Stockfish: More losts on time (ponder disabled), TC = 3m+2s

[Vinvin20]

From : http://www.talkchess.com/forum/viewtopic.php?p=697040#697040

2016-11-30 version loses more on time.

3 losts on time, 2 times in a winning position and 1 time in a losing position (worth 1.5% score = 10 Elo).
This has no impact on the Lichess analyze (JS or server).

Condition : 3min+2sec
There are some mate scores involved too.
Is it feasible to make SF plays 2% to 5% faster ? Or use less time in the end of the game ?

[Event "CCVA Stockfish zh 30Nov2016 Gauntlet Blitz Tour Nr 43"]
[Site "i7-2600K"]
[Date "2016.12.01"]
[Round "1"]
[White "Stockfish zh 30Nov2016 64bit"]
[Black "Imortal 2.17 64bit"]
[Result "0-1"]
[PlyCount "86"]
[Termination "time forfeit"]
[TimeControl "180+2"]
[Variant "crazyhouse"]

1. e4 {book} e5 {book} 2. Nf3 {book} Nc6 {book} 3. Bc4 {book} Bc5 {book}
4. O-O {book} d6 {book} 5. d3 {book} h6 {book} 6. Nc3 {book} Nf6 {book}
7. Nd5 {book} Bg4 {book} 8. h3 {book} Bh5 {book} 9. c3 {book} a6 {book}
10. d4 {+2.79/19 43s} exd4 {+0.55/11 9.0s} 11. Nxf6+ {+4.97/17 5.0s}
Qxf6 {+0.79/10 8.6s} 12. N@d5 {+4.92/17 6.4s} Qd8 {+0.41/12 8.3s}
13. Nf4 {+4.97/17 3.9s} Bxf3 {+0.17/9 8.0s} 14. Bxf7+ {+6.21/15 4.0s}
Kxf7 {-1.81/10 6.6s} 15. Qxf3 {+6.26/14 0.87s} N@e5 {-0.96/9 7.5s}
16. B@d5+ {+5.92/15 7.8s} Ke8 {-4.53/11 5.1s} 17. Qh5+ {+7.62/14 2.5s}
P@f7 {-4.92/11 7.0s} 18. Bxf7+ {+6.40/16 28s} Nxf7 {-5.08/12 6.8s}
19. P@e6 {+8.71/15 2.3s} N@g5 {-5.71/10 6.6s} 20. exf7+ {+9.37/16 2.3s}
Kd7 {-6.65/10 6.3s} 21. Qg4+ {+8.47/17 1.4s} B@e6 {-12.50/11 5.1s}
22. Nxe6 {+10.06/18 9.0s} Nxe6 {-11.45/9 6.0s} 23. P@d5 {+10.83/18 7.8s}
N@f8 {-12.73/10 5.4s} 24. dxe6+ {+11.66/17 4.0s} Nxe6 {-16.45/10 5.3s}
25. B@d5 {+12.00/18 9.5s} Kc8 {-11.64/10 5.2s} 26. Bxe6+ {+11.85/16 1.2s}
Kb8 {-12.28/12 4.6s} 27. N@d7+ {+11.21/16 14s} Ka7 {-12.99/11 5.0s}
28. a4 {+10.85/16 1.1s} B@d3 {-13.40/9 4.9s} 29. N@b5+ {+14.17/14 4.3s}
axb5 {-14.75/10 4.8s} 30. axb5+ {+13.02/12 1.2s} N@a6 {-15.29/10 4.7s}
31. bxc6 {+13.18/13 11s} P@e2 {-16.19/10 4.6s} 32. Re1 {+17.18/15 9.6s}
bxc6 {-15.46/9 4.6s} 33. N@a5 {+19.10/16 8.2s} P@b7 {-20.44/11 4.5s}
34. Nxb7 {+19.43/13 2.8s} dxc3 {-21.80/10 4.4s} 35. Nxd8 {+21.32/14 8.5s}
Rhxd8 {-20.28/9 4.3s} 36. P@e7 {+21.88/13 3.2s} Rxd7 {-20.70/7 4.3s}
37. Bxd7 {+20.58/13 16s} Bxf2+ {-23.27/7 4.2s} 38. Kxf2 {+22.64/11 1.6s}
P@g3+ {-24.15/7 4.1s} 39. Qxg3 {+25.93/11 3.3s} N@d1+ {-25.81/6 4.1s}
40. Kg1 {+28.84/12 11s} N@c5 {-30.99/6 4.0s} 41. f8=R {+M21/11 5.6s}
P@f2+ {-39.09/5 3.9s} 42. Qxf2 {+M9/14 1.8s} Rxf8 {-M8/7 3.9s}
43. Qxc5+ {+M7/16 2.9s} P@b6 {-M6/8 3.8s, White loses on time} 0-1

