Seer 2.0.0

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

Moderators: hgm, Rebel, chrisw

connor_mcmonigle
Posts: 532
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Seer 2.0.0

Post by connor_mcmonigle »

I've released Seer 2.0.0 here: https://github.com/connormcmonigle/seer ... tag/v2.0.0.

With any luck, it should be more than 100 elo stronger than v1.2.1. With this version, the network weights are now embedded into the binaries. The embedded network is trained entirely on data generated by Seer's search starting with a randomly initialized network. The network is no longer directly or indirectly trained on Stockfish derived training data. The training technique employed is a variant of semisupervised learning and involves starting with a large number of <=6 man positions labeled with WDL values using Syzygy EGTBs. I then iteratively back up these initial WDL scores to 32-man chess positions using Seer's search to find continuations from N-man chess positions to <N-man chess positions using a network trained on all the (thus far) labelled <N man positions.

All positions (~700 million) I used were sampled randomly from human Lichess games which resulted in a diverse, though not completely random, sample of positions. I then subsequently binned the positions based on the number of remaining pieces and assigned positions with <= 6 men WDL values using EGTBs. Then, for each iteration N in (7, 32), I trained a network on all the <N-1 man labelled positions. I then used the resultant network to play out every quiet N man position to a quiet <N man position and used the static WDL prediction of the network on the resultant <N man position as a label which I assigned to the original N-man position. I then added the label-position pair to the training dataset for use in the next iteration (N+1). After labelling every sampled position using this process, I trained a number of networks on the resultant labelled data in SL fashion which made tweaking hyperparameters and tinkering with the layer dimensions feasible on my limited hardware (regenerating labels everytime I was interested in testing a small model tweak would have been impractical).

Qualitatively, relative to my older networks trained on Stockfish NNUE data, my semi-supervised training based networks are largely inferior positionally and I find their pawn play in particular to be lacking. However, they prove massively superior at developing long term attacks and also exhibit an advantage in the endgame owing to the EGTBs their understanding of chess was iteratively constructed from. The strongest semi-supervised training based network I produced proved about 15 elo weaker than the strongest network I trained on Stockfish NNUE data. Here's an illustrative (if somewhat meaningless in isolation) game played against Halogen 10 at 1min+1s from the start position:

[pgn]
[Date "2021.04.23"]
[Round "?"]
[White "Seer v2.0.0"]
[Black "Halogen 10"]
[Result "1-0"]
[WhiteElo "?"]
[BlackElo "?"]
[Variant "Standard"]
[TimeControl "60+1"]
[ECO "B50"]
[Opening "Sicilian Defense"]
[Termination "Normal"]
[Annotator "lichess.org"]

1. e4 c5 2. Nf3 d6 { B50 Sicilian Defense } 3. a4 Nf6 4. Bb5+ Bd7 5. d3 g6 6. Bxd7+ Qxd7 7. c3 Bg7 8. O-O Nc6 9. h3 O-O 10. Re1 e5 11. Nbd2 b6 12. Nc4 Rfe8 13. a5?! { (0.26 → -0.37) Inaccuracy. Bg5 was best. } (13. Bg5 d5) 13... b5 14. Ne3 a6 15. Nh2 Qd8?! { (-0.51 → 0.40) Inaccuracy. h5 was best. } (15... h5) 16. Nhg4 Nxg4 17. hxg4 Nxa5 18. Nd5 Nc6 19. Be3 a5 20. g5 Ne7 21. Nf6+ Bxf6 22. gxf6 Nc6 23. Bg5 Qc8 24. Re3 a4 25. Rh3 Na5?? { (-0.37 → 1.84) Blunder. Nd8 was best. } (25... Nd8 26. Rh6) 26. g4 Nb3 27. f3 c4?? { (2.54 → 5.87) Blunder. Qe6 was best. } (27... Qe6 28. Kf2) 28. Kg2 h5 29. gxh5 Nxa1 30. dxc4 Qxc4?? { (6.02 → Mate in 11) Checkmate is now unavoidable. a3 was best. } (30... a3 31. bxa3) 31. hxg6 fxg6 32. Qh1 Qe2+ 33. Kg3 Qe1+ 34. Qxe1 d5 35. exd5 Ra7 36. Qh1 Rg7 37. Rh8+ Kf7 38. fxg7 Rg8 39. Rxg8 Kxg8 40. Qh8+ Kf7 41. g8=Q# { White wins by checkmate. } 1-0
Computer analysis[/pgn]
Top
User avatar
Guenther
Posts: 4606
Joined: Wed Oct 01, 2008 6:33 am
Location: Regensburg, Germany
Full name: Guenther Simon

Re: Seer 2.0.0

Post by Guenther »

connor_mcmonigle wrote: ↑Tue Apr 27, 2021 12:20 pm I've released Seer 2.0.0 here: https://github.com/connormcmonigle/seer ... tag/v2.0.0.
...

Qualitatively, relative to my older networks trained on Stockfish NNUE data, my semi-supervised training based networks are largely inferior positionally and I find their pawn play in particular to be lacking. However, they prove massively superior at developing long term attacks and also exhibit an advantage in the endgame owing to the EGTBs their understanding of chess was iteratively constructed from.
...
Hi Connor,

Thanks for the new release.

I just let it play a few games vs. Komodo3 which is already around 3100 at CCRLBlitz and already in the first four games I could see
some fine endgame improvement and also some long term attacking style.
For the endgame just look how it saved game 1 and especially game 3, where it very soon obviously knew it will end up with a superior
Knight vs. Bishop endgame! (it should be noted though that Komodo3 was not yet able to use tablebases, this came in later versions)

OTH I noticed that the time management for mps tc is not safe and it lost two of four (totally won) games on time before the second session ended.

I have added the pgn in column style for those lost on time games at the end again, as here it can be easily seen how fast it was forced to play
from around move 72-75 and didn't succeed. It also plays already move 39-40 in extreme time trouble, but here it succeeded always.
This was in CuteChessDev at 40/80s (with no extra time margin defined as usual) with 64MB hash for both and with adjudication nearly off,
for finding more bugs ;-)

[pgn][Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "1"]
[White "Komodo_3-64"]
[Black "Seer_200-64"]
[Result "1/2-1/2"]
[ECO "A06"]
[GameDuration "00:05:20"]
[GameEndTime "2021-04-27T13:59:30.574 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T13:54:10.334 Mitteleuropõische Sommerzeit"]
[Opening "Reti"]
[PlyCount "162"]
[Termination "adjudication"]
[TimeControl "40/80"]
[Variation "Nimzovich-Larsen attack"]

1. Nf3 {book} d5 {book}
2. b3 {book} Bg4 {book}
3. e3 {book} e5 {book}
4. h3 {+0.06/15 3.84} Bxf3 {-0.32/12 1.27}
5. Qxf3 {+0.03/15 0.75} e4 {-0.19/13 1.88}
6. Qg3 {+0.16/15 2.30} Nc6 {-0.32/12 1.29}
7. Bb5 {+0.31/14 1.18} Qd7 {-0.07/11 1.31}
8. Ba3 {+0.32/14 2.77} h5 {-0.14/13 1.31}
9. c4 {+0.39/13 2.29} Nf6 {-0.33/15 4.60}
10. Bxc6 {+0.45/14 2.80} bxc6 {-0.21/13 1.28}
11. Qe5+ {+0.41/16 3.09} Be7 {-0.50/17 5.41}
12. O-O {+0.41/16 2.72} Rc8 {-0.85/17 2.56}
13. Nc3 {+0.41/15 2.03} Qe6 {-0.17/13 1.20}
14. Qxe6 {+0.41/14 1.15} fxe6 {-0.56/15 8.33}
15. Bb2 {+0.38/15 4.54} Kf7 {-0.37/13 1.08}
16. Ne2 {+0.60/13 1.13} c5 {-0.76/14 3.17}
17. cxd5 {+0.97/15 4.04} exd5 {-0.72/14 2.54}
18. f3 {+0.92/15 3.38} Rhd8 {-1.24/12 1.03}
19. fxe4 {+1.02/14 1.85} dxe4 {-1.47/14 2.11}
20. Rad1 {+1.03/16 1.71} Bd6 {-1.06/14 1.22}
21. Bxf6 {+1.00/14 1.75} gxf6 {-2.36/18 5.62}
22. Rf5 {+0.86/15 2.45} Be5 {-1.26/15 1.70}
23. Rxh5 {+0.81/16 1.58} Rd3 {-1.44/14 1.60}
24. Kf2 {+0.87/16 2.91} Rcd8 {-1.80/14 0.95}
25. Ke1 {+0.71/16 1.93} Rg8 {-1.46/12 0.92}
26. Rh7+ {+0.71/15 1.16} Ke6 {-1.15/14 1.03}
27. g4 {+0.86/16 1.71} Rgd8 {-1.56/14 1.00}
28. Rh5 {+0.84/16 1.83} Kf7 {-1.61/13 0.95}
29. Nc1 {+0.76/15 2.25} R3d5 {-2.15/17 11.20}
30. Ke2 {+0.70/13 1.00} a6 {-1.78/16 2.59}
31. d3 {+0.82/14 1.27} Ke6 {-2.12/13 0.50}
32. Rh7 {+0.85/15 2.56} a5 {-2.67/13 0.64}
33. a4 {+0.90/14 1.80} Bg3 {-1.56/17 5.39}
34. h4 {+0.86/14 1.58} c6 {-2.71/14 2.54}
35. h5 {+1.27/14 1.04} f5 {-2.31/12 0.21}
36. Rh6+ {+0.82/13 2.54} Ke7 {-1.95/11 0.11}
37. Rg6 {+1.03/14 3.50} f4 {-2.57/11 0.13}
38. exf4 {+0.87/14 1.61} Bxf4 {-1.39/14 1.02}
39. Rf1 {+0.87/14 1.00} Bxc1 {-1.68/11 0.062}
40. Rxc1 {+0.82/13 0.81} exd3+ {-1.50/8 0.033}
41. Kd2 {+0.98/14 0.87} Rf8 {-2.38/13 1.20}
42. Rxc6 {+1.03/15 2.17} Rf2+ {-3.23/15 1.21}
43. Kc3 {+1.03/16 2.14} Kd7 {-3.71/16 6.83}
44. Ra6 {+1.48/13 1.43} Rh2 {-3.21/16 4.54}
45. Rxa5 {+1.50/13 1.39} Kc7 {-3.61/16 14.45}
46. Ra7+ {+1.51/15 1.60} Kb6 {-4.69/14 0.87}
47. Rg7 {+1.55/15 1.57} Ka5 {-3.61/18 11.62}
48. Rg8 {+1.86/16 5.51} Ka6 {-4.32/13 0.70}
49. Rh8 {+1.86/14 2.82} Ka7 {-4.23/16 9.24}
50. Rh6 {+1.78/15 2.07} Rd4 {-3.78/14 0.56}
51. g5 {+1.84/14 1.26} d2 {-4.06/13 0.55}
52. Rd1 {+2.06/15 1.79} Rh3+ {-1.66/17 7.30}
53. Kb2 {+1.85/16 11.50} Rdd3 {-1.59/14 0.48}
54. Rh7+ {+0.92/12 1.66} Kb6 {-0.99/13 0.51}
55. a5+ {+0.79/13 1.71} Kxa5 {-0.46/15 1.26}
56. Rb7 {+0.55/14 2.23} Ka6 {-0.06/15 2.33}
57. Rb8 {+0.55/16 2.40} Ka7 {-0.06/12 0.39}
58. Rb5 {+0.14/16 6.41} Ka6 {-0.06/13 0.40}
59. Rxc5 {+0.08/15 0.79} Rxb3+ {-0.05/12 0.40}
60. Kc2 {+0.08/15 0.56} Ra3 {-0.03/13 0.40}
61. Rc8 {+0.08/13 1.72} Kb7 {-0.01/12 0.40}
62. Rc4 {+0.05/14 1.39} Ra2+ {0.00/11 0.42}
63. Kb1 {+0.44/11 0.28} Ra5 {-0.42/13 0.43}
64. g6 {+0.08/13 1.70} Rhxh5 {-0.41/11 0.42}
65. Rf4 {+0.05/14 1.39} Rhg5 {-0.10/12 0.76}
66. Rf7+ {+0.58/11 0.55} Kb6 {-0.28/12 0.42}
67. Rf6+ {+0.56/14 0.87} Kb5 {-0.08/13 0.44}
68. Rxd2 {+0.56/12 0.65} Rg1+ {-0.02/13 0.43}
69. Kb2 {+0.05/13 0.57} Ra6 {-0.02/12 0.44}
70. Rxa6 {+0.05/15 1.87} Kxa6 {-0.02/13 0.47}
71. Rd6+ {+0.04/18 2.27} Kb5 {-0.01/13 0.47}
72. Kc3 {+0.04/19 1.55} Kc5 {0.00/12 0.47}
73. Ra6 {+0.05/18 2.01} Kd5 {0.00/12 0.50}
74. Kd3 {+0.04/18 1.62} Ke5 {0.00/13 0.50}
75. Kc4 {+0.04/14 0.36} Kf5 {0.00/12 0.52}
76. Kd5 {+0.04/15 1.23} Rxg6 {0.00/12 0.54}
77. Ra8 {+0.04/13 0.61} Kf6 {0.00/12 0.55}
78. Rf8+ {+0.04/14 3.68} Kg7 {0.00/12 0.57}
79. Rf2 {+0.03/16 3.24} Rg5+ {0.00/12 0.61}
80. Ke6 {+0.04/17 0.56} Rg3 {0.00/13 0.63}
81. Ra2 {+0.04/16 2.62} Kg6 {0.00/14 1.28}
{Draw by adjudication: user decision}
1/2-1/2

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "1"]
[White "Seer_200-64"]
[Black "Komodo_3-64"]
[Result "1-0"]
[ECO "A06"]
[GameDuration "00:03:49"]
[GameEndTime "2021-04-27T14:03:20.754 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T13:59:31.634 Mitteleuropõische Sommerzeit"]
[Opening "Reti"]
[PlyCount "121"]
[TimeControl "40/80"]
[Variation "Nimzovich-Larsen attack"]

