Novelty in the Sicilian Najdorf and interesting plan of Lc0 AF228

[Javier Ros]

I am doing a test with Lc0 32472, AF125 and AF228 against Stockfish 10, and some nice games are showing up.

The BalsaTop_25 test suite has been used and the Sicilian Najdorf has provided interesting games.
Lc0 on 1060x2, SF10 on i7 980X 6 cores, at 5 minutes plus 1 sec.

AF228 played the novelty 12...Kh8 with an original plan to attack Kingside that defeated SF10.


[pgn][Event "BalsaAFx2_5p1"] [Site "J-PC"] [Date "2019.01.26"] [Round "17"] [White "Stockfish_10_x64_popcnt"] [Black "AF228x2"] [Result "0-1"] [ECO "B92"] [WhiteElo "2200"] [BlackElo "2200"] [PlyCount "141"] [EventDate "2019.??.??"] [TimeControl "300+1"] 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be2 e5 7. Nb3 Be7 { -0.25/8 7} 8. O-O {0.49/25 19} O-O {-0.06/8 6} 9. Be3 {0.23/24 6} Be6 {-0.03/9 5} 10. h3 {0.26/25 8} Nbd7 {0.07/8 8} 11. a4 {0.11/24 4} Rc8 {0.03/8 6} 12. Bf3 {0.33/23 2} Kh8 {0.42/8 8} 13. a5 {0.69/23 3} Rg8 {0.51/11 14} 14. Kh2 { 0.33/27 32} g5 {0.77/9 3} 15. g4 {0.00/28 21} Nf8 {0.71/11 5} 16. Rg1 {-0.09/ 27 18} Ng6 {0.84/11 5} 17. Nd5 {-0.19/27 15} Nf4 {0.94/11 4} 18. c3 {0.00/24 6} Rg6 {1.17/10 9} 19. Bxf4 {0.00/26 12} gxf4 {1.34/9 7} 20. Nd2 {0.00/27 2} Qf8 { 1.33/8 11} 21. Qe2 {-0.20/26 18} Bd8 {1.39/8 10} 22. Nc4 {-0.29/25 6} Rh6 { 2.11/8 8} 23. Ncb6 {-0.23/21 3} Bxb6 {2.61/10 4} 24. Nxb6 {-0.30/24 5} Rd8 { 2.68/10 8} 25. Rh1 {-0.93/26 13} Rh4 {2.22/10 7} 26. Kg2 {-0.81/28 13} h5 { 1.88/11 10} 27. Rag1 {-1.24/26 4} Qh6 {1.75/12 6} 28. Kf1 {-0.57/21 2} Rg8 { 1.78/13 3} 29. gxh5 {-0.47/22 2} Rxg1+ {1.86/12 5} 30. Rxg1 {-1.09/25 