Crafty 23.2 JA x64: Little bug ...

[bob]

Hi Bob,

now as *.zip

http://www.amateurschach.de/download/x6 ... roblem.zip

Best
Frank

OK, got it. Delete the stuff. Am looking at it now.

[bob]

OK, some questions. You are stuffing the first N moves, probably by starting from some pre-selected opening? If so, where does the time control start? at move 1, or at the first search in the game? I think we saw an issue like this a while back and fixed it. Will check to see if the fix was somehow lost...

[Frank Quisinsky]

Hello Bob,

file delete.

1. I used the Shredder 12 opening book. Means Shredder Classic 4 GUI used the Shredder 12 opening book, GUI manage the book moves for booth engines. The reason I don't used book learning or the owns opening books which comes by engines. This books I delete in the engine directorys. You can see that book learning / own opening book = off in Crafty.rc.

2. Time control ...
40 moves in 10 minutes, 40 moves in 10 minutes ... repeadedly
started after an engine is out of Shredder's own opening book.

Best
Frank

[bob]

Never mind the questions. I did find a bug in the time usage for that kind of repeating time control. We never use that so it has not been tested since the last WCCC event we played in. The last move could set the target time to the total remaining time, which is dangerous. A simple fix has corrected this. I can update 23.2 if you want, or you can wait for 23.3. I could even do a 23.2a which would be more reasonable to keep the old and new separate... I'll put this on the ftp box in just a few minutes.

OK, there is a 23.2a version there, let me know how this does, but the losing on time on the last move of that repeating time control should be history.

[Frank Quisinsky]

Hi Bob,

for myself ...
I can wait of Crafty 23.3 version.
The games Crafty 23.2 lost will be replay for my ratinglist, clear own rules in such cases.

I have look yesterday in all of my Crafty "Lost of time" games.

1x in move 37
2x in move 38
13x in move 39
4x in move 40

and ...

1x in move 159
1x in move 158

In both games 4TBs Nalimov working. But I don't have the log files. Normally log file = off in my Crafty.rc.

No "lost on time" games in moves 78-80, 118-120 ... (each 40 moves)
Perhaps random, perhaps this could interesting for yourself.

Interesting is, this two games I have played without the parameter
"mode=tournament".

My first information was wrong. I wrote that I have no problem without the parameter "mode= tounrament".

But the other "lost on time" games before move 40 (first time control) are all with the parameter "mode= tournament".

Have now 140 games played with the parameter "mode= tournament" and 100 games without the parameter "mode= tournament".

With other words:

With parameter "mode= tournament" (140 games) = 20x lost on time before the first time control.

Without parameter "mode= tournament" (100 games) = 2x lost on time before the forth time control in endgame with 4-pieces tablebases Nalimov.

Thanks!

My ratinglist was playing on 4 different Intel Q9550 systems. The lost on time games are on different of these systems. The two lost on time games too ... means no problem with my harddisk or defect tablebases databases (checked this in the night).

Best
Frank

[bob]

Hi Bob,

for myself ...
I can wait of Crafty 23.3 version.
The games Crafty 23.2 lost will be replay for my ratinglist, clear own rules in such cases.

I have look yesterday in all of my Crafty "Lost of time" games.

1x in move 37
2x in move 38
13x in move 39
4x in move 40

and ...

1x in move 159
1x in move 158

In both games 4TBs Nalimov working. But I don't have the log files. Normally log file = off in my Crafty.rc.

No "lost on time" games in moves 78-80, 118-120 ... (each 40 moves)
Perhaps random, perhaps this could interesting for yourself.

Interesting is, this two games I have played without the parameter
"mode=tournament".

My first information was wrong. I wrote that I have no problem without the parameter "mode= tounrament".

But the other "lost on time" games before move 40 (first time control) are all with the parameter "mode= tournament".

Have now 140 games played with the parameter "mode= tournament" and 100 games without the parameter "mode= tournament".

With other words:

With parameter "mode= tournament" (140 games) = 20x lost on time before the first time control.

Without parameter "mode= tournament" (100 games) = 2x lost on time before the forth time control in endgame with 4-pieces tablebases Nalimov.

Thanks!

My ratinglist was playing on 4 different Intel Q9550 systems. The lost on time games are on different of these systems. The two lost on time games too ... means no problem with my harddisk or defect tablebases databases (checked this in the night).

Best
Frank

The problem is that in this code, where there is a "drop dead" number of moves and time limit, the program attempts to use _all_ of the time. Not more than what is left, but _exactly_ what is left. And near a time control, since the "absolute time limit" is set to the remaining time, it can burn that up and it then has to notify the GUI about the move, which will cause it to step over the time control and lose.

At one point, we had a small fudge factor set to prevent using ALL the time, but somewhere that was changed to "0" in data.c. As a result, rather than changing it back, we just make sure that we can't use more than 1/2 the remaining time. Testing showed that worked just fine. The "mode tournament" option disables the "run 'im out of time" detection code used when playing only against humans, so with no "mode tournament" and no "computer" that code would come into play and cause it to play the last couple of moves in the time control much faster and avoid the problem (in the wrong way)...

This is fixed in 23.2a and 23.3 (which will be released right after ACCA event). 23.3 is (so far) +55 elo over 23.2. Not huge, but also not insignificant, either.

[bob]

Hi Bob,

for myself ...
I can wait of Crafty 23.3 version.
The games Crafty 23.2 lost will be replay for my ratinglist, clear own rules in such cases.

I have look yesterday in all of my Crafty "Lost of time" games.

1x in move 37
2x in move 38
13x in move 39
4x in move 40

and ...

1x in move 159
1x in move 158

In both games 4TBs Nalimov working. But I don't have the log files. Normally log file = off in my Crafty.rc.

No "lost on time" games in moves 78-80, 118-120 ... (each 40 moves)
Perhaps random, perhaps this could interesting for yourself.

Interesting is, this two games I have played without the parameter
"mode=tournament".

My first information was wrong. I wrote that I have no problem without the parameter "mode= tounrament".

But the other "lost on time" games before move 40 (first time control) are all with the parameter "mode= tournament".

Have now 140 games played with the parameter "mode= tournament" and 100 games without the parameter "mode= tournament".

With other words:

With parameter "mode= tournament" (140 games) = 20x lost on time before the first time control.

Without parameter "mode= tournament" (100 games) = 2x lost on time before the forth time control in endgame with 4-pieces tablebases Nalimov.

Thanks!

My ratinglist was playing on 4 different Intel Q9550 systems. The lost on time games are on different of these systems. The two lost on time games too ... means no problem with my harddisk or defect tablebases databases (checked this in the night).

Best
Frank

There is not a real pattern related to this problem, other than it happening right around a time-control boundary. Once I had found that, the problem became apparent as well as a workable solution...

thanks for spotting this, it is significant for any that test using this type of time control (repeating, no inc).

[Frank Quisinsky]

Hi,

thats what I like on your messages.
A lot of interesting / readable information ... all the years ... for each reader (programmer or not).

55 ELO is a lot ...
Crafty need an higher ELO jumping because Crafty is cult.

The most problem Crafty has all the years are the tactical weaknesses. Very fine positional moves but if Crafty is playing against a tactical power engine your engine have some problems. In king safty too.

I think if you can solved that Crafty is a TOP-10 Engines (also with 1 Core, ponder = on ... compare to the other TOP engines).

I like Crafty, all the nice parameters ... an engine I can spend a lot of time with different experiments to computer chess.

Perfect work with WB so many years ...

Good luck with ACCA event.

Best
Frank