[Event "CCVA Stockfish zh 30Nov2016 Gauntlet Blitz Tour Nr 43"]
[Site "i7-2600K"]
[Date "2016.12.01"]
[Round "1"]
[White "Stockfish zh 30Nov2016 64bit"]
[Black "Imortal 2.17 64bit"]
[Result "0-1"]
[PlyCount "94"]
[Termination "time forfeit"]
[TimeControl "180+2"]
[Variant "crazyhouse"]

1. e4 {book} e5 {book} 2. Nc3 {book} Nc6 {book} 3. Nf3 {book} Bc5 {book}
4. Bc4 {+1.20/19 9.6s} d6 {-0.90/13 6.1s} 5. d3 {+1.35/18 4.6s}
Nf6 {-0.76/14 7.2s} 6. O-O {+1.95/18 4.1s} O-O {-1.03/13 6.0s}
7. Bg5 {+0.97/17 1.2s} Bg4 {-2.34/13 8.3s} 8. Nd5 {+2.10/16 3.2s}
Be6 {-2.55/13 8.0s} 9. Nxf6+ {+3.70/16 4.9s} gxf6 {-2.16/13 7.7s}
10. Bxe6 {+3.53/16 3.9s} fxe6 {-2.83/12 7.4s} 11. Bh6 {+3.48/17 16s}
B@b6 {-2.26/11 7.2s} 12. N@g4 {+2.54/16 7.6s} Bxf2+ {-1.87/11 6.9s}
13. Rxf2 {+3.60/16 4.2s} Bxf2+ {-2.64/11 6.7s} 14. Kxf2 {+4.18/18 39s}
Rf7 {-2.06/11 4.8s} 15. B@c4 {+5.74/17 2.2s} Qe7 {-3.18/9 6.3s}
16. B@b3 {+6.30/17 4.0s} N@c5 {-1.59/9 6.1s} 17. B@f5 {+5.45/18 7.3s}
P@d7 {-3.43/12 4.7s} 18. Be3 {+5.22/15 1.1s} Kh8 {-3.84/12 5.7s}
19. Nh4 {+4.60/17 14s} R@h5 {-0.46/13 5.5s} 20. Ng6+ {+4.29/17 11s}
hxg6 {+0.92/13 5.3s} 21. Bxg6 {+4.06/15 1.4s} Rh4 {+0.16/11 5.2s}
22. g3 {+5.23/14 2.5s} Rg7 {-0.67/11 5.0s} 23. gxh4 {+5.43/15 3.0s}
Rxg6 {-1.06/10 4.9s} 24. h5 {+6.39/14 1.7s} Rg7 {-2.86/9 4.8s}
25. R@h6+ {+5.80/15 8.1s} B@h7 {-5.63/9 4.7s} 26. Bxc5 {+6.87/16 6.6s}
N@f4 {-4.68/10 4.6s} 27. Be3 {+9.28/14 2.3s} Rag8 {-7.19/11 4.5s}
28. P@g6 {+8.30/16 7.1s} Nxg6 {-7.39/10 4.4s} 29. hxg6 {+9.50/17 4.6s}
Rxg6 {-7.91/10 4.4s} 30. N@f5 {+9.57/15 4.7s} exf5 {-7.70/10 4.3s}
31. Bxg8 {+10.13/14 18s} Rxh6 {-8.10/9 4.2s} 32. Nxh6 {+10.97/13 2.5s}
f4 {-8.43/9 4.1s} 33. N@f7+ {+11.29/11 3.8s} Kg7 {-8.47/8 4.1s}
34. R@g2+ {+12.88/12 12s} P@g5 {-10.79/8 4.0s} 35. Bxh7 {+15.08/13 5.4s}
fxe3+ {-13.27/7 4.0s} 36. Kxe3 {+17.84/12 1.6s} B@c5+ {-12.86/7 3.9s}
37. Kd2 {+23.91/12 1.6s} Bb4+ {-21.58/8 3.8s} 38. c3 {+26.22/13 2.8s}
Bxc3+ {-22.79/7 3.8s} 39. bxc3 {+28.27/14 2.8s} P@e3+ {-25.10/8 3.7s}
40. Kc1 {+34.62/13 4.4s} P@d2+ {-27.32/8 3.7s} 41. Kb2 {+40.65/13 1.6s}
N@a4+ {-33.90/8 3.6s} 42. Bxa4 {+M17/13 1.9s} R@b6+ {-42.57/8 3.6s}
43. Bb3 {+M13/16 2.6s} Rxb3+ {-34.69/6 3.5s} 44. axb3 {+M11/18 2.6s}
B@c1+ {-M10/8 3.5s} 45. Kb1 {+M9/19 2.7s} Qxf7 {-M8/8 3.4s}
46. N@f5+ {+M7/19 12s} Kxh7 {-M6/9 3.0s} 47. R@h8+ {+M5/21 6.6s}
Kxh8 {-M4/8 3.3s, White loses on time} 0-1

