[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [computer-go] game end protocol (kgsGtp ideas)




From: Don Dailey <drd@xxxxxxxxxxxxxxxxx>
Reply-To: drd@xxxxxxxxxxxxxxxxx, computer-go <computer-go@xxxxxxxxxxxxxxxxx>
To: computer-go@xxxxxxxxxxxxxxxxx
Subject: [computer-go] game end protocol
Date: Sat, 16 Jul 2005 07:40:42 -0400

Mel,


> But I could not debug robot, because I couldn't create situation
> when two robots have played game and not agree with result.
> This situation is appeared once a month at the tournament, and
> that is not sufficient to debug code of the algorithm.


This is simple.   Ask for another computer account and play 2 copies
of your program to debug.

Or just play manually against your program as a guest.  When both your
program and you pass, the protocol should kick in.
But in a non-rated game against a human the full protocol is not used. If there is a disagreement then the human's judgment is given priority.

I wonder if, in order to encourage program authors to fully support the game end protocol, it might be a good idea for WMS to allow use of the full protocol in non-rated games too? Perhaps he could make it an option in kgsGtp?

So there could be a property in the kgsGtp properties file called something like "allow.human.to.score" which if true would mirror the current behaviour (it would probably default to true). If false, it would mean that kgsGtp would use the full game end protocol even in non-rated games vs humans.

On the subject of ideas for kgsGtp, does anyone else wish there was a way to tell kgsGtp to exit at the end of the current game? My program plays various test matches on KGS and it tends to end up will a lot of unfinished games as there is easy no way to stop kgsGtp in between games. Perhaps a simple text interface could be added to kgsGtp which would implement this command? I imagine one could think of other commands to add too.

cheers,
Peter


Or if you need, I can put up a copy of my program that will accept
challenges only from your program. The protocol is debugged and works
fine with my program.

Or put up a copy of gnugo (only for this test) which is set to only accept
challenges from your program.

- Don








> I like the game-end protocol proposed by William M. Shubert.
> If robot don't agrees with opponent about captured stones,
> then game would continue to clear all captured stones.
> People do the same algorithm in the local position
> if they not agree with result.
>
> So, I have implemented this algorithm into bot quickly and simply.
> But it didn't work on the tournament.
> I think about one of three causes:
> 1. KgsGtp hasn't sent the command kgs_genmove_cleanup to robot.
> That command initiate the game-end protocol.
> 2. KgsGtp hasn't the logic module when one robot is using game-end protocol and another not.
> 3. Robot has bugs.
>
> But I could not debug robot, because I couldn't create situation
> when two robots have played game and not agree with result.
> This situation is appeared once a month at the tournament, and
> that is not sufficient to debug code of the algorithm.
>
> May be William will provide the game-end algorithm with instraction:
> how is it implemented and found bugs.





_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/
_________________________________________________________________
Looking for love? Check out XtraMSN Personals http://xtramsn.match.com/match/mt.cfm?pg=channel&tcid=200731

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/