9} Rxh3 { 1.90/11 2} 31. Qd1 {-1.06/25 5} Qf8 {1.87/11 7} 32. c4 {-1.30/28 23} Qe7 { 2.39/10 12} 33. b3 {-1.66/27 25} Kh7 {2.62/10 14} 34. b4 {-1.65/24 6} Kh6 { 4.08/9 4} 35. Ke2 {-2.98/26 13} Nh7 {3.86/11 6} 36. Rh1 {-3.49/24 5} Rxh1 { 4.37/12 5} 37. Qxh1 {-3.48/22 0} Ng5 {4.34/11 3} 38. Kd3 {-3.95/27 11} Nh3 { 5.15/11 7} 39. Qg2 {-4.04/26 3} Qh4 {6.25/11 5} 40. Nd5 {-3.92/23 0} Nxf2+ { 7.71/11 8} 41. Kd2 {-4.35/25 5} Bg4 {8.78/11 6} 42. b5 {-4.89/23 3} Qg3 { 11.00/12 7} 43. Qxg3 {-4.92/21 0} fxg3 {10.85/12 4} 44. Ke3 {-5.44/24 2} Kxh5 { 11.92/12 5} 45. Bg2 {-6.09/26 1} Kh4 {13.59/11 16} 46. Ne7 {-6.41/24 1} Nd1+ { 14.95/13 2} 47. Kd2 {-6.64/25 0} Nb2 {18.81/11 3} 48. Kc3 {-6.96/25 1} Na4+ { 20.25/11 4} 49. Kd2 {-7.25/24 0} Nc5 {25.85/10 5} 50. Ke1 {-7.36/22 1} Nb3 { 31.94/9 10} 51. bxa6 {-7.65/21 0} bxa6 {61.42/9 2} 52. Nc6 {-7.99/21 1} Be6 { 57.84/8 7} 53. c5 {-7.73/22 0} Nxc5 {66.45/7 2} 54. Nb4 {-8.15/23 1} Bc4 { 60.04/7 4} 55. Kd2 {-8.50/22 0} Nb3+ {54.72/7 3} 56. Kc3 {-8.69/23 1} Nxa5 { 60.95/8 1} 57. Nc2 {-8.86/22 0} Be6 {77.03/7 3} 58. Nb4 {-8.85/24 0} Bc8 { 79.77/7 3} 59. Nc2 {-8.93/22 0} Kg5 {75.60/7 2} 60. Nb4 {-8.97/24 1} Kf4 { 74.28/7 2} 61. Nd5+ {-8.41/19 0} Kg5 {62.59/7 2} 62. Nc7 {-8.77/23 1} Kf4 { 55.46/7 2} 63. Nd5+ {-9.02/24 0} Kg4 {50.28/8 1} 64. Ne3+ {-8.75/21 0} Kg5 { 64.23/7 1} 65. Kb4 {-9.22/21 1} Nb7 {68.66/7 2} 66. Nd5 {-8.75/22 0} Be6 { 77.17/7 1} 67. Nc3 {-9.02/22 0} Kf4 {71.03/7 1} 68. Ne2+ {-8.95/20 0} Ke3 { 71.22/7 1} 69. Nxg3 {-11.24/23 1} Kf2 {72.55/7 1} 70. Nh5 {-14.86/22 1} Kxg2 { 71.54/7 0} 71. Kc3 {adjudication -13.93/21 1 Arena Adjudication} 0-1 [/pgn]

