[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [computer-go] Competition of the GO-algorithms
Alexander wrote:
> So, I organized the competition of the GO-algorithms to calculate the result
> of the game played by Japanese Rules.
How finished are the games? In the example games I've looked at dame
is not filled, so I suppose that's generally the case. Can there also
be forced reinforcement moves (teire) omitted? Can there be an
unfinished ko left on the board?
> The algorithm has to contain almost all parts of playing program to solve this task.
I'm not so sure about that. I think e.g. that you can get a decent
score without a proper life and death analysis, using various
heuristics which are only valid when the game is known to be finished.
But it depends on how finished the games are of course. With an open
ko it's quite a challenge to determine who has the most ko threats and
play through them without losing points due to bad ko threats or
incorrect answers to ko threats.
> Rules of the competition is placed here:
> http://rusgolib.iponweb.net/AleksandrMel'nikov.html
> You can see statistics and preliminary results of the competition also.
Which version of GNU Go has been tested? Both GNU Go 3.6 and GNU Go
3.7.1 correctly score the example game 2527 on the web page.
> If you would not check your algorithm by the competition Database,
> you can check it by your own Database and publish the result to compare with other algorithms.
Is the competition database available or is it planned to be kept
secret? I did find the 50 example games in the Comp_CPP.zip file. If
anybody wants to have a GTP test file for those, send me a mail.
/Gunnar
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/