Cheese - Lc0 1 - 0 TCEC

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

Jouni
Posts: 3757
Joined: Wed Mar 08, 2006 8:15 pm
Full name: Jouni Uski

Cheese - Lc0 1 - 0 TCEC

Post by Jouni »

https://tcec-chess.com/#div=sw4&game=36&season=24

What happened??
Jouni
Top
User avatar
Eelco de Groot
Posts: 4692
Joined: Sun Mar 12, 2006 2:40 am
Full name:   Eelco de Groot

Re: Cheese - Lc0 1 - 0 TCEC

Post by Eelco de Groot »

Is this Blitz?

[pgn][Event "TCEC Season 24 - Swiss 4"]
[Site "https://tcec-chess.com"]
[Date "2023.01.26"]
[Round "1.36"]
[White "Cheese 3.1.1"]
[Black "LCZero 0.30-dag-6b5f9451-BT2-3250000"]
[Result "1-0"]
[Annotator "archive"]

1. d4 f5 2. g3 Nf6 3. Nf3 d6 4. Bg2 c6 5. O-O Qc7 6. Nbd2 Rg8 7. Re1 f4 8. gxf4 g5 9. fxg5 Nbd7 10. gxf6 Nxf6 11. Nf1 Bg4 12. Ng3 O-O-O 13. e4 Kb8 14. h3 Bxf3 15. Qxf3 Nd7 16. Qb3 Rg6 17. h4 Bg7 18. h5 Rxg3 19. Qxg3 Bxd4 20. c3 Bb6 21. e5 Nxe5 22. Bf4 Ka8 23. Bxe5 dxe5 24. Qxe5 Qc8 25. Qe6 Qb8 26. Qxc6 bxc6 27. Bxc6+ Qb7 28. Bxb7+ Kxb7 29. Rxe7+ Kc6 30. Rxh7 Bxf2+ 31. Kxf2 Rf8+ 32. Ke3 Rg8 33. Rf1 Kd5 34. Rh6 Ke5 35. Rg6 Rh8 36. h6 Re8 37. Rgf6 Kd5+ 38. Kd3 Re6 39. h7 a5 40. c4+ Kd6 41. Rxe6+ Kxe6 42. h8=Q Kd6 43. Rf6+ Kd7 44. Qg7+ Kc8 45. Rf8#
*[/pgn]

6...Rg8 {?!?} Rebel 16.1 would now play 7.c4


[fen]rnb1kbr1/ppq1p1pp/2pp1n2/5p2/3P4/5NP1/PPPNPPBP/R1BQ1RK1 w q -[/fen]

Engine: Rebel-16.1 (64 MB)
gemaakt door Ed Schroder and Chris Whittington (C) 2023

21/33 0:01 +3.17++ 7.c4 Th8 8.e4 (1.823.356) 1035

21/33 0:02 +3.03 7.c4 d5 8.Pb3 e6 9.Lf4 Ld6 10.e3 Th8
11.Dc2 Pe4 12.Pc1 a5 13.Pd3 De7
14.c5 Lc7 15.Tfe1 g5 16.Lxc7 Dxc7
17.Pfe5 (2.297.410) 1048

22/34 0:03 +3.04 7.c4 d5 8.Pb3 e6 9.Lf4 Ld6 10.e3 Th8
11.Dc2 Pe4 12.Pc1 a5 13.Pd3 Pd7
14.Tfe1 Dd8 15.Tad1 Le7 16.h4 a4
17.cxd5 exd5 (3.946.781) 1079

23/36 0:07 +3.04++ 7.c4 g6 8.Dc2 (8.662.011) 1104

23/40 0:11 +2.81 7.c4 d5 8.cxd5 cxd5 9.Pe5 e6 10.Pdf3 Db6
11.Lf4 Pc6 12.b4 Pxe5 13.dxe5 Pe4
14.a3 Ld7 15.Le3 Pc3 16.Dd3 (12.657.866) 1113

24/37 0:13 +2.97++ 7.c4 Th8 (14.537.373) 1111

24/38 0:13 +3.04 7.c4 d5 8.cxd5 cxd5 9.Pe5 e6 10.Pdf3 Db6
11.Lf4 Pc6 12.Tc1 Pxe5 13.Pxe5 Ld6
14.Da4+ Ld7 15.Pxd7 Pxd7 16.e3 Lxf4
17.exf4 Ke7 18.Tfe1 (15.063.456) 1111

