

if Fat Friz.exe is equivalent to lc0.exe as it was, yes.

[dkappe]

When there is new engine or network improvement… the settings that once worked will no longer work. You will have to find new settings that will work with latest engine improvement. dkappe you should stop beeing lazy and try and tune the engine yourself. Don't expect others to do your work for you….

Hmmm, not sure I understand the humor, wink, wink?

[mwyoung]

The latest update is Fat Fritz 1.1 with changes to Cpuct, Fpu, Policy temp, NN cache and couple of others.
Tests are required to see where it stands.

Here is how Fat Fritz 1.1 is performing currently against Stockfish. So far the results look very good for the Fat Fritz update!

Fat Fritz 1.1 vs Stockfish 040320 current standings +2 =48 -0 +14 Elo for Fat Fritz 1.1

Live results: https://www.youtube.com/watch?v=RffvjxgTaSw

[dkappe]

The Dorf-off nears its conclusion:

Code: Select all

   # PLAYER    :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 FFv471    :       7     16   161.5     307  52.6      81   51  221   35  72.0
   2 SF11      :       0   ----   298.0     613  48.6      91   76  444   93  72.4
   3 62627     :     -10     14   153.5     306  50.2     ---   42  223   41  72.9

[dkappe]

The Dorf-off is now complete. FF had a slight edge in the Najdorf with t60 underperforming.

Code: Select all

   # PLAYER    :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 FFv471    :      16     13   215.5     400  53.9      99   69  293   38  73.2
   2 SF11      :       0   ----   383.5     800  47.9      93   93  581  126  72.6
   3 62627     :      -9     13   201.0     400  50.2     ---   57  288   55  72.0

I’ll add some head to head in the Najdorf for good measure.

[dkappe]

It seems as if t60 doesn’t really get the pawn structure of the Najdorf.

[Event "?"] [Site "https://lichess.org/6sFidd5t"] [Date "2020.03.06"] [Round "2"] [White "62627"] [Black "FFv471"] [Result "0-1"] [WhiteElo "?"] [BlackElo "?"] [Variant "Standard"] [TimeControl "10800+1"] [ECO "B90"] [Opening "Sicilian Defense: Najdorf Variation"] [Termination "Normal"] [Annotator "lichess.org"] 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 { B90 Sicilian Defense: Najdorf Variation } 6. f3 e5 7. Nb3 Be7 8. Be3 Be6 9. Qd2 O-O 10. O-O-O Nbd7 11. g4 b5 12. Kb1 b4 13. Nd5 Bxd5 14. exd5 Nb6 15. Na5 Nbxd5 16. Nc6 Qc7 17. Nxe7+ Qxe7 18. Bg5 Qc7 19. Bxf6 Nxf6 20. g5 Nh5 21. Qxb4 Rfd8 22. Qc4 Qb6 23. Qe4 Rab8 24. b3 Rdc8 25. Bc4 g6 26. Rd2? { (-0.76 → -2.01) Mistake. Rhg1 was best. } (26. Rhg1 Nf4) 26... Qb4 27. Kc1? { (-1.76 → -3.61) Mistake. Bxf7+ was best. } (27. Bxf7+) 27... Qc3 28. Rdd1 Nf4 29. Rhe1 Rb4 30. Qe3 Rcxc4 31. bxc4 Qb2+ 32. Kd2 Rxc4 33. Qb3 Qd4+ 34. Kc1 Qa1+ 35. Qb1 Qc3 36. Qb3 Nd3+ 37. Rxd3 Qxe1+ 38. Rd1 Qh4 39. Rxd6? { (-2.70 → -4.64) Mistake. Rd2 was best. } (39. Rd2 Rc6 40. Qe3 Qc4 41. a3 Kg7 42. Qd3 Qc5 43. Rd1 d5 44. h4 Qf2 45. Rd2 Qe1+) 39... Qf4+ 40. Kd1?! { (-4.38 → -5.57) Inaccuracy. Rd2 was best. } (40. Rd2 Rd4) 40... Rb4 41. Qc3?! { (-5.20 → -8.32) Inaccuracy. Qa3 was best. } (41. Qa3) 41... Rb1+ 42. Ke2 Qxh2+ 43. Ke3 Qf4+ 44. Ke2 Qxg5 45. Rd5? { (-4.65 → -21.61) Mistake. Rxa6 was best. } (45. Rxa6) 45... Qg2+ 46. Ke3 Qg1+ 47. Ke2 Qf1+ 48. Ke3 Re1+ 49. Qxe1 Qxe1+ 50. Kd3 Qb1 51. Ra5 Qh1 52. c4 Qg1 53. c5 e4+ 54. Kc4?! { (-22.72 → Mate in 9) Checkmate is now unavoidable. fxe4 was best. } (54. fxe4 Qe1 55. Rxa6 Qf1+ 56. Kd4 Qxa6 57. a3 Qxa3 58. Kc4 Qa2+ 59. Kd4 Qe6 60. Kd3 Qe5) 54... exf3 55. Kd5 f2 56. c6 f1=Q 57. Kd6 Qa7 58. Rc5 Qa1 59. c7 Qb6+ 60. Kd7 Qe6+ 61. Kd8 Qaf6# { Black wins by checkmate. } 0-1

[bmp1974]

A small test between Lc0 (with Ls14 net) vs Fat fritz 1.1

Hardware: i7, 7700k, 4.2ghz turbo
GPU: Gtx 1060 ti
Book depth: 4 moves
Book: My custom book with games of 2700+ players from 2000 till 2019.