1. Nf3 {book} d5 {book}
2. b3 {book} Bg4 {book}
3. e3 {book} e5 {book}
4. h3 {+0.32/14 1.67} Bxf3 {-0.13/16 3.69}
5. Qxf3 {+0.31/12 1.27} Nc6 {-0.13/13 0.26}
6. Bb2 {+0.46/13 1.29} Nf6 {-0.12/15 3.40}
7. g4 {+0.64/13 1.31} e4 {-0.05/14 3.64}
8. Qg2 {+0.75/13 1.63} Nd7 {-0.13/15 4.60}
9. Nc3 {+1.69/15 4.01} Nce5 {-0.27/15 2.86}
10. O-O-O {+1.54/13 1.28} c6 {-0.35/15 1.74}
11. d3 {+1.25/16 4.08} exd3 {-0.32/12 1.05}
12. f4 {+1.71/15 1.81} dxc2 {-0.29/13 1.75}
13. Qxc2 {+1.53/16 5.86} Ng6 {-0.39/15 6.52}
14. Kb1 {+1.76/16 4.41} a5 {-0.55/15 10.44}
15. h4 {+2.26/14 1.49} h5 {-0.64/12 1.98}
16. g5 {+2.62/14 1.92} Qe7 {-0.76/13 1.57}
17. f5 {+5.28/18 14.12} Nge5 {-0.62/13 1.65}
18. Nxd5 {+5.17/14 0.81} cxd5 {-0.52/11 0.26}
19. Rxd5 {+5.45/13 0.83} Nc6 {-1.55/12 0.72}
20. Bb5 {+5.78/18 10.40} f6 {-2.44/13 2.57}
21. Rhd1 {+5.87/15 1.13} Rd8 {-3.59/15 7.96}
22. Bc3 {+6.96/15 0.58} Rc8 {-3.48/13 1.04}
23. Rxd7 {+8.06/15 1.45} Qxd7 {-3.56/13 0.54}
24. Rxd7 {+10.54/18 6.73} Kxd7 {-3.46/14 0.72}
25. Bxa5 {+11.72/13 0.43} Bd6 {-4.58/14 1.31}
26. Qd3 {+12.33/14 0.70} Rcd8 {-4.86/15 0.55}
27. g6 {+12.89/14 0.39} Rhe8 {-6.20/15 1.63}
28. a4 {+12.82/15 0.55} Kc8 {-5.37/18 1.24}
29. Bxd8 {+13.43/15 0.41} Rxd8 {-3.84/14 0.40}
30. Bxc6 {+14.13/14 0.41} bxc6 {-3.92/15 0.15}
31. Qa6+ {+14.47/14 0.41} Kd7 {-4.12/15 0.16}
32. a5 {+14.90/12 0.41} Rb8 {-4.74/17 1.27}
33. Qc4 {+16.25/18 4.58} Re8 {-5.85/18 3.12}
34. Kc2 {+16.99/15 2.15} Re5 {-4.82/16 1.35}
35. b4 {+16.24/11 0.13} Bc7 {-5.21/16 1.52}
36. Qg8 {+18.42/13 0.93} Re7 {-6.24/18 1.49}
37. a6 {+16.23/12 0.12} Bb6 {-6.95/18 2.36}
38. Qa8 {+17.88/8 0.029} Bxe3 {-6.82/17 1.38}
39. Kd3 {+16.75/11 0.082} Bg1 {-6.34/17 1.66}
40. a7 {+17.01/11 0.049} Bxa7 {-6.74/17 1.02}
41. Qxa7+ {+19.71/15 1.19} Kd6 {-6.82/18 0.50}
42. Qb8+ {+29.19/21 16.12} Kd7 {-7.38/19 1.78}
43. Kd4 {+31.51/20 13.11} Re1 {-9.78/17 2.68}
44. Kc5 {+28.38/16 0.80} Re5+ {-9.41/15 1.09}
45. Qxe5 {+36.83/19 10.65} fxe5 {-10.97/18 0.66}
46. f6 {+36.83/17 3.35} gxf6 {-10.97/18 0.62}
47. g7 {+36.83/14 1.19} Ke6 {-10.97/14 0.19}
48. g8=Q+ {+36.83/12 0.60} Kf5 {-10.97/16 0.37}
49. Kxc6 {+36.83/12 0.61} e4 {-M22/12 0.48}
50. Kd5 {+36.83/13 0.60} Kf4 {-M26/13 0.80}
51. Kd4 {+36.83/14 0.63} Kf3 {-M26/15 1.50}
52. Qb3+ {+36.83/14 0.63} Kg4 {-M22/14 0.65}
53. Qd1+ {+36.83/13 0.62} Kf5 {-M26/15 1.34}
54. Qxh5+ {+36.83/13 0.63} Kf4 {-M24/13 0.070}
55. Qh6+ {+36.83/15 0.64} Kf5 {-M22/12 0.061}
56. Qh7+ {+36.83/15 0.66} Kg4 {-M20/10 0.057}
57. Qxe4+ {+256.00/16 0.65} Kh3 {-M18/10 0.032}
58. Qg6 {+256.00/22 0.67} Kh2 {-M16/11 0.071}
59. Ke3 {+256.00/41 0.67} Kh1 {-M4/5 0.017}
60. Kf3 {+256.00/127 0.68} Kh2 {-M2/1 0.010}
61. Qg2# {+256.00/127 0.69}
{White mates}
1-0

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "2"]
[White "Komodo_3-64"]
[Black "Seer_200-64"]
[Result "1-0"]
[ECO "B20"]
[GameDuration "00:05:16"]
[GameEndTime "2021-04-27T14:08:37.937 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T14:03:21.784 Mitteleuropõische Sommerzeit"]
[Opening "Sicilian defense"]
[PlyCount "153"]
[Termination "time forfeit"]
[TimeControl "40/80"]

1. e4 {book} c5 {book}
2. d3 {book} Nc6 {book}
3. Be2 {book} g6 {book}
4. Be3 {+0.07/15 3.19} Nf6 {+0.17/12 1.29}
5. Nd2 {+0.17/15 2.12} e5 {+0.24/13 3.41}
6. Ngf3 {+0.03/13 1.34} d5 {+0.28/13 1.26}
7. c3 {+0.12/15 3.79} d4 {+0.34/13 1.27}
8. Bg5 {+0.03/15 2.58} h6 {+0.33/13 1.29}
9. Bh4 {+0.13/15 2.09} Bd6 {+0.43/11 1.29}
10. Nc4 {+0.19/14 2.26} Bc7 {+0.26/11 1.31}
11. cxd4 {+0.07/13 2.25} cxd4 {+0.43/12 1.32}
12. Rc1 {+0.25/14 2.63} Qe7 {+0.18/12 1.33}
13. O-O {+0.21/13 2.29} a5 {+0.27/13 1.34}
14. Qb3 {+0.42/12 2.30} g5 {+0.77/14 2.28}
15. Bg3 {+0.37/14 2.54} Nh5 {+0.99/12 1.35}
16. Nb6 {+0.39/13 3.83} Bxb6 {+0.96/14 1.36}
17. Qxb6 {+0.38/13 1.06} Nxg3 {+0.86/13 1.39}
18. fxg3 {+0.41/15 1.70} O-O {+0.89/14 2.79}
19. Qc5 {+0.37/16 3.14} Qxc5 {+0.43/15 1.60}
20. Rxc5 {+0.43/15 0.59} f6 {+0.26/19 12.82}
21. Nd2 {+0.37/15 1.41} Be6 {+0.71/19 13.78}
22. a3 {+0.40/15 0.43} Ra6 {+0.74/13 0.78}
23. Bd1 {+0.29/16 1.86} Kf7 {+0.53/12 0.78}
24. Nc4 {+0.45/16 2.03} Ke7 {+0.92/13 0.79}
25. Ba4 {+0.54/15 1.90} Bxc4 {+0.73/13 0.81}
26. Rxc4 {+0.49/16 0.96} Kd6 {+0.90/12 0.83}
27. Rf2 {+0.63/16 1.87} Ne7 {+0.90/14 1.31}
28. g4 {+0.32/15 2.59} Raa8 {+1.06/15 1.29}
29. Rc1 {+0.29/17 1.55} Nc8 {+1.06/14 0.84}
30. Bb5 {+0.26/18 2.76} Nb6 {+1.29/15 0.84}
31. a4 {+0.24/18 1.75} Rac8 {+0.54/16 3.30}
32. Rb1 {+0.17/18 4.26} Ke7 {+1.09/14 0.76}
33. g3 {+0.11/16 1.77} Rc5 {+0.73/13 0.80}
34. h3 {+0.09/17 2.48} Nc8 {+0.59/14 0.82}
35. Bc4 {+0.31/15 0.39} Nd6 {+0.47/16 2.07}
36. Bd5 {+0.20/17 1.64} b5 {+0.39/15 0.78}
37. axb5 {+0.36/15 0.60} Nxb5 {+0.30/13 0.80}
38. Rbf1 {+0.28/16 2.51} Nc7 {+1.01/14 1.16}
39. Bc4 {+0.23/17 1.83} Na8 {+0.92/14 0.84}
40. Ba2 {+0.15/17 1.62} Nb6 {+1.27/13 0.90}
41. Kg2 {+0.04/16 2.75} Rb5 {+1.98/15 1.30}
42. Ra1 {0.00/16 1.83} a4 {+1.95/15 1.88}
43. Rc1 {-0.04/16 1.77} Rc8 {+2.55/15 1.31}
44. Rxc8 {-0.05/18 2.87} Nxc8 {+2.87/15 1.33}
45. Rc2 {-0.05/20 1.92} Nb6 {+2.77/15 1.79}
46. Kf3 {-0.09/18 1.96} Kd6 {+2.33/15 1.34}
47. Ke2 {-0.21/18 2.17} Ra5 {+2.21/17 1.59}
48. Ke1 {-0.18/18 3.05} a3 {+2.67/17 3.03}
49. bxa3 {-0.20/19 2.83} Rxa3 {+2.82/17 6.50}
50. Ke2 {-0.19/17 0.58} Na4 {+3.64/13 1.24}
51. Bc4 {-0.18/17 1.36} Nc3+ {+3.19/15 1.54}
52. Kf3 {-0.18/17 0.56} Nd1 {+3.90/14 1.26}
53. Bb5 {-0.38/17 3.46} Rc3 {+4.55/15 1.27}
54. Rd2 {-0.39/18 3.48} Rc1 {+4.70/14 1.29}
55. Ra2 {-0.40/17 0.80} Kc5 {+8.80/19 16.78}
56. Be8 {-1.01/18 11.71} Kb4 {+9.09/15 2.53}
57. Bf7 {-1.21/16 6.58} Ne3 {+9.00/18 6.13}
58. h4 {-1.32/15 2.40} Kc3 {+10.41/21 6.75}
59. Ke2 {-1.10/13 1.43} Rc2+ {+11.95/22 8.81}
60. Rxc2+ {-0.92/15 0.28} Kxc2 {+10.94/19 1.51}
61. hxg5 {-1.48/17 1.38} fxg5 {+13.07/21 5.96}
62. Be6 {-2.27/18 1.25} Nd1 {+13.38/16 0.34}
63. Bc4 {-3.25/20 0.92} Nc3+ {+16.31/20 2.48}
64. Kf3 {-3.92/23 2.11} Kd2 {+21.85/20 3.35}
65. Ba6 {-4.73/23 1.16} Na4 {+22.03/19 0.58}
66. Bc4 {-5.54/23 0.56} Nb2 {+21.83/16 0.30}
67. Bd5 {-7.52/23 1.01} Nxd3 {+29.21/18 2.02}
68. Ba8 {-9.00/25 1.40} Nb2 {+30.25/18 1.23}
69. Bd5 {-9.00/21 0.13} d3 {+28.38/17 0.60}
70. Bc6 {-12.76/21 1.75} Kc1 {+28.01/11 0.046}
71. Bb5 {-9.00/15 0.22} d2 {+29.44/14 0.068}
72. Be2 {-M58/22 10.02} Nd1 {+29.69/15 0.46}
73. Kg2 {-M28/17 0.58} Kc2 {+29.64/14 0.20}
74. Bf3 {-13.27/17 0.35} Ne3+ {+27.52/12 0.13}
75. Kf2 {-5.82/17 0.24} Nc4 {+30.97/12 0.043}
76. Kg2 {-13.79/22 1.41} d1=Q {+31.62/8 0.014}
77. Bxd1+ {-7.97/16 0.20}
{Black loses on time}
1-0

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "2"]
[White "Seer_200-64"]
[Black "Komodo_3-64"]
[Result "0-1"]
[ECO "B20"]
[GameDuration "00:05:10"]
[GameEndTime "2021-04-27T14:13:49.327 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T14:08:38.997 Mitteleuropõische Sommerzeit"]
[Opening "Sicilian defense"]
[PlyCount "156"]
[Termination "time forfeit"]
[TimeControl "40/80"]