25/40 0:24 +2.88++ 7.c4 g5 8.Dc2 (27.228.525) 1115

25/38 0:25 +3.09 7.c4 d5 8.cxd5 cxd5 9.Pe5 e6 10.Pdf3 Db6
11.Lf4 Pc6 12.Tc1 Ld7 13.Pxd7 Kxd7
14.Da4 Ke8 15.e3 Le7 16.Pe5 Tc8
17.Pxc6 bxc6 (28.432.477) 1113

26/41 0:43 +2.93++ 7.c4 Th8 8.Te1 (48.109.534) 1116

26/43 0:50 +2.96 7.c4 d5 8.Te1 e6 9.Dc2 Th8 10.Pe5 a5
11.Pdf3 Pe4 12.Lf4 Dd8 13.h4 Pd7
14.e3 Le7 15.Pd3 h6 16.Le5 Pxe5
17.Pfxe5 Ld6 18.Tac1 g5 (56.308.929) 1122

27/41 1:04 +2.87 7.c4 d5 8.Dc2 e6 9.Pb3 Ld6 10.Pc5 De7
11.Pd3 a5 12.b3 Pbd7 13.a4 Th8
14.Pfe5 Pb8 15.Lf4 Pa6 16.cxd5 cxd5
17.Tfc1 Ld7 18.e3 Pb4 19.Pxb4 Lxb4 (72.023.708) 1124

28/40 1:25 +3.03++ 7.c4 Th8 8.Te1 (95.472.571) 1123

28/46 2:36 +3.10 7.c4 d5 8.b4 e6 9.Tb1 b5 10.cxb5 cxb5
11.Pb3 Ld6 12.Lg5 Pc6 13.Tc1 Ld7
14.a4 Db6 15.Pc5 Tc8 16.a5 Db8
17.Lxf6 gxf6 18.Dd2 Th8 19.Tfd1 Lxc5
20.Txc5 (175.174.428) 1118

29/43 3:03 +2.94++ 7.c4 e6 8.c5 (203.403.619) 1111

29/43 3:07 +2.81 7.c4 d5 8.b4 e6 9.Tb1 b5 10.cxb5 cxb5
11.a4 a6 12.Pb3 Ld6 13.Ld2 Pc6
14.Tc1 Db7 15.Pc5 Lxc5 16.bxc5 Ld7
17.Lf4 Kf7 18.Tb1 Tgc8 19.Pe5+ Pxe5
20.Lxe5 (208.187.037) 1111

30/44 3:15 +2.97++ 7.c4 g5 8.b4 (217.025.667) 1111

30/44 3:28 +3.13++ 7.c4 (230.206.779) 1102

30/45 3:41 +3.14 7.c4 d5 8.b4 e6 9.Tb1 b5 10.cxb5 cxb5
11.a4 a6 12.Pb3 Ld6 13.Lg5 Pc6
14.Tc1 h6 15.Pe5 hxg5 16.Txc6 Db8
17.Da1 Ld7 18.Pxd7 Kxd7 19.Tfc1 Db7
20.Da2 (243.500.135) 1101

Charisma MCTS from Eduard with the new Stockfish net would as well but not so high in eval:

Code: Select all

rnb1kbr1/ppq1p1pp/2pp1n2/5p2/3P4/5NP1/PPPNPPBP/R1BQ1RK1 w q -

Engine: Charisma MCTS 20230123 (512 MB)
gemaakt door E. Nemeth, K. Kiniama, A. Manzo and Stockfish developers

18/24  0:01   +1.60    7.c4 d5 8.b4 e6 9.Tb1 b5 10.cxb5 cxb5 
                       11.a4 Ld6 12.axb5 Ld7 13.Pb3 Pe4 
                       14.Ld2 Lxb5 15.Pg5 Pxg5 16.Lxg5 (1.342.414) 769 

19/28  0:02   +1.56    7.c4 d5 8.b4 e6 9.c5 a5 10.bxa5 g5 
                       11.Pb3 h6 12.Pe5 Pbd7 13.Pd3 Kf7 
                       14.f3 Kg7 (2.172.473) 773 

20/27  0:03   +1.63++  7.c4 (2.348.437) 771 

20/32  0:03   +1.50--  7.c4 d5 (2.664.316) 768 

20/32  0:03   +1.55    7.c4 d5 8.b4 e6 9.c5 a5 10.bxa5 Dxa5 
                       11.a4 Dd8 12.Pb3 Pbd7 13.Pg5 Pb8 
                       14.a5 h6 15.Pf3 g5 16.Dc2 Pbd7 
                       17.Pe1 h5 18.f3 (2.802.155) 765 