[bmp1974]

[Event "Fat fritz vs others, Blitz 3min+1sec"] [Site "?"] [Date "2020.03.07"] [Round "11.1"] [White "Fat Fritz (in lc0) cuda"] [Black "Lc0 v0.24.0-rc1+git.d76b8e2"] [Result "0-1"] [ECO "E17"] [Annotator "0.26;0.38"] [PlyCount "112"] [EventDate "2020.03.07"] [EventType "tourn"] [SourceTitle "Fritz Engine Tournament"] [Source "Prasanna"] [TimeControl "180+1"] {Intel(R) Core(TM) i7-7700 CPU @ 3.60GHz 3600 MHz W=10.4 plies; 4kN/s; 93 TBAs; Super duper_2013 book.ctg B=12.1 plies; 4kN/s; 298 TBAs; PowerBook2013.ctg} 1. c4 {B 0} e6 {B 0} 2. g3 {B 0} Nf6 {B 0} 3. Bg2 {B 0} d5 {B 0} 4. Nf3 {B 0} Be7 {B 0} 5. d4 {0.26/7 4} O-O {0.38/10 4} 6. Nc3 {0.24/8 4} Nbd7 {0.33/13 3} 7. b3 {0.26/9 3 (Qd3)} b6 {0.21/13 3} 8. O-O {0.25/9 5} Bb7 {0.20/13 3} 9. Bb2 { 0.22/9 5 (Bf4)} c5 {0.16/14 5} 10. Ne5 {0.20/11 6 (Ne1)} Rc8 {0.15/15 5} 11. cxd5 {0.22/12 2} Nxe5 {0.18/13 2} 12. dxe5 {0.25/13 4} Nxd5 {0.21/16 4} 13. Nb5 {0.31/12 2} Qd7 {0.24/14 3} 14. a4 {0.21/12 4} Rcd8 {0.18/17 5 (a6)} 15. Nxa7 { 0.26/13 6} Ba6 {0.11/19 3} 16. Nb5 {0.21/13 3} Bxb5 {0.06/17 3} 17. axb5 { 0.19/11 3} Qxb5 {0.02/15 2} 18. Qc2 {0.20/9 4} Nb4 {-0.06/14 7 (Rd7)} 19. Qc4 { 0.13/13 8} Qxc4 {-0.14/15 6} 20. bxc4 {0.03/13 3} Rd2 {-0.17/14 2} 21. Bc1 { 0.00/12 5} Rxe2 {-0.16/14 5} 22. Ra7 {-0.03/11 3} Re8 {-0.23/12 5} 23. Be3 { -0.13/11 15 (Rd1)} Rc2 {-0.26/12 9} 24. Rd1 {-0.15/12 2} Bf8 {-0.24/14 3} 25. Rb7 {-0.12/11 10} Rxc4 {-0.21/12 6} 26. Rdd7 {-0.10/10 1} Nc2 {-0.25/11 14} 27. Bf4 {-0.04/9 3 (Rxf7)} Rb4 {-0.15/10 7} 28. h4 {-0.03/9 5} Ne1 {-0.39/10 4} 29. h5 {-0.25/10 9} Nxg2 {-0.42/11 7} 30. Kxg2 {-0.31/12 1} h6 {-0.53/11 2 (Rc8)} 31. Bd2 {-0.40/9 9 (Rxf7)} Rb3 {-0.58/10 9} 32. Rxf7 {-0.35/10 3} Rc8 {-0.61/ 11 1} 33. Be3 {-0.36/11 3} b5 {-0.60/12 6} 34. g4 {-0.42/11 3} b4 {-1.14/11 5 (c4)} 35. g5 {-0.22/13 5} Rxe3 {-1.56/13 3 (c4)} 36. fxe3 {-0.84/11 5} hxg5 { -1.66/12 3} 37. Kf3 {-1.14/12 10} c4 {-1.67/12 3} 38. Kg4 {-1.26/12 3} c3 { -1.88/13 5} 39. Rbc7 {-1.37/14 5 (Rfc7)} Rxc7 {-1.85/14 4} 40. Rxc7 {-1.44/13 1 } Be7 {-2.03/12 4} 41. Rc8+ {-1.55/13 10} Kh7 {-2.15/12 4} 42. Rc4 {-1.59/10 1} Kh6 {-2.30/13 4 (g6)} 43. e4 {-2.25/10 8 (Rc8)} g6 {-2.55/13 5} 44. hxg6 { -2.33/10 1} Kxg6 {-2.71/11 3} 45. Rc6 {-2.42/10 4} Bd8 {-2.90/11 4} 46. Rxe6+ { -2.54/10 2} Kg7 {-3.09/11 5 (Kf7)} 47. Rc6 {-2.75/8 4} Ba5 {-3.26/10 2} 48. Kf5 {-3.23/8 7} b3 {-3.28/10 5} 49. e6 {-3.69/8 4 (Rg6+)} b2 {-5.40/8 6} 50. Rc8 { -4.64/7 2 (e7)} b1=Q {-10.75/8 5} 51. e7 {-5.33/7 2} Qb5+ {-7.72/9 7} 52. Ke6 { -5.35/7 5} Qa6+ {-10.21/8 2} 53. Kd7 {-6.41/6 3} Qb7+ {-12.44/7 4} 54. Kd6 { -7.73/5 3} Qxc8 {-26.69/6 3} 55. e8=Q {-9.08/5 2 (e8R)} Qxe8 {-43.66/6 3} 56. Kc5 {-70.82/5 2 (Kd5)} Qxe4 {-128.00/4 3 (Qh5)} 0-1

Sample game