1. e4 {book} c5 {book}
2. d3 {book} Nc6 {book}
3. Be2 {book} g6 {book}
4. Nf3 {+0.09/12 1.29} Nf6 {-0.06/15 4.68}
5. c3 {-0.02/14 4.00} Bg7 {0.00/16 7.38}
6. O-O {+0.07/12 1.25} O-O {+0.05/13 0.34}
7. Re1 {+0.15/13 1.26} d5 {+0.22/12 1.10}
8. e5 {+0.14/13 1.28} Ne8 {+0.06/13 0.64}
9. d4 {+0.20/13 1.29} Nc7 {+0.01/14 3.25}
10. Be3 {+0.19/12 1.30} Bf5 {+0.02/13 1.17}
11. dxc5 {+0.12/14 1.30} Nxe5 {-0.01/14 1.86}
12. Nxe5 {+0.23/14 1.32} Bxe5 {+0.13/13 1.15}
13. Na3 {+0.42/14 1.34} Bg7 {+0.08/14 2.38}
14. Qd2 {+0.35/12 1.34} Qd7 {+0.17/14 2.04}
15. Rad1 {+0.20/13 1.37} Rad8 {+0.17/13 0.43}
16. f4 {+0.34/12 1.38} Qa4 {+0.07/13 1.99}
17. Kh1 {-0.47/14 2.99} Rfe8 {+0.31/14 8.10}
18. h3 {-0.08/16 5.74} f6 {+0.44/14 2.42}
19. Kg1 {+0.14/13 2.01} e5 {+0.61/12 0.79}
20. fxe5 {+0.28/12 1.28} fxe5 {+0.68/14 1.21}
21. Rf1 {-0.13/15 3.87} Rf8 {+0.68/13 0.88}
22. Bf3 {-0.13/14 1.24} Be6 {+0.73/13 1.76}
23. Nc2 {+0.04/13 1.54} a5 {+0.77/13 2.10}
24. a3 {+0.26/11 1.26} Qb3 {+0.63/13 3.01}
25. Rb1 {-0.47/13 1.60} Rd7 {+1.03/14 3.10}
26. Be2 {-1.30/14 3.23} Rxf1+ {+1.30/14 5.15}
27. Bxf1 {-1.89/18 11.29} Qa2 {+1.47/15 1.18}
28. Qc1 {-1.54/16 0.90} d4 {+1.48/15 1.92}
29. cxd4 {-1.53/14 0.93} exd4 {+1.47/15 0.63}
30. Bd2 {-1.43/13 0.93} d3 {+1.39/14 1.37}
31. Ne1 {-1.66/15 2.19} Bf5 {+1.11/14 6.20}
32. b4 {-2.37/16 3.00} Qd5 {+1.32/12 1.69}
33. bxa5 {-1.76/15 9.05} Ne6 {+1.22/11 1.18}
34. a6 {-0.66/12 0.40} Nxc5 {+0.91/11 1.50}
35. a7 {-0.01/13 0.52} Rd8 {+0.47/12 1.18}
36. Nf3 {+0.37/12 0.42} Be4 {+0.49/12 1.81}
37. Be3 {+1.21/12 0.45} Ne6 {+0.30/12 1.95}
38. Qd1 {+1.61/12 0.45} Bf8 {+0.32/10 0.71}
39. Qb3 {+2.10/12 0.48} Bc5 {+0.23/12 0.57}
40. Bxc5 {+1.77/16 3.61} Qxc5+ {+0.24/11 0.23}
41. Kh1 {0.00/16 2.20} Bd5 {+0.22/14 2.13}
42. Qxd3 {0.00/17 1.20} Qxa7 {+0.13/16 4.17}
43. Rb5 {0.00/18 2.31} Ba2 {+0.27/14 2.15}
44. Qe2 {+1.08/15 1.20} Qxa3 {+0.09/13 1.29}
45. Rb6 {0.00/16 1.47} Re8 {0.00/14 1.19}
46. Qd1 {0.00/18 1.22} Re7 {+0.23/14 2.24}
47. Qa1 {+1.46/15 1.29} Qa4 {+0.05/14 0.58}
48. Rb1 {+2.76/19 7.35} Bb3 {+0.05/16 3.02}
49. Qe5 {+3.46/20 11.36} Rf7 {+0.03/14 0.87}
50. Bd3 {+2.87/15 0.96} b6 {-0.11/13 2.69}
51. Be4 {+3.01/18 5.30} Re7 {-0.14/14 1.88}
52. Kh2 {+3.16/16 0.89} Rd7 {-0.16/13 2.12}
53. h4 {+3.08/19 11.62} Qc4 {-0.29/12 0.56}
54. h5 {+2.85/16 2.93} Bd1 {-0.29/12 2.88}
55. Ng5 {+2.66/16 0.64} Qc7 {-0.23/15 1.84}
56. Qxc7 {+2.35/15 0.64} Nxc7 {-0.18/15 1.02}
57. h6 {+2.47/19 8.34} b5 {-0.05/15 0.93}
58. Kg3 {+2.81/15 1.14} Ba4 {-0.16/17 2.49}
59. Rc1 {+2.43/15 0.47} Bb3 {-0.16/15 0.26}
60. Rc6 {+2.80/16 0.89} Re7 {-0.41/15 2.07}
61. Kh4 {+2.76/15 0.47} Bf7 {-0.41/16 1.79}
62. g4 {+3.94/18 4.66} Be8 {-0.51/16 3.91}
63. Rf6 {+4.85/14 0.42} Bf7 {-0.57/17 1.82}
64. Rb6 {+5.00/15 0.85} Be8 {-0.60/18 2.06}
65. Rb7 {+5.21/14 0.43} Kh8 {-0.62/16 0.44}
66. Rb8 {+6.86/18 4.26} Na6 {-0.62/21 2.59}
67. Rc8 {+4.86/16 1.10} Nc7 {-0.62/20 0.93}
68. Bc2 {+6.70/17 2.02} Nd5 {-0.62/14 0.38}
69. Rd8 {+5.94/14 0.14} Nf6 {-0.62/15 1.02}
70. Bb3 {+6.09/14 0.14} b4 {-0.62/16 1.21}
71. Be6 {+7.22/16 1.49} b3 {-2.48/14 2.31}
72. Bxb3 {+7.15/12 0.087} Re1 {-2.41/17 3.89}
73. Be6 {+6.98/13 0.077} Rh1+ {-2.48/17 2.17}
74. Kg3 {+8.11/14 0.35} Rxh6 {-1.69/16 2.82}
75. Bf7 {+9.30/14 0.61} Rh1 {-1.60/16 1.63}
76. Bxe8 {+9.32/10 0.025} Nxe8 {-1.61/15 1.02}
77. Rxe8+ {+8.96/10 0.24} Kg7 {-1.62/14 0.29}
78. Re7+ {+8.08/11 0.066} Kg8 {-1.60/17 2.81}
{White loses on time}
0-1[/pgn]

Code: Select all

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "2"]
[White "Komodo_3-64"]
[Black "Seer_200-64"]
[Result "1-0"]
[ECO "B20"]
[GameDuration "00:05:16"]
[GameEndTime "2021-04-27T14:08:37.937 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T14:03:21.784 Mitteleuropõische Sommerzeit"]
[Opening "Sicilian defense"]
[PlyCount "153"]
[Termination "time forfeit"]
[TimeControl "40/80"]

1. e4 {book} c5 {book}
2. d3 {book} Nc6 {book}
3. Be2 {book} g6 {book}
4. Be3 {+0.07/15 3.19} Nf6 {+0.17/12 1.29}
5. Nd2 {+0.17/15 2.12} e5 {+0.24/13 3.41}
6. Ngf3 {+0.03/13 1.34} d5 {+0.28/13 1.26}
7. c3 {+0.12/15 3.79} d4 {+0.34/13 1.27}
8. Bg5 {+0.03/15 2.58} h6 {+0.33/13 1.29}
9. Bh4 {+0.13/15 2.09} Bd6 {+0.43/11 1.29}
10. Nc4 {+0.19/14 2.26} Bc7 {+0.26/11 1.31}
11. cxd4 {+0.07/13 2.25} cxd4 {+0.43/12 1.32}
12. Rc1 {+0.25/14 2.63} Qe7 {+0.18/12 1.33}
13. O-O {+0.21/13 2.29} a5 {+0.27/13 1.34}
14. Qb3 {+0.42/12 2.30} g5 {+0.77/14 2.28}
15. Bg3 {+0.37/14 2.54} Nh5 {+0.99/12 1.35}
16. Nb6 {+0.39/13 3.83} Bxb6 {+0.96/14 1.36}
17. Qxb6 {+0.38/13 1.06} Nxg3 {+0.86/13 1.39}
18. fxg3 {+0.41/15 1.70} O-O {+0.89/14 2.79}
19. Qc5 {+0.37/16 3.14} Qxc5 {+0.43/15 1.60}
20. Rxc5 {+0.43/15 0.59} f6 {+0.26/19 12.82}
21. Nd2 {+0.37/15 1.41} Be6 {+0.71/19 13.78}
22. a3 {+0.40/15 0.43} Ra6 {+0.74/13 0.78}
23. Bd1 {+0.29/16 1.86} Kf7 {+0.53/12 0.78}
24. Nc4 {+0.45/16 2.03} Ke7 {+0.92/13 0.79}
25. Ba4 {+0.54/15 1.90} Bxc4 {+0.73/13 0.81}
26. Rxc4 {+0.49/16 0.96} Kd6 {+0.90/12 0.83}
27. Rf2 {+0.63/16 1.87} Ne7 {+0.90/14 1.31}
28. g4 {+0.32/15 2.59} Raa8 {+1.06/15 1.29}
29. Rc1 {+0.29/17 1.55} Nc8 {+1.06/14 0.84}
30. Bb5 {+0.26/18 2.76} Nb6 {+1.29/15 0.84}
31. a4 {+0.24/18 1.75} Rac8 {+0.54/16 3.30}
32. Rb1 {+0.17/18 4.26} Ke7 {+1.09/14 0.76}
33. g3 {+0.11/16 1.77} Rc5 {+0.73/13 0.80}
34. h3 {+0.09/17 2.48} Nc8 {+0.59/14 0.82}
35. Bc4 {+0.31/15 0.39} Nd6 {+0.47/16 2.07}
36. Bd5 {+0.20/17 1.64} b5 {+0.39/15 0.78}
37. axb5 {+0.36/15 0.60} Nxb5 {+0.30/13 0.80}
38. Rbf1 {+0.28/16 2.51} Nc7 {+1.01/14 1.16}
39. Bc4 {+0.23/17 1.83} Na8 {+0.92/14 0.84}
40. Ba2 {+0.15/17 1.62} Nb6 {+1.27/13 0.90}
41. Kg2 {+0.04/16 2.75} Rb5 {+1.98/15 1.30}
42. Ra1 {0.00/16 1.83} a4 {+1.95/15 1.88}
43. Rc1 {-0.04/16 1.77} Rc8 {+2.55/15 1.31}
44. Rxc8 {-0.05/18 2.87} Nxc8 {+2.87/15 1.33}
45. Rc2 {-0.05/20 1.92} Nb6 {+2.77/15 1.79}
46. Kf3 {-0.09/18 1.96} Kd6 {+2.33/15 1.34}
47. Ke2 {-0.21/18 2.17} Ra5 {+2.21/17 1.59}
48. Ke1 {-0.18/18 3.05} a3 {+2.67/17 3.03}
49. bxa3 {-0.20/19 2.83} Rxa3 {+2.82/17 6.50}
50. Ke2 {-0.19/17 0.58} Na4 {+3.64/13 1.24}
51. Bc4 {-0.18/17 1.36} Nc3+ {+3.19/15 1.54}
52. Kf3 {-0.18/17 0.56} Nd1 {+3.90/14 1.26}
53. Bb5 {-0.38/17 3.46} Rc3 {+4.55/15 1.27}
54. Rd2 {-0.39/18 3.48} Rc1 {+4.70/14 1.29}
55. Ra2 {-0.40/17 0.80} Kc5 {+8.80/19 16.78}
56. Be8 {-1.01/18 11.71} Kb4 {+9.09/15 2.53}
57. Bf7 {-1.21/16 6.58} Ne3 {+9.00/18 6.13}
58. h4 {-1.32/15 2.40} Kc3 {+10.41/21 6.75}
59. Ke2 {-1.10/13 1.43} Rc2+ {+11.95/22 8.81}
60. Rxc2+ {-0.92/15 0.28} Kxc2 {+10.94/19 1.51}
61. hxg5 {-1.48/17 1.38} fxg5 {+13.07/21 5.96}
62. Be6 {-2.27/18 1.25} Nd1 {+13.38/16 0.34}
63. Bc4 {-3.25/20 0.92} Nc3+ {+16.31/20 2.48}
64. Kf3 {-3.92/23 2.11} Kd2 {+21.85/20 3.35}
65. Ba6 {-4.73/23 1.16} Na4 {+22.03/19 0.58}
66. Bc4 {-5.54/23 0.56} Nb2 {+21.83/16 0.30}
67. Bd5 {-7.52/23 1.01} Nxd3 {+29.21/18 2.02}
68. Ba8 {-9.00/25 1.40} Nb2 {+30.25/18 1.23}
69. Bd5 {-9.00/21 0.13} d3 {+28.38/17 0.60}
70. Bc6 {-12.76/21 1.75} Kc1 {+28.01/11 0.046}
71. Bb5 {-9.00/15 0.22} d2 {+29.44/14 0.068}
72. Be2 {-M58/22 10.02} Nd1 {+29.69/15 0.46}
73. Kg2 {-M28/17 0.58} Kc2 {+29.64/14 0.20}
74. Bf3 {-13.27/17 0.35} Ne3+ {+27.52/12 0.13}
75. Kf2 {-5.82/17 0.24} Nc4 {+30.97/12 0.043}
76. Kg2 {-13.79/22 1.41} d1=Q {+31.62/8 0.014}
77. Bxd1+ {-7.97/16 0.20}
{Black loses on time}
1-0

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "2"]
[White "Seer_200-64"]
[Black "Komodo_3-64"]
[Result "0-1"]
[ECO "B20"]
[GameDuration "00:05:10"]
[GameEndTime "2021-04-27T14:13:49.327 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T14:08:38.997 Mitteleuropõische Sommerzeit"]
[Opening "Sicilian defense"]
[PlyCount "156"]
[Termination "time forfeit"]
[TimeControl "40/80"]