[Event "CCVA Stockfish zh 30Nov2016 Gauntlet Blitz Tour Nr 43"]
[Site "i7-2600K"]
[Date "2016.12.01"]
[Round "1"]
[White "Imortal 2.17 64bit"]
[Black "Stockfish zh 30Nov2016 64bit"]
[Result "1-0"]
[PlyCount "87"]
[Termination "time forfeit"]
[TimeControl "180+2"]
[Variant "crazyhouse"]

1. e4 {book} Nf6 {book} 2. Nc3 {book} Nc6 {book} 3. Nf3 {book} e5 {-0.64/18 13s}
4. Ng5 {+1.00/15 7.7s} Bc5 {-0.24/16 4.4s} 5. Bc4 {+0.29/12 8.7s}
O-O {-0.34/18 19s} 6. d3 {+1.57/13 7.2s} Nd4 {-0.91/18 25s}
7. O-O {+0.91/14 8.1s} Bb4 {-0.57/17 5.4s} 8. Kh1 {+2.21/15 5.5s}
Ne6 {-0.51/16 4.8s} 9. Nd5 {+0.92/13 7.6s} Nxd5 {-1.29/18 7.8s}
10. Nxh7 {+3.18/15 6.5s} Kxh7 {-3.16/19 32s} 11. exd5 {+2.35/14 7.1s}
N@f4 {-3.43/18 18s} 12. dxe6 {+3.71/13 6.9s} dxe6 {-3.46/18 6.2s}
13. Bxf4 {+3.69/12 6.6s} exf4 {-3.58/14 0.59s} 14. Qg4 {+4.32/12 6.4s}
N@h6 {-3.08/15 9.0s} 15. Qxf4 {+5.06/11 6.2s} B@g5 {-2.38/13 2.0s}
16. Qe4+ {+5.86/12 6.0s} Kg8 {-2.47/15 4.3s} 17. P@h7+ {+6.82/12 5.8s}
Kh8 {-1.61/1 0s} 18. P@a6 {+7.59/14 4.6s} P@d5 {-1.20/16 3.4s}
19. Bxd5 {+8.46/13 5.5s} Qxd5 {-0.60/17 1.9s} 20. Qxb4 {+8.46/12 5.3s}
B@d6 {-1.76/15 6.7s} 21. Qa4 {+9.57/12 5.1s} b5 {-1.38/17 18s}
22. Qa5 {+9.49/13 5.0s} Bf6 {-2.42/17 11s} 23. B@g3 {+9.74/12 4.8s}
Bxg3 {-2.12/16 8.6s} 24. fxg3 {+10.20/11 4.7s} B@b6 {-3.64/16 8.7s}
25. Qa3 {+12.08/12 4.6s} Bxa6 {-2.03/15 1.3s} 26. B@e4 {+12.26/10 4.6s}
P@e2 {-3.59/15 4.5s} 27. Bxd5 {+11.12/10 4.5s} exf1=Q+ {-1.83/12 0.44s}
28. Rxf1 {+12.64/10 4.1s} exd5 {-1.88/15 3.7s} 29. N@d7 {+12.64/9 4.3s}
B@d6 {-5.17/15 7.4s} 30. Nxf6 {+17.15/9 4.3s} R@g6 {-7.45/15 6.0s}
31. Qxa6 {+14.23/9 4.2s} gxf6 {-8.03/16 3.1s} 32. N@h4 {+16.80/8 4.1s}
Rg7 {-8.37/13 2.0s} 33. P@b7 {+15.60/8 4.1s} Rae8 {-7.60/13 2.0s}
34. Q@f3 {+16.44/9 3.2s} N@f2+ {-6.85/12 1.7s} 35. Rxf2 {+15.85/10 4.0s}
Bxf2 {-7.47/14 2.3s} 36. Qxf2 {+15.19/8 3.9s} Ng4 {-9.34/14 2.0s}
37. Qfxa7 {+17.18/9 3.8s} R@b6 {-8.86/13 2.0s} 38. Q7xb6 {+15.21/9 3.8s}
cxb6 {-8.41/13 2.0s} 39. B@g1 {+15.95/9 3.7s} Re1 {-8.81/13 2.0s}
40. B@d4 {+15.04/8 3.7s} Q@f2 {-5.66/14 0.78s} 41. R@g8+ {+18.44/7 3.6s}
Rgxg8 {-3.11/12 0.90s} 42. hxg8=Q+ {+13.88/7 3.6s} Kxg8 {-2.66/15 1.4s}
43. P@h7+ {+25.50/9 3.5s} Kxh7 {-M16/17 3.1s}
44. N@g5+ {+25.67/7 3.5s, Black loses on time} 1-0

