RE: [computer-go] Fourth KGS tournament: results

["Peter McKenzie" <peter_mckenzie@xxxxxxxxxxx>]

> From: Nick Wedd <nick@xxxxxxxxxxxxxxxxx>
> Reply-To: computer-go <computer-go@xxxxxxxxxxxxxxxxx>
> To: computer-go <computer-go@xxxxxxxxxxxxxxxxx>
> Subject: [computer-go] Fourth KGS tournament: results
> Date: Tue, 12 Jul 2005 14:34:54 +0100
>
> The results of the 4th KGS Computer Go tournament are now at
> http://www.weddslist.com/kgs/past/4/index.html
>
> Congratulations to the winners, Aya and tlsBot!
>
> I have not "closed" (in KGS's sense) the Open division, because there is a 
> problem with it.  I would like to award gonzoBot a win in its final-round 
> game against Go81 (it killed all of Go81's stones), but the system has 
> recorded this as a win for Go81, and I don't have the power to change this 
> result myself.  There is more about the problem in the above-cited report.  
> I expect this particular problem won't recur, it was caused by a bug in 
> Gonzo which will be fixed,
I will describe this problem as there is a chance that others may fall into 
it.

Firstly, I will say that in the weeks prior to the tournament Gonzo played 
many hundreds of test games on KGS without problem.  Gonzo was able to 
finish all those games without intervention.  However in the tournament all 
of Gonzo's games required manual intervention to be completed.

I believe the problem is that Gonzo did not implement the final_status_list 
command.  I was under the impression that this command was only required for 
rated games and that if you didn't implement it kgsGtp would interpret that 
as being equivalent to implementing it and always returning an empty list of 
dead stones.  While this is true for non-tournament games it doesn't seem to 
hold true for tournament games!  I'm not sure if this is a bug in kgsGtp, or 
if it is intended behaviour.  Note that the kgsGtp documentation says that 
final_status_list is only required for rated games.

The solution (hopefully) is simply to implement the final_status_list 
command, even if the implementation is as simple as always returning an 
empty list of dead stones - which is exactly what I have done now in Gonzo.  
Unfortunately it may be difficult to test whether this actually solves the 
problem until the next tournament.  I wonder if I'm able to create a 
tournament myself to test this?

Anyway, the tournament was very enjoyable.  I would encourage others who may 
have a program that is in the early stages of development to make the effort 
to enter.

And finally, apologies for turning up a couple of minutes late and then 
having the nerve to be grumpy about missing out on the first round :-)  I'm 
looking forward to the next tournament, the start time is better time for me 
too (8am).

cheers,
Peter

> but I still find it odd than I can overrule a game result in all but the 
> final round.
>
> I am unsure whether it was right of me to enter a version of GNU Go myself. 
>  Reasons in favour:
>   The more entrants, the better
>   People like to see their bots compete against GNU Go
>   It is no extra work for me, my PC is connected anyway
> Reasons against:
>   The GNU Go was eight months old, a recent one would play better
>   As organiser, I ought not to help a program like this
> I will welcome views on this, particularly from GNU Go developers.
>
> Nick
> --
> Nick Wedd    nick@xxxxxxxxxxxxxxxxx
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://www.computer-go.org/mailman/listinfo/computer-go/
_________________________________________________________________
Become a fitness fanatic @  http://xtramsn.co.nz/health

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