1. e4 {book} c5 {book}
2. d3 {book} Nc6 {book}
3. Be2 {book} g6 {book}
4. Nf3 {+0.09/12 1.29} Nf6 {-0.06/15 4.68}
5. c3 {-0.02/14 4.00} Bg7 {0.00/16 7.38}
6. O-O {+0.07/12 1.25} O-O {+0.05/13 0.34}
7. Re1 {+0.15/13 1.26} d5 {+0.22/12 1.10}
8. e5 {+0.14/13 1.28} Ne8 {+0.06/13 0.64}
9. d4 {+0.20/13 1.29} Nc7 {+0.01/14 3.25}
10. Be3 {+0.19/12 1.30} Bf5 {+0.02/13 1.17}
11. dxc5 {+0.12/14 1.30} Nxe5 {-0.01/14 1.86}
12. Nxe5 {+0.23/14 1.32} Bxe5 {+0.13/13 1.15}
13. Na3 {+0.42/14 1.34} Bg7 {+0.08/14 2.38}
14. Qd2 {+0.35/12 1.34} Qd7 {+0.17/14 2.04}
15. Rad1 {+0.20/13 1.37} Rad8 {+0.17/13 0.43}
16. f4 {+0.34/12 1.38} Qa4 {+0.07/13 1.99}
17. Kh1 {-0.47/14 2.99} Rfe8 {+0.31/14 8.10}
18. h3 {-0.08/16 5.74} f6 {+0.44/14 2.42}
19. Kg1 {+0.14/13 2.01} e5 {+0.61/12 0.79}
20. fxe5 {+0.28/12 1.28} fxe5 {+0.68/14 1.21}
21. Rf1 {-0.13/15 3.87} Rf8 {+0.68/13 0.88}
22. Bf3 {-0.13/14 1.24} Be6 {+0.73/13 1.76}
23. Nc2 {+0.04/13 1.54} a5 {+0.77/13 2.10}
24. a3 {+0.26/11 1.26} Qb3 {+0.63/13 3.01}
25. Rb1 {-0.47/13 1.60} Rd7 {+1.03/14 3.10}
26. Be2 {-1.30/14 3.23} Rxf1+ {+1.30/14 5.15}
27. Bxf1 {-1.89/18 11.29} Qa2 {+1.47/15 1.18}
28. Qc1 {-1.54/16 0.90} d4 {+1.48/15 1.92}
29. cxd4 {-1.53/14 0.93} exd4 {+1.47/15 0.63}
30. Bd2 {-1.43/13 0.93} d3 {+1.39/14 1.37}
31. Ne1 {-1.66/15 2.19} Bf5 {+1.11/14 6.20}
32. b4 {-2.37/16 3.00} Qd5 {+1.32/12 1.69}
33. bxa5 {-1.76/15 9.05} Ne6 {+1.22/11 1.18}
34. a6 {-0.66/12 0.40} Nxc5 {+0.91/11 1.50}
35. a7 {-0.01/13 0.52} Rd8 {+0.47/12 1.18}
36. Nf3 {+0.37/12 0.42} Be4 {+0.49/12 1.81}
37. Be3 {+1.21/12 0.45} Ne6 {+0.30/12 1.95}
38. Qd1 {+1.61/12 0.45} Bf8 {+0.32/10 0.71}
39. Qb3 {+2.10/12 0.48} Bc5 {+0.23/12 0.57}
40. Bxc5 {+1.77/16 3.61} Qxc5+ {+0.24/11 0.23}
41. Kh1 {0.00/16 2.20} Bd5 {+0.22/14 2.13}
42. Qxd3 {0.00/17 1.20} Qxa7 {+0.13/16 4.17}
43. Rb5 {0.00/18 2.31} Ba2 {+0.27/14 2.15}
44. Qe2 {+1.08/15 1.20} Qxa3 {+0.09/13 1.29}
45. Rb6 {0.00/16 1.47} Re8 {0.00/14 1.19}
46. Qd1 {0.00/18 1.22} Re7 {+0.23/14 2.24}
47. Qa1 {+1.46/15 1.29} Qa4 {+0.05/14 0.58}
48. Rb1 {+2.76/19 7.35} Bb3 {+0.05/16 3.02}
49. Qe5 {+3.46/20 11.36} Rf7 {+0.03/14 0.87}
50. Bd3 {+2.87/15 0.96} b6 {-0.11/13 2.69}
51. Be4 {+3.01/18 5.30} Re7 {-0.14/14 1.88}
52. Kh2 {+3.16/16 0.89} Rd7 {-0.16/13 2.12}
53. h4 {+3.08/19 11.62} Qc4 {-0.29/12 0.56}
54. h5 {+2.85/16 2.93} Bd1 {-0.29/12 2.88}
55. Ng5 {+2.66/16 0.64} Qc7 {-0.23/15 1.84}
56. Qxc7 {+2.35/15 0.64} Nxc7 {-0.18/15 1.02}
57. h6 {+2.47/19 8.34} b5 {-0.05/15 0.93}
58. Kg3 {+2.81/15 1.14} Ba4 {-0.16/17 2.49}
59. Rc1 {+2.43/15 0.47} Bb3 {-0.16/15 0.26}
60. Rc6 {+2.80/16 0.89} Re7 {-0.41/15 2.07}
61. Kh4 {+2.76/15 0.47} Bf7 {-0.41/16 1.79}
62. g4 {+3.94/18 4.66} Be8 {-0.51/16 3.91}
63. Rf6 {+4.85/14 0.42} Bf7 {-0.57/17 1.82}
64. Rb6 {+5.00/15 0.85} Be8 {-0.60/18 2.06}
65. Rb7 {+5.21/14 0.43} Kh8 {-0.62/16 0.44}
66. Rb8 {+6.86/18 4.26} Na6 {-0.62/21 2.59}
67. Rc8 {+4.86/16 1.10} Nc7 {-0.62/20 0.93}
68. Bc2 {+6.70/17 2.02} Nd5 {-0.62/14 0.38}
69. Rd8 {+5.94/14 0.14} Nf6 {-0.62/15 1.02}
70. Bb3 {+6.09/14 0.14} b4 {-0.62/16 1.21}
71. Be6 {+7.22/16 1.49} b3 {-2.48/14 2.31}
72. Bxb3 {+7.15/12 0.087} Re1 {-2.41/17 3.89}
73. Be6 {+6.98/13 0.077} Rh1+ {-2.48/17 2.17}
74. Kg3 {+8.11/14 0.35} Rxh6 {-1.69/16 2.82}
75. Bf7 {+9.30/14 0.61} Rh1 {-1.60/16 1.63}
76. Bxe8 {+9.32/10 0.025} Nxe8 {-1.61/15 1.02}
77. Rxe8+ {+8.96/10 0.24} Kg7 {-1.62/14 0.29}
78. Re7+ {+8.08/11 0.066} Kg8 {-1.60/17 2.81}
{White loses on time}
0-1
https://rwbc-chess.de

trollwatch:
Chessqueen + chessica + AlexChess + Eduard + Sylwy
Top
Archimedes
Posts: 135
Joined: Tue Mar 05, 2019 3:43 pm
Full name: Archimedes

Re: Seer 2.0.0

Post by Archimedes »

I've had problems, compiling Seer for Android. In simd.h there are the following lines:

Code: Select all

#ifdef __AVX__
constexpr size_t alignment = 32;
#elif __SSE__
constexpr size_t alignment = 16;
#endif
When compiling for ARM architecture constexpr variable will not be initialized and therefore an error occourred.

Changing to

Code: Select all

#ifdef __AVX__
constexpr size_t alignment = 32;
#else
constexpr size_t alignment = 16;
#endif
solves the problem for me (hopefully alignment 16 is correct for Android builds).
Top
connor_mcmonigle
Posts: 532
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Seer 2.0.0

Post by connor_mcmonigle »

Guenther wrote: ↑Tue Apr 27, 2021 2:37 pm
connor_mcmonigle wrote: ↑Tue Apr 27, 2021 12:20 pm I've released Seer 2.0.0 here: https://github.com/connormcmonigle/seer ... tag/v2.0.0.
...

Qualitatively, relative to my older networks trained on Stockfish NNUE data, my semi-supervised training based networks are largely inferior positionally and I find their pawn play in particular to be lacking. However, they prove massively superior at developing long term attacks and also exhibit an advantage in the endgame owing to the EGTBs their understanding of chess was iteratively constructed from.
...
Hi Connor,

Thanks for the new release.

I just let it play a few games vs. Komodo3 which is already around 3100 at CCRLBlitz and already in the first four games I could see
some fine endgame improvement and also some long term attacking style.
For the endgame just look how it saved game 1 and especially game 3, where it very soon obviously knew it will end up with a superior
Knight vs. Bishop endgame! (it should be noted though that Komodo3 was not yet able to use tablebases, this came in later versions)

OTH I noticed that the time management for mps tc is not safe and it lost two of four (totally won) games on time before the second session ended.

I have added the pgn in column style for those lost on time games at the end again, as here it can be easily seen how fast it was forced to play
from around move 72-75 and didn't succeed. It also plays already move 39-40 in extreme time trouble, but here it succeeded always.
This was in CuteChessDev at 40/80s (with no extra time margin defined as usual) with 64MB hash for both and with adjudication nearly off,
for finding more bugs ;-)

[pgn][Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "1"]
[White "Komodo_3-64"]
[Black "Seer_200-64"]
[Result "1/2-1/2"]
[ECO "A06"]
[GameDuration "00:05:20"]
[GameEndTime "2021-04-27T13:59:30.574 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T13:54:10.334 Mitteleuropõische Sommerzeit"]
[Opening "Reti"]
[PlyCount "162"]
[Termination "adjudication"]
[TimeControl "40/80"]
[Variation "Nimzovich-Larsen attack"]

1. Nf3 {book} d5 {book}
2. b3 {book} Bg4 {book}
3. e3 {book} e5 {book}
4. h3 {+0.06/15 3.84} Bxf3 {-0.32/12 1.27}
5. Qxf3 {+0.03/15 0.75} e4 {-0.19/13 1.88}
6. Qg3 {+0.16/15 2.30} Nc6 {-0.32/12 1.29}
7. Bb5 {+0.31/14 1.18} Qd7 {-0.07/11 1.31}
8. Ba3 {+0.32/14 2.77} h5 {-0.14/13 1.31}
9. c4 {+0.39/13 2.29} Nf6 {-0.33/15 4.60}
10. Bxc6 {+0.45/14 2.80} bxc6 {-0.21/13 1.28}
11. Qe5+ {+0.41/16 3.09} Be7 {-0.50/17 5.41}
12. O-O {+0.41/16 2.72} Rc8 {-0.85/17 2.56}
13. Nc3 {+0.41/15 2.03} Qe6 {-0.17/13 1.20}
14. Qxe6 {+0.41/14 1.15} fxe6 {-0.56/15 8.33}
15. Bb2 {+0.38/15 4.54} Kf7 {-0.37/13 1.08}
16. Ne2 {+0.60/13 1.13} c5 {-0.76/14 3.17}
17. cxd5 {+0.97/15 4.04} exd5 {-0.72/14 2.54}
18. f3 {+0.92/15 3.38} Rhd8 {-1.24/12 1.03}
19. fxe4 {+1.02/14 1.85} dxe4 {-1.47/14 2.11}
20. Rad1 {+1.03/16 1.71} Bd6 {-1.06/14 1.22}
21. Bxf6 {+1.00/14 1.75} gxf6 {-2.36/18 5.62}
22. Rf5 {+0.86/15 2.45} Be5 {-1.26/15 1.70}
23. Rxh5 {+0.81/16 1.58} Rd3 {-1.44/14 1.60}
24. Kf2 {+0.87/16 2.91} Rcd8 {-1.80/14 0.95}
25. Ke1 {+0.71/16 1.93} Rg8 {-1.46/12 0.92}
26. Rh7+ {+0.71/15 1.16} Ke6 {-1.15/14 1.03}
27. g4 {+0.86/16 1.71} Rgd8 {-1.56/14 1.00}
28. Rh5 {+0.84/16 1.83} Kf7 {-1.61/13 0.95}
29. Nc1 {+0.76/15 2.25} R3d5 {-2.15/17 11.20}
30. Ke2 {+0.70/13 1.00} a6 {-1.78/16 2.59}
31. d3 {+0.82/14 1.27} Ke6 {-2.12/13 0.50}
32. Rh7 {+0.85/15 2.56} a5 {-2.67/13 0.64}
33. a4 {+0.90/14 1.80} Bg3 {-1.56/17 5.39}
34. h4 {+0.86/14 1.58} c6 {-2.71/14 2.54}
35. h5 {+1.27/14 1.04} f5 {-2.31/12 0.21}
36. Rh6+ {+0.82/13 2.54} Ke7 {-1.95/11 0.11}
37. Rg6 {+1.03/14 3.50} f4 {-2.57/11 0.13}
38. exf4 {+0.87/14 1.61} Bxf4 {-1.39/14 1.02}
39. Rf1 {+0.87/14 1.00} Bxc1 {-1.68/11 0.062}
40. Rxc1 {+0.82/13 0.81} exd3+ {-1.50/8 0.033}
41. Kd2 {+0.98/14 0.87} Rf8 {-2.38/13 1.20}
42. Rxc6 {+1.03/15 2.17} Rf2+ {-3.23/15 1.21}
43. Kc3 {+1.03/16 2.14} Kd7 {-3.71/16 6.83}
44. Ra6 {+1.48/13 1.43} Rh2 {-3.21/16 4.54}
45. Rxa5 {+1.50/13 1.39} Kc7 {-3.61/16 14.45}
46. Ra7+ {+1.51/15 1.60} Kb6 {-4.69/14 0.87}
47. Rg7 {+1.55/15 1.57} Ka5 {-3.61/18 11.62}
48. Rg8 {+1.86/16 5.51} Ka6 {-4.32/13 0.70}
49. Rh8 {+1.86/14 2.82} Ka7 {-4.23/16 9.24}
50. Rh6 {+1.78/15 2.07} Rd4 {-3.78/14 0.56}
51. g5 {+1.84/14 1.26} d2 {-4.06/13 0.55}
52. Rd1 {+2.06/15 1.79} Rh3+ {-1.66/17 7.30}
53. Kb2 {+1.85/16 11.50} Rdd3 {-1.59/14 0.48}
54. Rh7+ {+0.92/12 1.66} Kb6 {-0.99/13 0.51}
55. a5+ {+0.79/13 1.71} Kxa5 {-0.46/15 1.26}
56. Rb7 {+0.55/14 2.23} Ka6 {-0.06/15 2.33}
57. Rb8 {+0.55/16 2.40} Ka7 {-0.06/12 0.39}
58. Rb5 {+0.14/16 6.41} Ka6 {-0.06/13 0.40}
59. Rxc5 {+0.08/15 0.79} Rxb3+ {-0.05/12 0.40}
60. Kc2 {+0.08/15 0.56} Ra3 {-0.03/13 0.40}
61. Rc8 {+0.08/13 1.72} Kb7 {-0.01/12 0.40}
62. Rc4 {+0.05/14 1.39} Ra2+ {0.00/11 0.42}
63. Kb1 {+0.44/11 0.28} Ra5 {-0.42/13 0.43}
64. g6 {+0.08/13 1.70} Rhxh5 {-0.41/11 0.42}
65. Rf4 {+0.05/14 1.39} Rhg5 {-0.10/12 0.76}
66. Rf7+ {+0.58/11 0.55} Kb6 {-0.28/12 0.42}
67. Rf6+ {+0.56/14 0.87} Kb5 {-0.08/13 0.44}
68. Rxd2 {+0.56/12 0.65} Rg1+ {-0.02/13 0.43}
69. Kb2 {+0.05/13 0.57} Ra6 {-0.02/12 0.44}
70. Rxa6 {+0.05/15 1.87} Kxa6 {-0.02/13 0.47}
71. Rd6+ {+0.04/18 2.27} Kb5 {-0.01/13 0.47}
72. Kc3 {+0.04/19 1.55} Kc5 {0.00/12 0.47}
73. Ra6 {+0.05/18 2.01} Kd5 {0.00/12 0.50}
74. Kd3 {+0.04/18 1.62} Ke5 {0.00/13 0.50}
75. Kc4 {+0.04/14 0.36} Kf5 {0.00/12 0.52}
76. Kd5 {+0.04/15 1.23} Rxg6 {0.00/12 0.54}
77. Ra8 {+0.04/13 0.61} Kf6 {0.00/12 0.55}
78. Rf8+ {+0.04/14 3.68} Kg7 {0.00/12 0.57}
79. Rf2 {+0.03/16 3.24} Rg5+ {0.00/12 0.61}
80. Ke6 {+0.04/17 0.56} Rg3 {0.00/13 0.63}
81. Ra2 {+0.04/16 2.62} Kg6 {0.00/14 1.28}
{Draw by adjudication: user decision}
1/2-1/2