21/29  0:03   +1.52--  7.c4 d5 (2.964.187) 766 

21/29  0:03   +1.53    7.c4 d5 8.b4 e6 9.c5 a5 10.bxa5 Dxa5 
                       11.a4 Dd8 12.Pe1 g5 13.Pb3 h6 14.Pd3 Lg7 
                       15.f3 Th8 16.a5 Pa6 17.e4 (2.990.585) 766 

22/28  0:04   +1.50--  7.c4 d5 (3.340.805) 765 

22/28  0:04   +1.47--  7.c4 d5 (3.611.245) 767 

22/29  0:05   +1.43--  7.c4 d5 (4.044.151) 768 

22/30  0:05   +1.43    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 Ld7 
                       14.Tb3 g5 15.Pe5 Pc6 16.Pxc6 Lxc6 
                       17.Pf3 h6 18.Pe5 Ta4 19.Pxc6 (4.253.185) 768 

23/28  0:06   +1.54++  7.c4 (4.677.798) 760 

23/28  0:06   +1.64++  7.c4 (4.842.191) 761 

23/33  0:06   +1.57    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 La6 
                       14.Tb2 Pc6 15.Pb3 Pe4 16.Lf4 Db7 
                       17.Pe5 Pc3 18.Dc2 b4 19.Pxc6 Pxe2+ 
                       20.Kh1 (5.078.934) 759 

24/34  0:07   +1.55--  7.c4 d5 (5.620.875) 759 

24/34  0:07   +1.52--  7.c4 d5 (5.855.390) 760 

24/34  0:08   +1.48--  7.c4 d5 (6.102.053) 762 

24/34  0:08   +1.43--  7.c4 d5 (6.376.215) 763 

24/34  0:09   +1.48++  7.c4 (7.082.279) 759 

24/34  0:09   +1.49    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 La6 
                       14.Tb2 Pc6 15.Pb3 b4 16.Lf4 Db7 
                       17.Pe5 g5 18.Pxc6 Dxc6 19.Le5 Pg4 
                       20.Dd2 (7.312.026) 760 

25/34  0:11   +1.58++  7.c4 (8.484.787) 753 

25/36  0:12   +1.46    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 La6 
                       14.Tb2 Pc6 15.Pb3 b4 16.Lf4 Db7 
                       17.Pe5 g5 18.Ld2 Le7 19.Pxc6 Dxc6 
                       20.Pa5 (9.146.244) 752 

26/28  0:12   +1.46--  7.c4 d5 (9.368.902) 753 

26/35  0:12   +1.52++  7.c4 (9.648.058) 752 

26/35  0:13   +1.49    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 La6 
                       14.Tb2 Pc6 15.Pb3 b4 16.Lf4 Db7 
                       17.Pe5 g5 18.Ld2 Le7 19.Pxc6 Dxc6 
                       20.Pc1 (9.872.113) 753 

27/37  0:16   +1.60++  7.c4 (12.307.418) 745 

27/38  0:17   +1.49    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 La6 
                       14.Tb2 Pc6 15.Pb3 b4 16.Lf4 Db7 
                       17.Pe5 Pxe5 18.Lxe5 Kf7 19.Dd2 Lc4 
                       20.Tfb1 (12.880.682) 746 

28/40  0:21   +1.49--  7.c4 d5 (15.820.396) 738 

28/40  0:22   +1.55++  7.c4 (16.252.020) 738 

28/40  0:25   +1.48    7.c4 d5 8.b4 e6 9.c5 g5 10.Pb3 g4 
                       11.Lf4 Dg7 12.Pe5 h5 13.b5 h4 14.Pa5 Ph5 
                       15.Da4 Pxf4 16.gxf4 Dc7 17.Tfb1 h3 
                       18.Lf1 Lg7 (18.861.984) 734 