[ddugovic]

Unlike #134 in this case you're saying SF was winning then lost on time.

Here is SF's time management algorithm. It's possible the best solution requires changing something other than this code.

[ddugovic]

According to Mosca, duplication conditions are:

1. Stockfish zh 30Nov2016 64bit as gauntlet
2. Blitz, TC 3m base + 2s increment/move
3. 32 games per pairing, each opening is played twice
4. Own book, ponder and learning are off
5. Starting positions are from ccva-520start.pgn selected randomly per pairing
6. Hash is 256 MB, threads 1
7. Hardware is i7-2600K, Quad-core, 12 GB RAM, 3.4 GHz, Win 7 OS

[ddugovic]

Possibly related to old issue official-stockfish/Stockfish#603 ?

[ianfab]

I guess so, since they still use move overhead equal to 1000 for TCEC, so this issue probably has not been solved.

[Vinvin20]

I'll ask to use "move overhead=100" for the Crazyhouse Computer Championships 2016 .

[MichaelB7]

What's the time control - you might want to go with 1000 - which is just one second - but should do trick. Anything longer than 5 minutes and it will be worth it.

Sent from my iPhone

On Dec 1, 2016, at 7:24 AM, Vinvin20 [e-mail] wrote:

I'll ask to use "move overhead=100" for this tournament.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.

[ddugovic]

Here are the tournament conditions and I agree with @MichaelB7 , SF can afford a conservative value until someone identifies and explains the root cause.

[Vinvin20]

What's the time control - you might want to go with 1000 - which is just one second - but should do trick. Anything longer than 5 minutes and it will be worth it.

1 second but how many times ?
"move overhead" is used for time lost at each move. If the system estimates it will play 40 moves and it costs 1 second at each move, that's 40 seconds.
In TCEC the TC is more than 3 hours, not a big change if it's 1 minute shorter.
But in a 3 minutes game, 1 minute is too much.

[ianfab]

According to the link above the time control is at least 60'+10'' except for tie-breaks. But even in the case of 3'+2'' it should not be a problem as long as the increment is larger than the move overhead, at least that is what I would deduce from the formula for hypMyTime.

[ddugovic]

Right, using a conservative value has a small risk of SF "thinking" it is in time pressure.

Using an aggressive value could put Stockfish at risk of losing on time, and I don't understand official-stockfish/Stockfish#603 so I'm not qualified to comment on what's considered "aggressive".

[Vinvin20]

According to the link above the time control is at least 60'+10''

OK, I mismatched TC. Let's set to 1000.