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "1"]
[White "Seer_200-64"]
[Black "Komodo_3-64"]
[Result "1-0"]
[ECO "A06"]
[GameDuration "00:03:49"]
[GameEndTime "2021-04-27T14:03:20.754 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T13:59:31.634 Mitteleuropõische Sommerzeit"]
[Opening "Reti"]
[PlyCount "121"]
[TimeControl "40/80"]
[Variation "Nimzovich-Larsen attack"]

1. Nf3 {book} d5 {book}
2. b3 {book} Bg4 {book}
3. e3 {book} e5 {book}
4. h3 {+0.32/14 1.67} Bxf3 {-0.13/16 3.69}
5. Qxf3 {+0.31/12 1.27} Nc6 {-0.13/13 0.26}
6. Bb2 {+0.46/13 1.29} Nf6 {-0.12/15 3.40}
7. g4 {+0.64/13 1.31} e4 {-0.05/14 3.64}
8. Qg2 {+0.75/13 1.63} Nd7 {-0.13/15 4.60}
9. Nc3 {+1.69/15 4.01} Nce5 {-0.27/15 2.86}
10. O-O-O {+1.54/13 1.28} c6 {-0.35/15 1.74}
11. d3 {+1.25/16 4.08} exd3 {-0.32/12 1.05}
12. f4 {+1.71/15 1.81} dxc2 {-0.29/13 1.75}
13. Qxc2 {+1.53/16 5.86} Ng6 {-0.39/15 6.52}
14. Kb1 {+1.76/16 4.41} a5 {-0.55/15 10.44}
15. h4 {+2.26/14 1.49} h5 {-0.64/12 1.98}
16. g5 {+2.62/14 1.92} Qe7 {-0.76/13 1.57}
17. f5 {+5.28/18 14.12} Nge5 {-0.62/13 1.65}
18. Nxd5 {+5.17/14 0.81} cxd5 {-0.52/11 0.26}
19. Rxd5 {+5.45/13 0.83} Nc6 {-1.55/12 0.72}
20. Bb5 {+5.78/18 10.40} f6 {-2.44/13 2.57}
21. Rhd1 {+5.87/15 1.13} Rd8 {-3.59/15 7.96}
22. Bc3 {+6.96/15 0.58} Rc8 {-3.48/13 1.04}
23. Rxd7 {+8.06/15 1.45} Qxd7 {-3.56/13 0.54}
24. Rxd7 {+10.54/18 6.73} Kxd7 {-3.46/14 0.72}
25. Bxa5 {+11.72/13 0.43} Bd6 {-4.58/14 1.31}
26. Qd3 {+12.33/14 0.70} Rcd8 {-4.86/15 0.55}
27. g6 {+12.89/14 0.39} Rhe8 {-6.20/15 1.63}
28. a4 {+12.82/15 0.55} Kc8 {-5.37/18 1.24}
29. Bxd8 {+13.43/15 0.41} Rxd8 {-3.84/14 0.40}
30. Bxc6 {+14.13/14 0.41} bxc6 {-3.92/15 0.15}
31. Qa6+ {+14.47/14 0.41} Kd7 {-4.12/15 0.16}
32. a5 {+14.90/12 0.41} Rb8 {-4.74/17 1.27}
33. Qc4 {+16.25/18 4.58} Re8 {-5.85/18 3.12}
34. Kc2 {+16.99/15 2.15} Re5 {-4.82/16 1.35}
35. b4 {+16.24/11 0.13} Bc7 {-5.21/16 1.52}
36. Qg8 {+18.42/13 0.93} Re7 {-6.24/18 1.49}
37. a6 {+16.23/12 0.12} Bb6 {-6.95/18 2.36}
38. Qa8 {+17.88/8 0.029} Bxe3 {-6.82/17 1.38}
39. Kd3 {+16.75/11 0.082} Bg1 {-6.34/17 1.66}
40. a7 {+17.01/11 0.049} Bxa7 {-6.74/17 1.02}
41. Qxa7+ {+19.71/15 1.19} Kd6 {-6.82/18 0.50}
42. Qb8+ {+29.19/21 16.12} Kd7 {-7.38/19 1.78}
43. Kd4 {+31.51/20 13.11} Re1 {-9.78/17 2.68}
44. Kc5 {+28.38/16 0.80} Re5+ {-9.41/15 1.09}
45. Qxe5 {+36.83/19 10.65} fxe5 {-10.97/18 0.66}
46. f6 {+36.83/17 3.35} gxf6 {-10.97/18 0.62}
47. g7 {+36.83/14 1.19} Ke6 {-10.97/14 0.19}
48. g8=Q+ {+36.83/12 0.60} Kf5 {-10.97/16 0.37}
49. Kxc6 {+36.83/12 0.61} e4 {-M22/12 0.48}
50. Kd5 {+36.83/13 0.60} Kf4 {-M26/13 0.80}
51. Kd4 {+36.83/14 0.63} Kf3 {-M26/15 1.50}
52. Qb3+ {+36.83/14 0.63} Kg4 {-M22/14 0.65}
53. Qd1+ {+36.83/13 0.62} Kf5 {-M26/15 1.34}
54. Qxh5+ {+36.83/13 0.63} Kf4 {-M24/13 0.070}
55. Qh6+ {+36.83/15 0.64} Kf5 {-M22/12 0.061}
56. Qh7+ {+36.83/15 0.66} Kg4 {-M20/10 0.057}
57. Qxe4+ {+256.00/16 0.65} Kh3 {-M18/10 0.032}
58. Qg6 {+256.00/22 0.67} Kh2 {-M16/11 0.071}
59. Ke3 {+256.00/41 0.67} Kh1 {-M4/5 0.017}
60. Kf3 {+256.00/127 0.68} Kh2 {-M2/1 0.010}
61. Qg2# {+256.00/127 0.69}
{White mates}
1-0

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "2"]
[White "Komodo_3-64"]
[Black "Seer_200-64"]
[Result "1-0"]
[ECO "B20"]
[GameDuration "00:05:16"]
[GameEndTime "2021-04-27T14:08:37.937 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T14:03:21.784 Mitteleuropõische Sommerzeit"]
[Opening "Sicilian defense"]
[PlyCount "153"]
[Termination "time forfeit"]
[TimeControl "40/80"]

1. e4 {book} c5 {book}
2. d3 {book} Nc6 {book}
3. Be2 {book} g6 {book}
4. Be3 {+0.07/15 3.19} Nf6 {+0.17/12 1.29}
5. Nd2 {+0.17/15 2.12} e5 {+0.24/13 3.41}
6. Ngf3 {+0.03/13 1.34} d5 {+0.28/13 1.26}
7. c3 {+0.12/15 3.79} d4 {+0.34/13 1.27}
8. Bg5 {+0.03/15 2.58} h6 {+0.33/13 1.29}
9. Bh4 {+0.13/15 2.09} Bd6 {+0.43/11 1.29}
10. Nc4 {+0.19/14 2.26} Bc7 {+0.26/11 1.31}
11. cxd4 {+0.07/13 2.25} cxd4 {+0.43/12 1.32}
12. Rc1 {+0.25/14 2.63} Qe7 {+0.18/12 1.33}
13. O-O {+0.21/13 2.29} a5 {+0.27/13 1.34}
14. Qb3 {+0.42/12 2.30} g5 {+0.77/14 2.28}
15. Bg3 {+0.37/14 2.54} Nh5 {+0.99/12 1.35}
16. Nb6 {+0.39/13 3.83} Bxb6 {+0.96/14 1.36}
17. Qxb6 {+0.38/13 1.06} Nxg3 {+0.86/13 1.39}
18. fxg3 {+0.41/15 1.70} O-O {+0.89/14 2.79}
19. Qc5 {+0.37/16 3.14} Qxc5 {+0.43/15 1.60}
20. Rxc5 {+0.43/15 0.59} f6 {+0.26/19 12.82}
21. Nd2 {+0.37/15 1.41} Be6 {+0.71/19 13.78}
22. a3 {+0.40/15 0.43} Ra6 {+0.74/13 0.78}
23. Bd1 {+0.29/16 1.86} Kf7 {+0.53/12 0.78}
24. Nc4 {+0.45/16 2.03} Ke7 {+0.92/13 0.79}
25. Ba4 {+0.54/15 1.90} Bxc4 {+0.73/13 0.81}
26. Rxc4 {+0.49/16 0.96} Kd6 {+0.90/12 0.83}
27. Rf2 {+0.63/16 1.87} Ne7 {+0.90/14 1.31}
28. g4 {+0.32/15 2.59} Raa8 {+1.06/15 1.29}
29. Rc1 {+0.29/17 1.55} Nc8 {+1.06/14 0.84}
30. Bb5 {+0.26/18 2.76} Nb6 {+1.29/15 0.84}
31. a4 {+0.24/18 1.75} Rac8 {+0.54/16 3.30}
32. Rb1 {+0.17/18 4.26} Ke7 {+1.09/14 0.76}
33. g3 {+0.11/16 1.77} Rc5 {+0.73/13 0.80}
34. h3 {+0.09/17 2.48} Nc8 {+0.59/14 0.82}
35. Bc4 {+0.31/15 0.39} Nd6 {+0.47/16 2.07}
36. Bd5 {+0.20/17 1.64} b5 {+0.39/15 0.78}
37. axb5 {+0.36/15 0.60} Nxb5 {+0.30/13 0.80}
38. Rbf1 {+0.28/16 2.51} Nc7 {+1.01/14 1.16}
39. Bc4 {+0.23/17 1.83} Na8 {+0.92/14 0.84}
40. Ba2 {+0.15/17 1.62} Nb6 {+1.27/13 0.90}
41. Kg2 {+0.04/16 2.75} Rb5 {+1.98/15 1.30}
42. Ra1 {0.00/16 1.83} a4 {+1.95/15 1.88}
43. Rc1 {-0.04/16 1.77} Rc8 {+2.55/15 1.31}
44. Rxc8 {-0.05/18 2.87} Nxc8 {+2.87/15 1.33}
45. Rc2 {-0.05/20 1.92} Nb6 {+2.77/15 1.79}
46. Kf3 {-0.09/18 1.96} Kd6 {+2.33/15 1.34}
47. Ke2 {-0.21/18 2.17} Ra5 {+2.21/17 1.59}
48. Ke1 {-0.18/18 3.05} a3 {+2.67/17 3.03}
49. bxa3 {-0.20/19 2.83} Rxa3 {+2.82/17 6.50}
50. Ke2 {-0.19/17 0.58} Na4 {+3.64/13 1.24}
51. Bc4 {-0.18/17 1.36} Nc3+ {+3.19/15 1.54}
52. Kf3 {-0.18/17 0.56} Nd1 {+3.90/14 1.26}
53. Bb5 {-0.38/17 3.46} Rc3 {+4.55/15 1.27}
54. Rd2 {-0.39/18 3.48} Rc1 {+4.70/14 1.29}
55. Ra2 {-0.40/17 0.80} Kc5 {+8.80/19 16.78}
56. Be8 {-1.01/18 11.71} Kb4 {+9.09/15 2.53}
57. Bf7 {-1.21/16 6.58} Ne3 {+9.00/18 6.13}
58. h4 {-1.32/15 2.40} Kc3 {+10.41/21 6.75}
59. Ke2 {-1.10/13 1.43} Rc2+ {+11.95/22 8.81}
60. Rxc2+ {-0.92/15 0.28} Kxc2 {+10.94/19 1.51}
61. hxg5 {-1.48/17 1.38} fxg5 {+13.07/21 5.96}
62. Be6 {-2.27/18 1.25} Nd1 {+13.38/16 0.34}
63. Bc4 {-3.25/20 0.92} Nc3+ {+16.31/20 2.48}
64. Kf3 {-3.92/23 2.11} Kd2 {+21.85/20 3.35}
65. Ba6 {-4.73/23 1.16} Na4 {+22.03/19 0.58}
66. Bc4 {-5.54/23 0.56} Nb2 {+21.83/16 0.30}
67. Bd5 {-7.52/23 1.01} Nxd3 {+29.21/18 2.02}
68. Ba8 {-9.00/25 1.40} Nb2 {+30.25/18 1.23}
69. Bd5 {-9.00/21 0.13} d3 {+28.38/17 0.60}
70. Bc6 {-12.76/21 1.75} Kc1 {+28.01/11 0.046}
71. Bb5 {-9.00/15 0.22} d2 {+29.44/14 0.068}
72. Be2 {-M58/22 10.02} Nd1 {+29.69/15 0.46}
73. Kg2 {-M28/17 0.58} Kc2 {+29.64/14 0.20}
74. Bf3 {-13.27/17 0.35} Ne3+ {+27.52/12 0.13}
75. Kf2 {-5.82/17 0.24} Nc4 {+30.97/12 0.043}
76. Kg2 {-13.79/22 1.41} d1=Q {+31.62/8 0.014}
77. Bxd1+ {-7.97/16 0.20}
{Black loses on time}
1-0

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "2"]
[White "Seer_200-64"]
[Black "Komodo_3-64"]
[Result "0-1"]
[ECO "B20"]
[GameDuration "00:05:10"]
[GameEndTime "2021-04-27T14:13:49.327 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T14:08:38.997 Mitteleuropõische Sommerzeit"]
[Opening "Sicilian defense"]
[PlyCount "156"]
[Termination "time forfeit"]
[TimeControl "40/80"]