29/35  0:30   +1.47--  7.c4 d5 (22.082.518) 728 

29/38  0:34   +1.44--  7.c4 d5 (25.145.671) 727 

29/38  0:36   +1.48++  7.c4 (26.860.855) 726 

29/38  0:40   +1.40--  7.c4 d5 (29.611.706) 727 

29/44  0:47   +1.53++  7.c4 (34.047.551) 722 

29/44  0:49   +1.42    7.c4 d5 8.b4 e6 9.a3 g5 10.Pb3 Pe4 
                       11.Pc5 h6 12.Pxe4 fxe4 13.Pe1 Pd7 
                       14.f3 dxc4 15.Dc2 Pb6 16.fxe4 e5 
                       17.Pf3 Le6 18.Pxe5 Lg7 (35.464.799) 722 

30/33  0:52   +1.41--  7.c4 d5 (38.105.653) 722 

30/36  0:58   +1.48++  7.c4 (42.061.773) 720 

30/37  0:59   +1.49    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 Ld7 
                       14.Dc2 Pc6 15.Txb5 g5 16.Pb3 Pe4 
                       17.Tb6 Le7 18.Le3 Kf7 19.Pe5+ Pxe5 
                       20.dxe5 (42.898.254) 720 

31/40  1:09   +1.58++  7.c4 (49.937.438) 715 

31/43  1:12   +1.45--  7.c4 d5 (51.791.099) 716 

31/43  1:20   +1.41--  7.c4 d5 (57.241.565) 715 

31/43  1:22   +1.49++  7.c4 (58.970.320) 714 

31/43  1:24   +1.54    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 Pc6 
                       14.Txb5 g5 15.Pb3 Pe4 16.Dc2 Le7 
                       17.Le3 Ld7 18.Tb6 Lf6 19.Pe5 Lxe5 
                       20.Lxe4 (60.120.994) 714 

32/44  1:34   +1.59    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 Pc6 
                       14.Txb5 g5 15.Pb3 Pe4 16.Dc2 h5 
                       17.Le3 f4 18.Ld2 h4 19.g4 h3 20.Lxh3 (67.351.998) 710 

33/41  1:42   +1.54--  7.c4 d5 (72.541.597) 708 

33/41  1:44   +1.51--  7.c4 d5 (74.289.212) 709 

33/43  1:47   +1.47--  7.c4 d5 (76.408.420) 710 

33/44  1:51   +1.51++  7.c4 (79.336.355) 709 

33/44  1:57   +1.45    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 Pc6 
                       14.Txb5 g5 15.Pb3 Pe4 16.Dc2 h5 
                       17.Le3 h4 18.Tb6 f4 19.Pxg5 Pxg5 
                       20.Lxf4 (83.210.537) 708 

34/40  2:32   +1.44--  7.c4 d5 (107.248.453) 703 

34/42  2:37   +1.50++  7.c4 (110.003.264) 699 

34/43  3:16   +1.54    7.c4 d5 8.Pb3 e6 9.Pc5 Ld6 10.Pd3 De7 
                       11.c5 Lc7 12.b4 a5 13.Da4 Pe4 14.b5 cxb5 
                       15.Dxb5+ Pc6 16.Lf4 Lxf4 17.Pxf4 g5 
                       18.Pd3 Kf8 19.Pfe5 Pxd4 20.Db2 (135.854.628) 692 

35/43  3:30   +1.49--  7.c4 d5 (145.277.391) 691 

35/47  3:33   +1.47--  7.c4 d5 (147.800.917) 691 

35/47  3:47   +1.43--  7.c4 d5 (156.792.873) 688 

35/47  4:02   +1.47++  7.c4 (166.940.922) 687 

35/47  4:10   +1.48    7.c4 d5 8.Pb3 e6 9.Pc5 Ld6 10.Pd3 De7 
                       11.c5 Lc7 12.b4 a5 13.Da4 h6 14.Lf4 Lxf4 
                       15.Pxf4 g5 16.Pd3 Kf8 17.Pfe5 Tg7 
                       18.Tab1 Pfd7 19.Pxd7+ Dxd7 20.b5 (172.782.415) 688 

36/47  4:34   +1.44--  7.c4 d5 (189.103.528) 688 

36/47  4:56   +1.42--  7.c4 d5 (204.139.227) 688 

36/47  5:30   +1.46++  7.c4 (227.144.175) 686 

36/47  5:53   +1.63++  7.c4 (242.588.601) 686 

36/48  6:10   +1.45    7.c4 d5 8.b4 e6 9.a3 g5 10.Dc2 Ld6 
                       11.Pb3 h6 12.Pc5 a5 13.Ld2 axb4 
                       14.axb4 Txa1 15.Txa1 Pe4 16.Lc1 b5 
                       17.Pxe4 fxe4 18.c5 Le7 19.Pe5 Pd7 
                       20.Lb2 (254.414.937) 687 