In the same game SF10-AF125, AF125 played 12..g6 with a plan to mobilize the bad Black bishop successfully used by Lc0 in a previous game from my tests. But this time the game was draw.

[pgn][Event "BalsaAFx2_5p1"] [Site "J-PC"] [Date "2019.01.26"] [Round "17"] [White "Stockfish_10_x64_popcnt"] [Black "AF125x2"] [Result "1/2-1/2"] [ECO "B92"] [WhiteElo "2200"] [BlackElo "2200"] [PlyCount "92"] [EventDate "2019.??.??"] [TimeControl "300+1"] 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be2 e5 7. Nb3 Be7 { -0.27/8 7} 8. O-O {0.38/25 15} O-O {-0.15/8 7} 9. Be3 {0.44/24 2} Be6 {-0.20/9 4} 10. Qd3 {0.41/25 7} Nbd7 {-0.14/10 7} 11. Nd5 {0.47/23 2} Bxd5 {-0.11/9 4} 12. exd5 {0.43/24 8} g6 {-0.03/9 8} 13. Bh6 {0.79/22 5} Re8 {0.02/8 7} 14. Rfd1 {0.69/24 21} Bf8 {0.24/8 10} 15. Bg5 {0.54/26 18} h6 {0.26/9 9} 16. Be3 { 0.72/23 3} Rc8 {0.29/10 6} 17. c4 {0.37/26 28} h5 {0.33/9 7} 18. h3 {0.54/21 3} a5 {0.40/8 6} 19. Nd2 {0.66/23 7} b6 {0.44/9 15} 20. Rab1 {0.51/25 5} Bg7 { 0.58/8 5} 21. Bg5 {0.30/27 21} Qc7 {0.50/8 5} 22. Qc2 {0.24/27 14} Nh7 { 0.42/9 8} 23. Be3 {0.52/23 3} f5 {0.45/10 3} 24. Nf3 {0.46/24 5} Qd8 {0.33/11 9 } 25. a3 {0.60/25 11} Qf6 {0.26/10 10} 26. b4 {0.25/25 11} axb4 {0.29/11 3} 27. axb4 {0.96/25 2} g5 {0.21/13 4} 28. Nd2 {1.37/23 4} g4 {0.02/14 5} 29. hxg4 { 0.68/25 14} hxg4 {-0.06/15 4} 30. c5 {0.65/25 5} bxc5 {-0.17/14 12} 31. bxc5 { 0.94/23 1} Rxc5 {-0.07/14 14} 32. Bxc5 {0.09/26 28} Nxc5 {-0.09/16 0} 33. Rb5 { 0.09/27 11} e4 {-0.08/14 5} 34. Nf1 {0.12/28 5} Qf8 {0.19/11 9} 35. Ne3 { 0.09/26 1} Ng5 {0.23/14 6} 36. Rxc5 {0.28/23 1} dxc5 {0.17/16 2} 37. d6 { 0.09/28 6} Bd4 {0.09/16 4} 38. Qc4+ {0.09/28 3} Kh7 {-0.13/15 12} 39. d7 { 0.00/30 8} Rd8 {-0.17/16 3} 40. Qd5 {0.00/32 1} g3 {-0.14/15 14} 41. Nxf5 { 0.00/32 1} Qf6 {-0.11/14 4} 42. Kf1 {0.00/32 4} Kg6 {-0.06/11 7} 43. Nh4+ { 0.00/30 2} Kh7 {-0.04/10 7} 44. Nf5 {0.00/32 2} Kg6 {-0.02/7 6} 45. Bg4 { 0.00/33 3} Qa6+ {0.00/4 6} 46. Be2 {0.00/37 0} Qf6 {normal 0.00/2 6 3-fold repetition} 1/2-1/2 [/pgn]

Finally the same game SF10-Lc0_32742 was a victory for SF after a tactical blunder of Lc0 42..Qh6?? in an even position.