1. e4 {book} c5 {book}
2. d3 {book} Nc6 {book}
3. Be2 {book} g6 {book}
4. Nf3 {+0.09/12 1.29} Nf6 {-0.06/15 4.68}
5. c3 {-0.02/14 4.00} Bg7 {0.00/16 7.38}
6. O-O {+0.07/12 1.25} O-O {+0.05/13 0.34}
7. Re1 {+0.15/13 1.26} d5 {+0.22/12 1.10}
8. e5 {+0.14/13 1.28} Ne8 {+0.06/13 0.64}
9. d4 {+0.20/13 1.29} Nc7 {+0.01/14 3.25}
10. Be3 {+0.19/12 1.30} Bf5 {+0.02/13 1.17}
11. dxc5 {+0.12/14 1.30} Nxe5 {-0.01/14 1.86}
12. Nxe5 {+0.23/14 1.32} Bxe5 {+0.13/13 1.15}
13. Na3 {+0.42/14 1.34} Bg7 {+0.08/14 2.38}
14. Qd2 {+0.35/12 1.34} Qd7 {+0.17/14 2.04}
15. Rad1 {+0.20/13 1.37} Rad8 {+0.17/13 0.43}
16. f4 {+0.34/12 1.38} Qa4 {+0.07/13 1.99}
17. Kh1 {-0.47/14 2.99} Rfe8 {+0.31/14 8.10}
18. h3 {-0.08/16 5.74} f6 {+0.44/14 2.42}
19. Kg1 {+0.14/13 2.01} e5 {+0.61/12 0.79}
20. fxe5 {+0.28/12 1.28} fxe5 {+0.68/14 1.21}
21. Rf1 {-0.13/15 3.87} Rf8 {+0.68/13 0.88}
22. Bf3 {-0.13/14 1.24} Be6 {+0.73/13 1.76}
23. Nc2 {+0.04/13 1.54} a5 {+0.77/13 2.10}
24. a3 {+0.26/11 1.26} Qb3 {+0.63/13 3.01}
25. Rb1 {-0.47/13 1.60} Rd7 {+1.03/14 3.10}
26. Be2 {-1.30/14 3.23} Rxf1+ {+1.30/14 5.15}
27. Bxf1 {-1.89/18 11.29} Qa2 {+1.47/15 1.18}
28. Qc1 {-1.54/16 0.90} d4 {+1.48/15 1.92}
29. cxd4 {-1.53/14 0.93} exd4 {+1.47/15 0.63}
30. Bd2 {-1.43/13 0.93} d3 {+1.39/14 1.37}
31. Ne1 {-1.66/15 2.19} Bf5 {+1.11/14 6.20}
32. b4 {-2.37/16 3.00} Qd5 {+1.32/12 1.69}
33. bxa5 {-1.76/15 9.05} Ne6 {+1.22/11 1.18}
34. a6 {-0.66/12 0.40} Nxc5 {+0.91/11 1.50}
35. a7 {-0.01/13 0.52} Rd8 {+0.47/12 1.18}
36. Nf3 {+0.37/12 0.42} Be4 {+0.49/12 1.81}
37. Be3 {+1.21/12 0.45} Ne6 {+0.30/12 1.95}
38. Qd1 {+1.61/12 0.45} Bf8 {+0.32/10 0.71}
39. Qb3 {+2.10/12 0.48} Bc5 {+0.23/12 0.57}
40. Bxc5 {+1.77/16 3.61} Qxc5+ {+0.24/11 0.23}
41. Kh1 {0.00/16 2.20} Bd5 {+0.22/14 2.13}
42. Qxd3 {0.00/17 1.20} Qxa7 {+0.13/16 4.17}
43. Rb5 {0.00/18 2.31} Ba2 {+0.27/14 2.15}
44. Qe2 {+1.08/15 1.20} Qxa3 {+0.09/13 1.29}
45. Rb6 {0.00/16 1.47} Re8 {0.00/14 1.19}
46. Qd1 {0.00/18 1.22} Re7 {+0.23/14 2.24}
47. Qa1 {+1.46/15 1.29} Qa4 {+0.05/14 0.58}
48. Rb1 {+2.76/19 7.35} Bb3 {+0.05/16 3.02}
49. Qe5 {+3.46/20 11.36} Rf7 {+0.03/14 0.87}
50. Bd3 {+2.87/15 0.96} b6 {-0.11/13 2.69}
51. Be4 {+3.01/18 5.30} Re7 {-0.14/14 1.88}
52. Kh2 {+3.16/16 0.89} Rd7 {-0.16/13 2.12}
53. h4 {+3.08/19 11.62} Qc4 {-0.29/12 0.56}
54. h5 {+2.85/16 2.93} Bd1 {-0.29/12 2.88}
55. Ng5 {+2.66/16 0.64} Qc7 {-0.23/15 1.84}
56. Qxc7 {+2.35/15 0.64} Nxc7 {-0.18/15 1.02}
57. h6 {+2.47/19 8.34} b5 {-0.05/15 0.93}
58. Kg3 {+2.81/15 1.14} Ba4 {-0.16/17 2.49}
59. Rc1 {+2.43/15 0.47} Bb3 {-0.16/15 0.26}
60. Rc6 {+2.80/16 0.89} Re7 {-0.41/15 2.07}
61. Kh4 {+2.76/15 0.47} Bf7 {-0.41/16 1.79}
62. g4 {+3.94/18 4.66} Be8 {-0.51/16 3.91}
63. Rf6 {+4.85/14 0.42} Bf7 {-0.57/17 1.82}
64. Rb6 {+5.00/15 0.85} Be8 {-0.60/18 2.06}
65. Rb7 {+5.21/14 0.43} Kh8 {-0.62/16 0.44}
66. Rb8 {+6.86/18 4.26} Na6 {-0.62/21 2.59}
67. Rc8 {+4.86/16 1.10} Nc7 {-0.62/20 0.93}
68. Bc2 {+6.70/17 2.02} Nd5 {-0.62/14 0.38}
69. Rd8 {+5.94/14 0.14} Nf6 {-0.62/15 1.02}
70. Bb3 {+6.09/14 0.14} b4 {-0.62/16 1.21}
71. Be6 {+7.22/16 1.49} b3 {-2.48/14 2.31}
72. Bxb3 {+7.15/12 0.087} Re1 {-2.41/17 3.89}
73. Be6 {+6.98/13 0.077} Rh1+ {-2.48/17 2.17}
74. Kg3 {+8.11/14 0.35} Rxh6 {-1.69/16 2.82}
75. Bf7 {+9.30/14 0.61} Rh1 {-1.60/16 1.63}
76. Bxe8 {+9.32/10 0.025} Nxe8 {-1.61/15 1.02}
77. Rxe8+ {+8.96/10 0.24} Kg7 {-1.62/14 0.29}
78. Re7+ {+8.08/11 0.066} Kg8 {-1.60/17 2.81}
{White loses on time}
0-1[/pgn]

Code: Select all

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "2"]
[White "Komodo_3-64"]
[Black "Seer_200-64"]
[Result "1-0"]
[ECO "B20"]
[GameDuration "00:05:16"]
[GameEndTime "2021-04-27T14:08:37.937 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T14:03:21.784 Mitteleuropõische Sommerzeit"]
[Opening "Sicilian defense"]
[PlyCount "153"]
[Termination "time forfeit"]
[TimeControl "40/80"]

1. e4 {book} c5 {book}
2. d3 {book} Nc6 {book}
3. Be2 {book} g6 {book}
4. Be3 {+0.07/15 3.19} Nf6 {+0.17/12 1.29}
5. Nd2 {+0.17/15 2.12} e5 {+0.24/13 3.41}
6. Ngf3 {+0.03/13 1.34} d5 {+0.28/13 1.26}
7. c3 {+0.12/15 3.79} d4 {+0.34/13 1.27}
8. Bg5 {+0.03/15 2.58} h6 {+0.33/13 1.29}
9. Bh4 {+0.13/15 2.09} Bd6 {+0.43/11 1.29}
10. Nc4 {+0.19/14 2.26} Bc7 {+0.26/11 1.31}
11. cxd4 {+0.07/13 2.25} cxd4 {+0.43/12 1.32}
12. Rc1 {+0.25/14 2.63} Qe7 {+0.18/12 1.33}
13. O-O {+0.21/13 2.29} a5 {+0.27/13 1.34}
14. Qb3 {+0.42/12 2.30} g5 {+0.77/14 2.28}
15. Bg3 {+0.37/14 2.54} Nh5 {+0.99/12 1.35}
16. Nb6 {+0.39/13 3.83} Bxb6 {+0.96/14 1.36}
17. Qxb6 {+0.38/13 1.06} Nxg3 {+0.86/13 1.39}
18. fxg3 {+0.41/15 1.70} O-O {+0.89/14 2.79}
19. Qc5 {+0.37/16 3.14} Qxc5 {+0.43/15 1.60}
20. Rxc5 {+0.43/15 0.59} f6 {+0.26/19 12.82}
21. Nd2 {+0.37/15 1.41} Be6 {+0.71/19 13.78}
22. a3 {+0.40/15 0.43} Ra6 {+0.74/13 0.78}
23. Bd1 {+0.29/16 1.86} Kf7 {+0.53/12 0.78}
24. Nc4 {+0.45/16 2.03} Ke7 {+0.92/13 0.79}
25. Ba4 {+0.54/15 1.90} Bxc4 {+0.73/13 0.81}
26. Rxc4 {+0.49/16 0.96} Kd6 {+0.90/12 0.83}
27. Rf2 {+0.63/16 1.87} Ne7 {+0.90/14 1.31}
28. g4 {+0.32/15 2.59} Raa8 {+1.06/15 1.29}
29. Rc1 {+0.29/17 1.55} Nc8 {+1.06/14 0.84}
30. Bb5 {+0.26/18 2.76} Nb6 {+1.29/15 0.84}
31. a4 {+0.24/18 1.75} Rac8 {+0.54/16 3.30}
32. Rb1 {+0.17/18 4.26} Ke7 {+1.09/14 0.76}
33. g3 {+0.11/16 1.77} Rc5 {+0.73/13 0.80}
34. h3 {+0.09/17 2.48} Nc8 {+0.59/14 0.82}
35. Bc4 {+0.31/15 0.39} Nd6 {+0.47/16 2.07}
36. Bd5 {+0.20/17 1.64} b5 {+0.39/15 0.78}
37. axb5 {+0.36/15 0.60} Nxb5 {+0.30/13 0.80}
38. Rbf1 {+0.28/16 2.51} Nc7 {+1.01/14 1.16}
39. Bc4 {+0.23/17 1.83} Na8 {+0.92/14 0.84}
40. Ba2 {+0.15/17 1.62} Nb6 {+1.27/13 0.90}
41. Kg2 {+0.04/16 2.75} Rb5 {+1.98/15 1.30}
42. Ra1 {0.00/16 1.83} a4 {+1.95/15 1.88}
43. Rc1 {-0.04/16 1.77} Rc8 {+2.55/15 1.31}
44. Rxc8 {-0.05/18 2.87} Nxc8 {+2.87/15 1.33}
45. Rc2 {-0.05/20 1.92} Nb6 {+2.77/15 1.79}
46. Kf3 {-0.09/18 1.96} Kd6 {+2.33/15 1.34}
47. Ke2 {-0.21/18 2.17} Ra5 {+2.21/17 1.59}
48. Ke1 {-0.18/18 3.05} a3 {+2.67/17 3.03}
49. bxa3 {-0.20/19 2.83} Rxa3 {+2.82/17 6.50}
50. Ke2 {-0.19/17 0.58} Na4 {+3.64/13 1.24}
51. Bc4 {-0.18/17 1.36} Nc3+ {+3.19/15 1.54}
52. Kf3 {-0.18/17 0.56} Nd1 {+3.90/14 1.26}
53. Bb5 {-0.38/17 3.46} Rc3 {+4.55/15 1.27}
54. Rd2 {-0.39/18 3.48} Rc1 {+4.70/14 1.29}
55. Ra2 {-0.40/17 0.80} Kc5 {+8.80/19 16.78}
56. Be8 {-1.01/18 11.71} Kb4 {+9.09/15 2.53}
57. Bf7 {-1.21/16 6.58} Ne3 {+9.00/18 6.13}
58. h4 {-1.32/15 2.40} Kc3 {+10.41/21 6.75}
59. Ke2 {-1.10/13 1.43} Rc2+ {+11.95/22 8.81}
60. Rxc2+ {-0.92/15 0.28} Kxc2 {+10.94/19 1.51}
61. hxg5 {-1.48/17 1.38} fxg5 {+13.07/21 5.96}
62. Be6 {-2.27/18 1.25} Nd1 {+13.38/16 0.34}
63. Bc4 {-3.25/20 0.92} Nc3+ {+16.31/20 2.48}
64. Kf3 {-3.92/23 2.11} Kd2 {+21.85/20 3.35}
65. Ba6 {-4.73/23 1.16} Na4 {+22.03/19 0.58}
66. Bc4 {-5.54/23 0.56} Nb2 {+21.83/16 0.30}
67. Bd5 {-7.52/23 1.01} Nxd3 {+29.21/18 2.02}
68. Ba8 {-9.00/25 1.40} Nb2 {+30.25/18 1.23}
69. Bd5 {-9.00/21 0.13} d3 {+28.38/17 0.60}
70. Bc6 {-12.76/21 1.75} Kc1 {+28.01/11 0.046}
71. Bb5 {-9.00/15 0.22} d2 {+29.44/14 0.068}
72. Be2 {-M58/22 10.02} Nd1 {+29.69/15 0.46}
73. Kg2 {-M28/17 0.58} Kc2 {+29.64/14 0.20}
74. Bf3 {-13.27/17 0.35} Ne3+ {+27.52/12 0.13}
75. Kf2 {-5.82/17 0.24} Nc4 {+30.97/12 0.043}
76. Kg2 {-13.79/22 1.41} d1=Q {+31.62/8 0.014}
77. Bxd1+ {-7.97/16 0.20}
{Black loses on time}
1-0

[Event "My Tournament"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.04.27"]
[Round "2"]
[White "Seer_200-64"]
[Black "Komodo_3-64"]
[Result "0-1"]
[ECO "B20"]
[GameDuration "00:05:10"]
[GameEndTime "2021-04-27T14:13:49.327 Mitteleuropõische Sommerzeit"]
[GameStartTime "2021-04-27T14:08:38.997 Mitteleuropõische Sommerzeit"]
[Opening "Sicilian defense"]
[PlyCount "156"]
[Termination "time forfeit"]
[TimeControl "40/80"]