37/31  6:18   +1.46--  7.c4 d5 (260.612.748) 687 

37/50  6:56   +1.52++  7.c4 (285.475.151) 685 

37/50  7:12   +1.59    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 Pc6 
                       14.Txb5 g5 15.Pb3 Pe4 16.Dc2 h5 
                       17.Tb6 h4 18.Le3 f4 19.Pxg5 Pxg5 
                       20.Lxf4 (296.289.735) 684 

38/44  7:39   +1.53--  7.c4 d5 (314.846.282) 684 

38/44  7:54   +1.50--  7.c4 d5 (324.966.398) 685 

38/48  8:27   +1.55++  7.c4 (347.589.516) 685 

38/49  8:50   +1.47--  7.c4 d5 (364.251.721) 686 

38/53  9:06   +1.48    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 Pc6 
                       14.Txb5 g5 15.Pb3 Pe4 16.Dc2 h5 
                       17.Tb6 h4 18.Le3 f4 19.Pxg5 Pxg5 
                       20.Lxf4 (375.076.291) 686 

39/56  10:00  +1.45--  7.c4 d5 (412.752.050) 687 

39/56  10:21  +1.52++  7.c4 (427.232.717) 687 

39/56  10:22  +1.52    7.c4 d5 8.b4 e6 9.Tb1 b5 10.c5 a5 
                       11.a4 axb4 12.axb5 cxb5 13.Txb4 Pc6 
                       14.Txb5 g5 15.Pb3 Pe4 16.Dc2 h5 
                       17.Tb6 h4 18.Le3 f4 19.Pxg5 Pxg5 
                       20.Lxf4 (427.965.893) 687
After 7.c4 d6 8. b4 e6

[fen]rnb1kbr1/ppq3pp/2p1pn2/3p1p2/1PPP4/5NP1/P2NPPBP/R1BQ1RK1 w q -[/fen]
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
Top
Patrice Duhamel
Posts: 203
Joined: Sat May 25, 2013 11:17 am
Location: France
Full name: Patrice Duhamel

Re: Cheese - Lc0 1 - 0 TCEC

Post by Patrice Duhamel »

Happy to see Cheese can win sometimes :)
Anything that can go wrong will go wrong.
Top
Guenther
Posts: 4718
Joined: Wed Oct 01, 2008 6:33 am
Location: Regensburg, Germany
Full name: Guenther Simon

Re: Cheese - Lc0 1 - 0 TCEC

Post by Guenther »

Eelco de Groot wrote: ↑Thu Jan 26, 2023 9:52 am Is this Blitz?

[pgn][Event "TCEC Season 24 - Swiss 4"]
[Site "https://tcec-chess.com"]
[Date "2023.01.26"]
[Round "1.36"]
[White "Cheese 3.1.1"]
[Black "LCZero 0.30-dag-6b5f9451-BT2-3250000"]
[Result "1-0"]
[Annotator "archive"]

1. d4 f5 2. g3 Nf6 3. Nf3 d6 4. Bg2 c6 5. O-O Qc7 6. Nbd2 Rg8 7. Re1 f4 8. gxf4 g5 9. fxg5 Nbd7 10. gxf6 Nxf6 11. Nf1 Bg4 12. Ng3 O-O-O 13. e4 Kb8 14. h3 Bxf3 15. Qxf3 Nd7 16. Qb3 Rg6 17. h4 Bg7 18. h5 Rxg3 19. Qxg3 Bxd4 20. c3 Bb6 21. e5 Nxe5 22. Bf4 Ka8 23. Bxe5 dxe5 24. Qxe5 Qc8 25. Qe6 Qb8 26. Qxc6 bxc6 27. Bxc6+ Qb7 28. Bxb7+ Kxb7 29. Rxe7+ Kc6 30. Rxh7 Bxf2+ 31. Kxf2 Rf8+ 32. Ke3 Rg8 33. Rf1 Kd5 34. Rh6 Ke5 35. Rg6 Rh8 36. h6 Re8 37. Rgf6 Kd5+ 38. Kd3 Re6 39. h7 a5 40. c4+ Kd6 41. Rxe6+ Kxe6 42. h8=Q Kd6 43. Rf6+ Kd7 44. Qg7+ Kc8 45. Rf8#
*[/pgn]

