Dorky 4.2 64-bit 2CPU time losses - very long post

[Graham Banks]

Somehow, around move 60, the engine exceeds time. It almost seems it thinks it has to make only 20 moves in the second time control.

Same for the 32-bit 1CPU version.
I'm going to have to pull it from my next series of amateur tournaments.

[hgm]

This could very well be the classical error from which many WinBoard engines suffer: not correctly counting moves when in force mode. This only manifests itself when you play the engine with an external book.

This case seems a bit unusual, though: you feed it 10 opening moves, and it forfeits on move 60. Usually they forfeit on move 50, in that case, because they expect new time then (40 moves after the first move they started thinking on), using up all time they had gotten at move 40, and then they don't get the new time, because they get it only on move 80.

[Graham Banks]

This could very well be the classical error from which many WinBoard engines suffer: not correctly counting moves when in force mode. This only manifests itself when you play the engine with an external book.

I thought the same so I checked it without book and the same happened. It is definitely not an error of miscounting moves. Moreover, I can't remember seeing this behaviour with versions 4.0 and 4.1.

I switched to the 4.1 exe and got the same problem. 4.0 was okay though.

[Sven]

I have taken a look at the time utilization of Dorky in two of these three games based on the given PGN. The engine did clearly respect the first time control at move 40 and used the expected average time per move for that purpose (~4,5 .. 5 sec/move after leaving the book). Immediately after move 40 time consumption increased to some 8 .. 9 sec/move which is almost a doubling, indicating that

1) the problem is obviously not related to wrong move counting w.r.t. opening book,
2) the engine somehow thought that it can use the extra time it got after move 40 for the next 20 (not 40) moves, i.e. assuming next time control were at move 60.

In the first game I looked at (round 43 with black vs. AnMon) Dorky had about 6 seconds left after its move 60...e1=Q, and in the second game (round 129 with white vs. Counter) it had about 7 seconds left after its last move 58.Kd5.

Sven

[Graham Banks]

I have taken a look at the time utilization of Dorky in two of these three games based on the given PGN. The engine did clearly respect the first time control at move 40 and used the expected average time per move for that purpose (~4,5 .. 5 sec/move after leaving the book). Immediately after move 40 time consumption increased to some 8 .. 9 sec/move which is almost a doubling, indicating that

1) the problem is obviously not related to wrong move counting w.r.t. opening book,
2) the engine somehow thought that it can use the extra time it got after move 40 for the next 20 (not 40) moves, i.e. assuming next time control were at move 60.

In the first game I looked at (round 43 with black vs. AnMon) Dorky had about 6 seconds left after its move 60...e1=Q, and in the second game (round 129 with white vs. Counter) it had about 7 seconds left after its last move 58.Kd5.

Sven

Hopefully Matt reads all this.

[mmcknight]

Thanks for bringing this to my attention. I will work on the fix.

[Graham Banks]

Thanks for bringing this to my attention. I will work on the fix.

Thanks Matt.

[mmcknight]

Would one of you possibly be able to email me a debug file? I am not seeing a second "level" command being issued in the forum post.

Thanks,
Matt

[Graham Banks]

Would one of you possibly be able to email me a debug file? I am not seeing a second "level" command being issued in the forum post.

Thanks,
Matt

On their way to you now.

[hgm]

Would one of you possibly be able to email me a debug file? I am not seeing a second "level" command being issued in the forum post.

What do you mean, second level command? You will get only one per game, unless you are in a multi-session TC game, (which only ChessGUI currently implements, I think), and the next session has parameters different from what was implied by the original level command.