1. e4 {book} c5 {book}
2. d3 {book} Nc6 {book}
3. Be2 {book} g6 {book}
4. Nf3 {+0.09/12 1.29} Nf6 {-0.06/15 4.68}
5. c3 {-0.02/14 4.00} Bg7 {0.00/16 7.38}
6. O-O {+0.07/12 1.25} O-O {+0.05/13 0.34}
7. Re1 {+0.15/13 1.26} d5 {+0.22/12 1.10}
8. e5 {+0.14/13 1.28} Ne8 {+0.06/13 0.64}
9. d4 {+0.20/13 1.29} Nc7 {+0.01/14 3.25}
10. Be3 {+0.19/12 1.30} Bf5 {+0.02/13 1.17}
11. dxc5 {+0.12/14 1.30} Nxe5 {-0.01/14 1.86}
12. Nxe5 {+0.23/14 1.32} Bxe5 {+0.13/13 1.15}
13. Na3 {+0.42/14 1.34} Bg7 {+0.08/14 2.38}
14. Qd2 {+0.35/12 1.34} Qd7 {+0.17/14 2.04}
15. Rad1 {+0.20/13 1.37} Rad8 {+0.17/13 0.43}
16. f4 {+0.34/12 1.38} Qa4 {+0.07/13 1.99}
17. Kh1 {-0.47/14 2.99} Rfe8 {+0.31/14 8.10}
18. h3 {-0.08/16 5.74} f6 {+0.44/14 2.42}
19. Kg1 {+0.14/13 2.01} e5 {+0.61/12 0.79}
20. fxe5 {+0.28/12 1.28} fxe5 {+0.68/14 1.21}
21. Rf1 {-0.13/15 3.87} Rf8 {+0.68/13 0.88}
22. Bf3 {-0.13/14 1.24} Be6 {+0.73/13 1.76}
23. Nc2 {+0.04/13 1.54} a5 {+0.77/13 2.10}
24. a3 {+0.26/11 1.26} Qb3 {+0.63/13 3.01}
25. Rb1 {-0.47/13 1.60} Rd7 {+1.03/14 3.10}
26. Be2 {-1.30/14 3.23} Rxf1+ {+1.30/14 5.15}
27. Bxf1 {-1.89/18 11.29} Qa2 {+1.47/15 1.18}
28. Qc1 {-1.54/16 0.90} d4 {+1.48/15 1.92}
29. cxd4 {-1.53/14 0.93} exd4 {+1.47/15 0.63}
30. Bd2 {-1.43/13 0.93} d3 {+1.39/14 1.37}
31. Ne1 {-1.66/15 2.19} Bf5 {+1.11/14 6.20}
32. b4 {-2.37/16 3.00} Qd5 {+1.32/12 1.69}
33. bxa5 {-1.76/15 9.05} Ne6 {+1.22/11 1.18}
34. a6 {-0.66/12 0.40} Nxc5 {+0.91/11 1.50}
35. a7 {-0.01/13 0.52} Rd8 {+0.47/12 1.18}
36. Nf3 {+0.37/12 0.42} Be4 {+0.49/12 1.81}
37. Be3 {+1.21/12 0.45} Ne6 {+0.30/12 1.95}
38. Qd1 {+1.61/12 0.45} Bf8 {+0.32/10 0.71}
39. Qb3 {+2.10/12 0.48} Bc5 {+0.23/12 0.57}
40. Bxc5 {+1.77/16 3.61} Qxc5+ {+0.24/11 0.23}
41. Kh1 {0.00/16 2.20} Bd5 {+0.22/14 2.13}
42. Qxd3 {0.00/17 1.20} Qxa7 {+0.13/16 4.17}
43. Rb5 {0.00/18 2.31} Ba2 {+0.27/14 2.15}
44. Qe2 {+1.08/15 1.20} Qxa3 {+0.09/13 1.29}
45. Rb6 {0.00/16 1.47} Re8 {0.00/14 1.19}
46. Qd1 {0.00/18 1.22} Re7 {+0.23/14 2.24}
47. Qa1 {+1.46/15 1.29} Qa4 {+0.05/14 0.58}
48. Rb1 {+2.76/19 7.35} Bb3 {+0.05/16 3.02}
49. Qe5 {+3.46/20 11.36} Rf7 {+0.03/14 0.87}
50. Bd3 {+2.87/15 0.96} b6 {-0.11/13 2.69}
51. Be4 {+3.01/18 5.30} Re7 {-0.14/14 1.88}
52. Kh2 {+3.16/16 0.89} Rd7 {-0.16/13 2.12}
53. h4 {+3.08/19 11.62} Qc4 {-0.29/12 0.56}
54. h5 {+2.85/16 2.93} Bd1 {-0.29/12 2.88}
55. Ng5 {+2.66/16 0.64} Qc7 {-0.23/15 1.84}
56. Qxc7 {+2.35/15 0.64} Nxc7 {-0.18/15 1.02}
57. h6 {+2.47/19 8.34} b5 {-0.05/15 0.93}
58. Kg3 {+2.81/15 1.14} Ba4 {-0.16/17 2.49}
59. Rc1 {+2.43/15 0.47} Bb3 {-0.16/15 0.26}
60. Rc6 {+2.80/16 0.89} Re7 {-0.41/15 2.07}
61. Kh4 {+2.76/15 0.47} Bf7 {-0.41/16 1.79}
62. g4 {+3.94/18 4.66} Be8 {-0.51/16 3.91}
63. Rf6 {+4.85/14 0.42} Bf7 {-0.57/17 1.82}
64. Rb6 {+5.00/15 0.85} Be8 {-0.60/18 2.06}
65. Rb7 {+5.21/14 0.43} Kh8 {-0.62/16 0.44}
66. Rb8 {+6.86/18 4.26} Na6 {-0.62/21 2.59}
67. Rc8 {+4.86/16 1.10} Nc7 {-0.62/20 0.93}
68. Bc2 {+6.70/17 2.02} Nd5 {-0.62/14 0.38}
69. Rd8 {+5.94/14 0.14} Nf6 {-0.62/15 1.02}
70. Bb3 {+6.09/14 0.14} b4 {-0.62/16 1.21}
71. Be6 {+7.22/16 1.49} b3 {-2.48/14 2.31}
72. Bxb3 {+7.15/12 0.087} Re1 {-2.41/17 3.89}
73. Be6 {+6.98/13 0.077} Rh1+ {-2.48/17 2.17}
74. Kg3 {+8.11/14 0.35} Rxh6 {-1.69/16 2.82}
75. Bf7 {+9.30/14 0.61} Rh1 {-1.60/16 1.63}
76. Bxe8 {+9.32/10 0.025} Nxe8 {-1.61/15 1.02}
77. Rxe8+ {+8.96/10 0.24} Kg7 {-1.62/14 0.29}
78. Re7+ {+8.08/11 0.066} Kg8 {-1.60/17 2.81}
{White loses on time}
0-1
Thanks for testing. I've not modified Seer's TM since the last release, though I did try somewhat less aggressive cyclical TM in the past due to time loss concerns. On my hardware, (Seer gets around 800-1000knps) less aggressive cyclical TM resulted in a slight loss of elo. The time loses are likely related to how slow Seer runs on older hardware. I'll try simulating your hardware to reproduce the time losses. What type of NPS were you observing?
Top
connor_mcmonigle
Posts: 532
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Seer 2.0.0

Post by connor_mcmonigle »

Archimedes wrote: ↑Tue Apr 27, 2021 3:47 pm I've had problems, compiling Seer for Android. In simd.h there are the following lines:

Code: Select all

#ifdef __AVX__
constexpr size_t alignment = 32;
#elif __SSE__
constexpr size_t alignment = 16;
#endif
When compiling for ARM architecture constexpr variable will not be initialized and therefore an error occourred.

Changing to

Code: Select all

#ifdef __AVX__
constexpr size_t alignment = 32;
#else
constexpr size_t alignment = 16;
#endif
solves the problem for me (hopefully alignment 16 is correct for Android builds).
Thanks for pointing this out. That was an oversight on my part when I added SSE intrinsics for better performance on older x86 systems yesterday. For ARM, 16 byte alignment should suffice for now as the generic dot product path doesn't make any assumptions about alignment. Seer's speed on ARM will likely be pretty disappointing until I add a NEON path though.
Top
User avatar
Guenther
Posts: 4606
Joined: Wed Oct 01, 2008 6:33 am
Location: Regensburg, Germany
Full name: Guenther Simon

Re: Seer 2.0.0

Post by Guenther »

connor_mcmonigle wrote: ↑Tue Apr 27, 2021 4:56 pm
Guenther wrote: ↑Tue Apr 27, 2021 2:37 pm

Code: Select all

...
70. Bb3 {+6.09/14 0.14} b4 {-0.62/16 1.21}
71. Be6 {+7.22/16 1.49} b3 {-2.48/14 2.31}
72. Bxb3 {+7.15/12 0.087} Re1 {-2.41/17 3.89}
73. Be6 {+6.98/13 0.077} Rh1+ {-2.48/17 2.17}
74. Kg3 {+8.11/14 0.35} Rxh6 {-1.69/16 2.82}
75. Bf7 {+9.30/14 0.61} Rh1 {-1.60/16 1.63}
76. Bxe8 {+9.32/10 0.025} Nxe8 {-1.61/15 1.02}
77. Rxe8+ {+8.96/10 0.24} Kg7 {-1.62/14 0.29}
78. Re7+ {+8.08/11 0.066} Kg8 {-1.60/17 2.81}
{White loses on time}
0-1
Thanks for testing. I've not modified Seer's TM since the last release, though I did try somewhat less aggressive cyclical TM in the past due to time loss concerns. On my hardware, (Seer gets around 800-1000knps) less aggressive cyclical TM resulted in a slight loss of elo. The time loses are likely related to how slow Seer runs on older hardware. I'll try simulating your hardware to reproduce the time losses. What type of NPS were you observing?
Of course this was the (official) core2 build.
Around 150-160 kn/s at start position, but less with low time, it speeds up at later depths.

But I am surprised, why it should influence how much time is used in mps tc?
If people would use extreme fast mps tc wouldn't it have the same issue then, despite fast hardware?

Code: Select all

uci
id name Seer 2.0.0
id author Connor McMonigle
option name Threads type spin default 1 min 1 max 512
option name Hash type spin default 16 min 1 max 65536
option name Weights type string default EMBEDDED
uciok
go infinite
...
info depth 11 seldepth 24 score cp 46 nodes 38142 nps 60542 time 629 pv e2e4 d7d5 e4d5 c7c6 d2d4 c6d5 c2c4 g8f6 c4d5 f6d5 g1f3
info depth 12 seldepth 24 score cp 42 nodes 138018 nps 114158 time 1208 pv e2e4 c7c5 b1c3 b8c6 g1f3 g8f6 f1b5 g7g6 h2h3 d7d6 a2a4 f8g7 a4a5 a7a6 b5c6
info depth 13 seldepth 27 score cp 36 nodes 272650 nps 137632 time 1980 pv e2e4 c7c5 g1f3 b8c6 d2d4 c5d4 c2c3 e7e5 c3d4 e5d4 f1c4 f8b4 b1d2 g8f6 e1g1
info depth 14 seldepth 24 score cp 39 nodes 458203 nps 148670 time 3081 pv e2e4 c7c5 g1f3 d7d6 b1c3 g8f6 d2d4 c5d4 f3d4 b8c6 f1b5 c8d7 c1e3 a7a6 b5a4 e7e6
info depth 15 seldepth 37 score cp 45 nodes 1553033 nps 159040 time 9764 pv e2e4 c7c5 c2c3 d7d5 e4d5 g8f6 d2d4 f6d5 g1f3 c5d4 c3d4 e7e5 f1b5 c8d7 b5d7 d8d7 d4e5
info depth 16 seldepth 29 score cp 42 nodes 2419848 nps 160032 time 15120 pv e2e4 c7c5 g1f3 d7d6 f1b5 b8d7 b1c3 a7a6 b5d7 c8d7 a2a4 e7e6 d2d4 c5d4 f3d4 g8f6 a4a5 f8e7
info depth 17 seldepth 31 score cp 38 nodes 9045935 nps 159405 time 56747 pv e2e4 c7c5 g1f3 d7d6 b1c3 g8f6 f1b5 c8d7 a2a4 b8c6 b5c6 d7c6 d2d4 c6e4 c3e4 f6e4 e1g1 d6d5 d4c5 e7e6
info depth 18 seldepth 37 score cp 31 nodes 11060683 nps 159327 time 69420 pv e2e4 c7c5 g1f3 d7d6 b1c3 g8f6 f1b5 b8c6 e1g1 e7e5 a2a4 a7a6 b5c6 b7c6 a4a5 f8e7 a1a3 h7h5 d2d3 a8b8 h2h3 h5h4 f1e1
May be I try to fiddle with it and do some tests

Code: Select all

    if(get<go::movestogo>().has_value()){
      // handle cyclical time controls (x / y + z)
      const go::movestogo::type moves_to_go = get<go::movestogo>().value();
      min_budget =  2 * (remaining - over_head) / (3 * (5 + moves_to_go)) + inc;
      max_budget = 10 * (remaining - over_head) / (10 + moves_to_go) + inc;
Edit:
BTW that problem was already hidden in that post but overlayed by a mate bug
http://talkchess.com/forum3/viewtopic.p ... 10#p873774

Check time usage in moves 70-80 it was always 0s (GUI used had zero digit precision - that was for CEGT)
https://rwbc-chess.de

trollwatch:
Chessqueen + chessica + AlexChess + Eduard + Sylwy
Top
chrisw
Posts: 4317
Joined: Tue Apr 03, 2012 4:28 pm

Re: Seer 2.0.0

Post by chrisw »

Guenther wrote: ↑Tue Apr 27, 2021 7:01 pm
connor_mcmonigle wrote: ↑Tue Apr 27, 2021 4:56 pm
Guenther wrote: ↑Tue Apr 27, 2021 2:37 pm

Code: Select all

...
70. Bb3 {+6.09/14 0.14} b4 {-0.62/16 1.21}
71. Be6 {+7.22/16 1.49} b3 {-2.48/14 2.31}
72. Bxb3 {+7.15/12 0.087} Re1 {-2.41/17 3.89}
73. Be6 {+6.98/13 0.077} Rh1+ {-2.48/17 2.17}
74. Kg3 {+8.11/14 0.35} Rxh6 {-1.69/16 2.82}
75. Bf7 {+9.30/14 0.61} Rh1 {-1.60/16 1.63}
76. Bxe8 {+9.32/10 0.025} Nxe8 {-1.61/15 1.02}
77. Rxe8+ {+8.96/10 0.24} Kg7 {-1.62/14 0.29}
78. Re7+ {+8.08/11 0.066} Kg8 {-1.60/17 2.81}
{White loses on time}
0-1
Thanks for testing. I've not modified Seer's TM since the last release, though I did try somewhat less aggressive cyclical TM in the past due to time loss concerns. On my hardware, (Seer gets around 800-1000knps) less aggressive cyclical TM resulted in a slight loss of elo. The time loses are likely related to how slow Seer runs on older hardware. I'll try simulating your hardware to reproduce the time losses. What type of NPS were you observing?
Of course this was the (official) core2 build.
Around 150-160 kn/s at start position, but less with low time, it speeds up at later depths.