6...Rg8 {?!?} Rebel 16.1 would now play 7.c4

Just looks like a complete setup error for LC0... or they mixed it up with a giveaway chess version of LC0 ;)
(no need to investigate this game more than a few seconds)
https://rwbc-chess.de

[Trolls n'existent pas...]
Top
Damir
Posts: 2900
Joined: Mon Feb 11, 2008 3:53 pm
Location: Denmark
Full name: Damir Desevac

Re: Cheese - Lc0 1 - 0 TCEC

Post by Damir »

Jouni wrote: ↑Thu Jan 26, 2023 9:17 am https://tcec-chess.com/#div=sw4&game=36&season=24

What happened??
Garbage opening moves, sacrificing pawns in the opening for nothing... :) :)
Top
Jouni
Posts: 3757
Joined: Wed Mar 08, 2006 8:15 pm
Full name: Jouni Uski

Re: Cheese - Lc0 1 - 0 TCEC

Post by Jouni »

I don't have BT2 net. But my Lc0 shows clear white win already at move 10:

Analysis by Lc0 v0.30.0-dag+git.540ddc2:

11.Nf1 Bh3 12.Ng3 Qd7 13.e4 0-0-0 14.e5 Bxg2 15.Kxg2 dxe5 16.Nxe5 Qxd4 17.Qxd4
+- (13.69) Depth: 7/14 00:00:03 4kN
11.Nf1 Bh3 12.Ng3 Qd7 13.e4 0-0-0 14.e5 Bxg2 15.Kxg2 dxe5 16.Nxe5 Qxd4 17.Qxd4
+- (14.58) Depth: 7/15 00:00:05 7kN
11.Nf1 Bh3 12.Ng3 Qd7 13.e4 0-0-0 14.e5 Bxg2 15.Kxg2 dxe5 16.Nxe5 Qxd4 17.Qxd4
+- (14.58) Depth: 7/16 00:00:07 8kN
11.Nf1 e5 12.e4 Bg4 13.Ng3 0-0-0 14.h3 Be6 15.a4 Qg7 16.a5
+- (14.59) Depth: 7/17 00:00:07 9kN
11.Nf1 Bh3 12.Ng3 Bxg2 13.Kxg2 0-0-0 14.e4 e5 15.Bg5 Be7 16.Bxf6 Bxf6 17.dxe5
+- (14.58) Depth: 7/17 00:00:12 15kN
11.Nf1 Bg4 12.Ng3 0-0-0 13.c4 e5 14.d5 c5 15.b4 e4 16.Ng5 Qg7 17.Rb1
+- (15.03) Depth: 7/17 00:00:17 21kN
11.Nf1 Bg4 12.Ng3 0-0-0 13.c4 e5 14.d5 c5 15.b4 e4 16.Ng5 Qg7 17.Rb1
+- (14.78) Depth: 7/17 00:00:22 27kN
11.Nf1 Bg4 12.Ng3 0-0-0 13.c4 e5 14.d5 c5 15.b4 e4 16.Ng5 Qg7 17.Rb1
+- (14.79) Depth: 7/17 00:00:27 33kN
11.Nf1 Bg4 12.Ng3 0-0-0 13.c4 e5 14.d5 c5 15.b4 e4 16.Ng5 Qg7 17.Rb1
+- (14.84) Depth: 8/17 00:00:28 34kN
11.Nf1 Bg4 12.Ng3 0-0-0 13.c4 e5 14.d5 c5 15.b4 e4 16.Ng5 Qg7 17.Rb1
+- (14.85) Depth: 8/18 00:00:30 36kN

Ok not normal game. There was already crash with Lc0 in one game.
Jouni
Top
Plutie
Posts: 20
Joined: Sun Jan 30, 2022 6:14 am
Full name: Evan Engler

Re: Cheese - Lc0 1 - 0 TCEC

Post by Plutie »

to explain what exactly happened: this version of leela is using a newer form of contempt based on the elo difference between players. the general idea behind it is to transform the network's WDL output to more accurately represent the chances of winning based on the opponent's skill. unfortunately, at such a high elo difference, leela expects that almost anything can be played and still finish in a win. in this case, with a 660 elo difference (internally treated as 60% of that value,) the contempt, with how high it was, ended up giving not only the book exit, but also the position after multiple unreasonable piece sacs, an 80% win chance for black. bottom line is: it couldn't exactly handle such a high elo difference, so it played really poorly as black. there's now a cap to how high the elo difference used by leela can be, at 420 elo. (note: currently not in effect at TCEC, since there probably won't be an update allowed.)