[pgn][Event "BalsaAFx2_5p1"] [Site "J-PC"] [Date "2019.01.26"] [Round "17"] [White "Stockfish_10_x64_popcnt"] [Black "Lc0201_32742x2"] [Result "1-0"] [ECO "B92"] [WhiteElo "2200"] [BlackElo "2200"] [PlyCount "94"] [EventDate "2019.??.??"] [TimeControl "300+1"] 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be2 e5 7. Nb3 Be7 { -0.28/9 7} 8. O-O {0.46/27 29} O-O {-0.34/10 8} 9. Be3 {0.33/27 9} Be6 { -0.35/11 3} 10. Qd3 {0.50/26 9} Nbd7 {-0.33/11 6} 11. Nd5 {0.51/25 3} Bxd5 { -0.34/11 4} 12. exd5 {0.45/25 4} Ne8 {-0.31/10 6} 13. a4 {0.33/25 14} Bg5 { -0.34/10 14} 14. a5 {0.50/24 2} Rc8 {-0.38/11 5} 15. c4 {0.45/24 3} f5 { -0.39/13 5} 16. Qc3 {0.46/24 10} Bxe3 {-0.37/10 8} 17. Qxe3 {0.42/26 4} Nef6 { -0.38/10 5} 18. Nd2 {0.42/24 5} Qe8 {-0.46/9 11} 19. b4 {0.46/23 4} Qg6 { -0.47/9 6} 20. Rae1 {0.70/23 9} h5 {-0.30/8 9} 21. Nf3 {1.10/22 5} f4 {-0.23/ 13 6} 22. Nh4 {0.00/27 6} Qh7 {-0.11/16 5} 23. Qd2 {0.00/28 6} g5 {-0.10/15 3} 24. Nf3 {0.00/30 3} Qg6 {-0.06/16 1} 25. h4 {0.00/30 5} gxh4 {-0.01/16 8} 26. Nxh4 {0.00/33 10} Qg5 {-0.01/15 1} 27. Nf3 {0.00/36 4} Qg7 {0.00/13 12} 28. Bd3 {0.79/23 2} Rfe8 {0.12/15 8} 29. Qc2 {0.13/27 11} Kh8 {0.15/15 8} 30. Bf5 { 0.09/30 8} e4 {-0.03/19 13} 31. Bxe4 {0.00/30 6} Rg8 {-0.17/22 3} 32. Nh4 { 0.00/28 4} Qg5 {-0.30/21 5} 33. Ng6+ {0.00/29 4} Rxg6 {-0.32/21 1} 34. Bxg6 { 0.00/33 5} Rg8 {-0.43/19 6} 35. Be4 {0.00/33 4} Ne5 {-0.50/18 5} 36. f3 { 1.02/26 3} Nxc4 {-0.52/17 5} 37. Rf2 {1.25/24 4} Ne3 {-0.18/14 15} 38. Qb2 { 1.02/25 9} h4 {-0.23/13 14} 39. Rc1 {0.59/26 13} Re8 {-0.23/16 6} 40. Kh1 { 0.86/25 5} Kg7 {-0.08/12 7} 41. b5 {0.29/28 20} axb5 {-0.14/11 7} 42. Rc7+ { 0.22/26 2} Kg8 {0.01/12 7} 43. Re2 {0.22/27 12} Qh6 {-0.18/15 8} ({Stockfish 10 64 BMI2:} 43... Nxe4 44. fxe4 h3 45. gxh3 Qg3 46. Rg2 Qxg2+ 47. Qxg2+ Nxg2 48. Kxg2 Rxe4 49. Rxb7 Ra4 50. Rxb5 Kf7 51. Kf3 Kf6 52. Kg4 Ke5 53. Rb8 Rxa5 54. Re8+ Kxd5 55. h4 Ra4 56. Kf3 Ra1 57. Kxf4 Rf1+ 58. Kg5 Rg1+ 59. Kf5 Rf1+ 60. Kg6 Rg1+ 61. Kf5 {0.00/36}) 44. Bf5 {6.17/20 1} b6 {-7.60/10 11} 45. Be6+ { 10.94/21 1} Rxe6 {-9.65/11 3} 46. dxe6 {11.12/22 1} h3 {-9.82/10 2} 47. Rxe3 { 16.78/24 2} fxe3 {adjudication -12.87/10 12 Arena Adjudication} 1-0 [/pgn]

[yanquis1972]

I recall seeing some KS attacks with test30 some time ago; haven’t been able to observe many games so I don’t know if it’s still doing it or playing in a more traditional manner. I think it will always try to shape the game to favored patterns so eg a net that likes or is trained on the KID (something I’d still like to see, that being opening-specific.NNs) may end up utilizing its themes in surprising places.

[carldaman]

I am doing a test with Lc0 32472, AF125 and AF228 against Stockfish 10, and some nice games are showing up.

Hi, where were you able to get AF228 and AF125?
I first tried doing a Google search for Antifish, but websearch engines are completely useless nowadays.

Thanks,
CL

[Javier Ros]

I am doing a test with Lc0 32472, AF125 and AF228 against Stockfish 10, and some nice games are showing up.

Hi, where were you able to get AF228 and AF125?
I first tried doing a Google search for Antifish, but websearch engines are completely useless nowadays.

Thanks,
CL

Here

https://drive.google.com/file/d/1v5XvTN ... drive_open


https://groups.google.com/forum/m/#!top ... hiNzETf5jc

[carldaman]

Thanks for the links - interesting project!