On an i7 4-core, 4.2 GHz, using seer_core2.exe, e2e4 startpos gave 520-580 Knps for mine. It doesn’t seem to pump out a hashful number, btw.

What’s the difference between the various executables? I can try nps on a threadripper 3990X x64, 256Gb RAM, especially if somebody tells me (command lines for dummies!!)which exe to use and the command line options for 64 threads and some suitable hash.

Connor, very neat idea for training. Are you using a unique net for each material bin, or maintain one net through the iterations? I’m assuming unique net for each and then throw each away?

But I am surprised, why it should influence how much time is used in mps tc?
If people would use extreme fast mps tc wouldn't it have the same issue then, despite fast hardware?

Code: Select all

uci
id name Seer 2.0.0
id author Connor McMonigle
option name Threads type spin default 1 min 1 max 512
option name Hash type spin default 16 min 1 max 65536
option name Weights type string default EMBEDDED
uciok
go infinite
...
info depth 11 seldepth 24 score cp 46 nodes 38142 nps 60542 time 629 pv e2e4 d7d5 e4d5 c7c6 d2d4 c6d5 c2c4 g8f6 c4d5 f6d5 g1f3
info depth 12 seldepth 24 score cp 42 nodes 138018 nps 114158 time 1208 pv e2e4 c7c5 b1c3 b8c6 g1f3 g8f6 f1b5 g7g6 h2h3 d7d6 a2a4 f8g7 a4a5 a7a6 b5c6
info depth 13 seldepth 27 score cp 36 nodes 272650 nps 137632 time 1980 pv e2e4 c7c5 g1f3 b8c6 d2d4 c5d4 c2c3 e7e5 c3d4 e5d4 f1c4 f8b4 b1d2 g8f6 e1g1
info depth 14 seldepth 24 score cp 39 nodes 458203 nps 148670 time 3081 pv e2e4 c7c5 g1f3 d7d6 b1c3 g8f6 d2d4 c5d4 f3d4 b8c6 f1b5 c8d7 c1e3 a7a6 b5a4 e7e6
info depth 15 seldepth 37 score cp 45 nodes 1553033 nps 159040 time 9764 pv e2e4 c7c5 c2c3 d7d5 e4d5 g8f6 d2d4 f6d5 g1f3 c5d4 c3d4 e7e5 f1b5 c8d7 b5d7 d8d7 d4e5
info depth 16 seldepth 29 score cp 42 nodes 2419848 nps 160032 time 15120 pv e2e4 c7c5 g1f3 d7d6 f1b5 b8d7 b1c3 a7a6 b5d7 c8d7 a2a4 e7e6 d2d4 c5d4 f3d4 g8f6 a4a5 f8e7
info depth 17 seldepth 31 score cp 38 nodes 9045935 nps 159405 time 56747 pv e2e4 c7c5 g1f3 d7d6 b1c3 g8f6 f1b5 c8d7 a2a4 b8c6 b5c6 d7c6 d2d4 c6e4 c3e4 f6e4 e1g1 d6d5 d4c5 e7e6
info depth 18 seldepth 37 score cp 31 nodes 11060683 nps 159327 time 69420 pv e2e4 c7c5 g1f3 d7d6 b1c3 g8f6 f1b5 b8c6 e1g1 e7e5 a2a4 a7a6 b5c6 b7c6 a4a5 f8e7 a1a3 h7h5 d2d3 a8b8 h2h3 h5h4 f1e1
May be I try to fiddle with it and do some tests

Code: Select all

    if(get<go::movestogo>().has_value()){
      // handle cyclical time controls (x / y + z)
      const go::movestogo::type moves_to_go = get<go::movestogo>().value();
      min_budget =  2 * (remaining - over_head) / (3 * (5 + moves_to_go)) + inc;
      max_budget = 10 * (remaining - over_head) / (10 + moves_to_go) + inc;
Edit:
BTW that problem was already hidden in that post but overlayed by a mate bug
http://talkchess.com/forum3/viewtopic.p ... 10#p873774

Check time usage in moves 70-80 it was always 0s (GUI used had zero digit precision - that was for CEGT)
Top
connor_mcmonigle
Posts: 532
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Seer 2.0.0

Post by connor_mcmonigle »

Guenther wrote: ↑Tue Apr 27, 2021 7:01 pm
connor_mcmonigle wrote: ↑Tue Apr 27, 2021 4:56 pm
Guenther wrote: ↑Tue Apr 27, 2021 2:37 pm

Code: Select all

...
70. Bb3 {+6.09/14 0.14} b4 {-0.62/16 1.21}
71. Be6 {+7.22/16 1.49} b3 {-2.48/14 2.31}
72. Bxb3 {+7.15/12 0.087} Re1 {-2.41/17 3.89}
73. Be6 {+6.98/13 0.077} Rh1+ {-2.48/17 2.17}
74. Kg3 {+8.11/14 0.35} Rxh6 {-1.69/16 2.82}
75. Bf7 {+9.30/14 0.61} Rh1 {-1.60/16 1.63}
76. Bxe8 {+9.32/10 0.025} Nxe8 {-1.61/15 1.02}
77. Rxe8+ {+8.96/10 0.24} Kg7 {-1.62/14 0.29}
78. Re7+ {+8.08/11 0.066} Kg8 {-1.60/17 2.81}
{White loses on time}
0-1
Thanks for testing. I've not modified Seer's TM since the last release, though I did try somewhat less aggressive cyclical TM in the past due to time loss concerns. On my hardware, (Seer gets around 800-1000knps) less aggressive cyclical TM resulted in a slight loss of elo. The time loses are likely related to how slow Seer runs on older hardware. I'll try simulating your hardware to reproduce the time losses. What type of NPS were you observing?
Of course this was the (official) core2 build.
Around 150-160 kn/s at start position, but less with low time, it speeds up at later depths.

But I am surprised, why it should influence how much time is used in mps tc?
If people would use extreme fast mps tc wouldn't it have the same issue then, despite fast hardware?

Code: Select all

uci
id name Seer 2.0.0
id author Connor McMonigle
option name Threads type spin default 1 min 1 max 512
option name Hash type spin default 16 min 1 max 65536
option name Weights type string default EMBEDDED
uciok
go infinite
...
info depth 11 seldepth 24 score cp 46 nodes 38142 nps 60542 time 629 pv e2e4 d7d5 e4d5 c7c6 d2d4 c6d5 c2c4 g8f6 c4d5 f6d5 g1f3
info depth 12 seldepth 24 score cp 42 nodes 138018 nps 114158 time 1208 pv e2e4 c7c5 b1c3 b8c6 g1f3 g8f6 f1b5 g7g6 h2h3 d7d6 a2a4 f8g7 a4a5 a7a6 b5c6
info depth 13 seldepth 27 score cp 36 nodes 272650 nps 137632 time 1980 pv e2e4 c7c5 g1f3 b8c6 d2d4 c5d4 c2c3 e7e5 c3d4 e5d4 f1c4 f8b4 b1d2 g8f6 e1g1
info depth 14 seldepth 24 score cp 39 nodes 458203 nps 148670 time 3081 pv e2e4 c7c5 g1f3 d7d6 b1c3 g8f6 d2d4 c5d4 f3d4 b8c6 f1b5 c8d7 c1e3 a7a6 b5a4 e7e6
info depth 15 seldepth 37 score cp 45 nodes 1553033 nps 159040 time 9764 pv e2e4 c7c5 c2c3 d7d5 e4d5 g8f6 d2d4 f6d5 g1f3 c5d4 c3d4 e7e5 f1b5 c8d7 b5d7 d8d7 d4e5
info depth 16 seldepth 29 score cp 42 nodes 2419848 nps 160032 time 15120 pv e2e4 c7c5 g1f3 d7d6 f1b5 b8d7 b1c3 a7a6 b5d7 c8d7 a2a4 e7e6 d2d4 c5d4 f3d4 g8f6 a4a5 f8e7
info depth 17 seldepth 31 score cp 38 nodes 9045935 nps 159405 time 56747 pv e2e4 c7c5 g1f3 d7d6 b1c3 g8f6 f1b5 c8d7 a2a4 b8c6 b5c6 d7c6 d2d4 c6e4 c3e4 f6e4 e1g1 d6d5 d4c5 e7e6
info depth 18 seldepth 37 score cp 31 nodes 11060683 nps 159327 time 69420 pv e2e4 c7c5 g1f3 d7d6 b1c3 g8f6 f1b5 b8c6 e1g1 e7e5 a2a4 a7a6 b5c6 b7c6 a4a5 f8e7 a1a3 h7h5 d2d3 a8b8 h2h3 h5h4 f1e1
May be I try to fiddle with it and do some tests

Code: Select all

    if(get<go::movestogo>().has_value()){
      // handle cyclical time controls (x / y + z)
      const go::movestogo::type moves_to_go = get<go::movestogo>().value();
      min_budget =  2 * (remaining - over_head) / (3 * (5 + moves_to_go)) + inc;
      max_budget = 10 * (remaining - over_head) / (10 + moves_to_go) + inc;
Edit:
BTW that problem was already hidden in that post but overlayed by a mate bug
http://talkchess.com/forum3/viewtopic.p ... 10#p873774

Check time usage in moves 70-80 it was always 0s (GUI used had zero digit precision - that was for CEGT)
I'll have to do some more investigation... That is the correct region of code to fiddle with and thanks for looking into it. The CEGT game time usage is weird. I might be setting the max_budget too high which means Seer will think for too long when the evaluation or best move are unstable. You'll note that the max_budget gets clamped later so that only 4/5 of the remaining time can ever be used (I'm not using the over_head margin in the clamp though which might be part of my problem as well). Additionally, min_budget might be too low explaining the 0s in the CEGT game as the evaluation was super stable (just mate score the whole 40 moves as I didn't adjust mate scores based on distance back then :P).

Time losses are usually because Seer can't complete a depth 1 search before it runs out of time... I think. On faster hardware, this is less of a problem. I'm not sure, of course.
Top
connor_mcmonigle
Posts: 532
Joined: Sun Sep 06, 2020 4:40 am
Full name: Connor McMonigle

Re: Seer 2.0.0

Post by connor_mcmonigle »

chrisw wrote: ↑Tue Apr 27, 2021 8:08 pm On an i7 4-core, 4.2 GHz, using seer_core2.exe, e2e4 startpos gave 520-580 Knps for mine. It doesn’t seem to pump out a hashful number, btw.

What’s the difference between the various executables? I can try nps on a threadripper 3990X x64, 256Gb RAM, especially if somebody tells me (command lines for dummies!!)which exe to use and the command line options for 64 threads and some suitable hash.

Connor, very neat idea for training. Are you using a unique net for each material bin, or maintain one net through the iterations? I’m assuming unique net for each and then throw each away?

I don't report hashfull numbers at the moment as I don't understand this information to be very important (if enough people request it, it wouldn't be very difficult to implement though).

Depending on the generation of your core-i7, either the ivybridge or skylake compile should be optimal. Notably, the ivybridge compile uses AVX (256bit wide) intrinsics, but without FMA. The skylake compile has FMA. For a threadripper 3990x, the zen version 2 compile would correspond to the architecture. Seer is a UCI chess engine so from the CLI, you can just issue the following for 16GB hash and 64 threads:

Code: Select all

setoption name Threads value 64
setoption name Hash value 16384
and can enter "uci" to see the list of available options...

I maintain one net through each iteration over piece count as I found this to result in significantly faster training (the N man network needs to evaluate every <N man position correctly anyways). The network starts overfit to only <=6 man positions, then overfit to <=7 man positions and so on up to <=32 man positions. The training code is available here: https://github.com/connormcmonigle/seer ... s/train.py. Afterwards, I do throw the network used to generate labels away as I usually am interested in testing different architectures and tweaking the model, even though the label generating net usually ended up quite strong when I let training continue.
Top
User avatar
Rebel
Posts: 6991
Joined: Thu Aug 18, 2011 12:04 pm

Re: Seer 2.0.0

Post by Rebel »

Division Three

40 moves in 2 minutes

Code: Select all

No. Engine              1     2     3     4     5     6     7     8  Score  Games  Perc 
-----------------------------------------------------------------------------------------
 1 Seer 2.0.0        xxxx  52.0  61.5  62.0  70.5  60.0  64.0  66.5  436.5 / 700 (62.36%)
 2 Weiss 1.3         48.0  xxxx  61.5  68.5  53.5  58.0  68.5  71.5  429.5 / 700 (61.36%)
 3 Topple 0.8.0      38.5  38.5  xxxx  52.5  48.5  52.5  54.5  60.0  345.0 / 700 (49.29%)
 4 Counter 3.7       38.0  31.5  47.5  xxxx  56.0  56.5  47.0  60.0  336.5 / 700 (48.07%)
 5 Seer 1.2.1        29.5  46.5  51.5  44.0  xxxx  51.0  52.5  57.5  332.5 / 700 (47.50%)
 6 Cheng 4.41        40.0  42.0  47.5  43.5  49.0  xxxx  51.5  58.5  332.0 / 700 (47.43%)
 7 FabChess 1.16     36.0  31.5  45.5  53.0  47.5  48.5  xxxx  56.0  318.0 / 700 (45.43%)
 8 Cheng 4.40        33.5  28.5  40.0  40.0  42.5  41.5  44.0  xxxx  270.0 / 700 (38.57%)
Nice progress Connor!
90% of coding is debugging, the other 10% is writing bugs.
Top

Return to “Computer Chess Club: General Topics”