a more in-depth explanation can be found here, I'm just going off of my surface level knowledge -
Top
Krzysztof Grzelak
Posts: 1586
Joined: Tue Jul 15, 2014 12:47 pm

Re: Cheese - Lc0 1 - 0 TCEC

Post by Krzysztof Grzelak »

Plutie wrote: ↑Thu Jan 26, 2023 4:31 pm to explain what exactly happened: this version of leela is using a newer form of contempt based on the elo difference between players. the general idea behind it is to transform the network's WDL output to more accurately represent the chances of winning based on the opponent's skill. unfortunately, at such a high elo difference, leela expects that almost anything can be played and still finish in a win. in this case, with a 660 elo difference (internally treated as 60% of that value,) the contempt, with how high it was, ended up giving not only the book exit, but also the position after multiple unreasonable piece sacs, an 80% win chance for black. bottom line is: it couldn't exactly handle such a high elo difference, so it played really poorly as black. there's now a cap to how high the elo difference used by leela can be, at 420 elo. (note: currently not in effect at TCEC, since there probably won't be an update allowed.)

a more in-depth explanation can be found here, I'm just going off of my surface level knowledge -
Plutie what version should play on TCEC to fix this problem.
Top
Plutie
Posts: 20
Joined: Sun Jan 30, 2022 6:14 am
Full name: Evan Engler

Re: Cheese - Lc0 1 - 0 TCEC

Post by Plutie »

Krzysztof Grzelak wrote: ↑Thu Jan 26, 2023 4:47 pm
Plutie wrote: ↑Thu Jan 26, 2023 4:31 pm to explain what exactly happened: this version of leela is using a newer form of contempt based on the elo difference between players. the general idea behind it is to transform the network's WDL output to more accurately represent the chances of winning based on the opponent's skill. unfortunately, at such a high elo difference, leela expects that almost anything can be played and still finish in a win. in this case, with a 660 elo difference (internally treated as 60% of that value,) the contempt, with how high it was, ended up giving not only the book exit, but also the position after multiple unreasonable piece sacs, an 80% win chance for black. bottom line is: it couldn't exactly handle such a high elo difference, so it played really poorly as black. there's now a cap to how high the elo difference used by leela can be, at 420 elo. (note: currently not in effect at TCEC, since there probably won't be an update allowed.)

a more in-depth explanation can be found here, I'm just going off of my surface level knowledge -
Plutie what version should play on TCEC to fix this problem.
well, it's been fixed in this commit, but since the event already started, I don't think there will be a chance to update it.
Top
User avatar
hgm
Posts: 28426
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: Cheese - Lc0 1 - 0 TCEC

Post by hgm »

Plutie wrote: ↑Thu Jan 26, 2023 4:31 pm to explain what exactly happened: this version of leela is using a newer form of contempt based on the elo difference between players. the general idea behind it is to transform the network's WDL output to more accurately represent the chances of winning based on the opponent's skill. unfortunately, at such a high elo difference, leela expects that almost anything can be played and still finish in a win. in this case, with a 660 elo difference (internally treated as 60% of that value,) the contempt, with how high it was, ended up giving not only the book exit, but also the position after multiple unreasonable piece sacs, an 80% win chance for black. bottom line is: it couldn't exactly handle such a high elo difference, so it played really poorly as black. there's now a cap to how high the elo difference used by leela can be, at 420 elo. (note: currently not in effect at TCEC, since there probably won't be an update allowed.)

a more in-depth explanation can be found here, I'm just going off of my surface level knowledge -
It means there is something fundamentally wrong in the way you correct the winning probability for the Elo difference. For one, it is not so much the Elo difference that matters, but the Elo of the weak player. When you are in KQK it doesn't really matter whether the bare King is handled by a 1500-Elo player or a 3500-Elo one, and it doesn't matter whether the Queen is handled by a 1000-Elo player or a 2500-Elo one.

Of course it also points to a gigantic Leela flaw, that it thinks it can be in any way best to sacrifice material because it thinks 'it will win anyway'. This is exactly the same flaw as that makes it play embarrasingly indecisive in won positions when playing without any contempt. It just doesn't know how to win won positions, and will often bungle them.
Top

Return to “Computer Chess Club: General Topics”