Can you both try running your matches with the version I made for the Stockfish testing framework, and then report back with your results? The binary package is hosted at http://koti.mbnet.fi/~ilaripih/bin/cute ... -win32.zip
The version number may say "0.5.1-SF" but it has the same features as 0.6.0.
Thanks!
Same, with the same command line, stalled after 2,600 games without any warning, CPU usage 0%, with hanging engines. EPD 800 positions, random.
Can you both try running your matches with the version I made for the Stockfish testing framework, and then report back with your results? The binary package is hosted at http://koti.mbnet.fi/~ilaripih/bin/cute ... -win32.zip
The version number may say "0.5.1-SF" but it has the same features as 0.6.0.
Thanks!
Same, with the same command line, stalled after 2,600 games without any warning, CPU usage 0%, with hanging engines. EPD 800 positions, random.
Thanks, I managed to reproduce it myself. Hoping to fix it soon...
BTW, this problem only affects the Windows version, so Linux and Mac users need not worry. Windows users should stick to 0.5.1 until an update is released.
Can you both try running your matches with the version I made for the Stockfish testing framework, and then report back with your results? The binary package is hosted at http://koti.mbnet.fi/~ilaripih/bin/cute ... -win32.zip
The version number may say "0.5.1-SF" but it has the same features as 0.6.0.
Thanks!
I tried to run with debug and this is worser, I only managed to finish 200 games.
Last part of debug from first CLI, all 100 games completed.
While console was still opened, I checked the task manager and executed to end the process of one cutechess-cli there. After that my third CLI was run and the match continued, engines were running again.
I think there is a problem when using the engine option 'whitepov' together with -resign option. I have many, many games wrongly adjudicated as a win for the losing side, when the winning side was playing with the black pieces.
I need the 'whitepov' option to get an appropriate eval graph in scid.
Ilari, sorry if this is a dumb question, but there's only one feature that still makes me use Chessbase over cutechess-cli.
Is it possible to create a tournament /gauntlet or 1vs1 match/ that will stack and in case for example of a power failure will continue from where it ends - i.e. from game No.100 out of 10000 for example.
The way Chessbase is doing that is by reserving all the 10000 games like empty slots and then just fills every single one on game end.
If tournament gets interrupted, it continues from the last game that have no recorded result.
Thanks!
Indeed, that is how WinBoard does it too. This 'result string' is stored in the tourney file with the other tourney parameters, and agents playing on the tourney simply pick the first game from it that doesn't have a result and is not marked as 'in progress'. So you can start and stop playing agents at any time. When a tourney was interrupted, you just restart WinBoard on the same tourney file (by double-clicking the latter), and it will resume where it left off. (The game that was interrupted would have to be replayed, though.)
There is one snag, though. If you would really exit by a power failure, the 'in progress' markers are not cleared. (They would be if you would shutdown WinBoard under program control, even when you do it through the window close button). I guess I really should provide a button 'Replay ongoing games' that you could use to remove all 'in progress' markers, to be used only for recovering from such a fatal crash. (Currently the only way to do that is edit the tourney file, to replace the * results by spaces.)
Thanks for releasing this, but I'm seeing a problem that I also saw in the previous version. Basically, when I run a gauntlet with multiple opponents, cutechess-cli seems to lose the connection between the output PGN file and the games. The latest example was today when I kicked off a gauntlet of 34,000 games against five opponents. I killed it after 5300 games because the pgn file only contained the first 4703 games and no new games were being added.
Thanks for releasing this, but I'm seeing a problem that I also saw in the previous version. Basically, when I run a gauntlet with multiple opponents, cutechess-cli seems to lose the connection between the output PGN file and the games. The latest example was today when I kicked off a gauntlet of 34,000 games against five opponents. I killed it after 5300 games because the pgn file only contained the first 4703 games and no new games were being added.
I had that problem before. It went away when I quit using all 4 cores of my computer and started using 3 cores.
Thanks for releasing this, but I'm seeing a problem that I also saw in the previous version. Basically, when I run a gauntlet with multiple opponents, cutechess-cli seems to lose the connection between the output PGN file and the games. The latest example was today when I kicked off a gauntlet of 34,000 games against five opponents. I killed it after 5300 games because the pgn file only contained the first 4703 games and no new games were being added.
Cutechess-cli saves games in the PGN in the same order they were created, so if game #1 stalls or just takes a really long time, and games 2 - 10 finish in the meantime, the PGN file will still be empty because it's waiting for game #1 to finish. When game #1 finally finishes, games 1 to 10 are all written in the file at the same time.
In your test you were running 8 games concurrently, but when you killed the match I can only see 5 "Finished game..." lines. I think there should be 8.