Re: [computer-go] Third KGS tournament: game-end protocol

[David G Doshay <ddoshay@xxxxxxx>]

On 10, Jun 2005, at 7:49 AM, Michal Bazynski wrote:

> where does the protocol say you have to fill down to 2 libs??
> the protocol assumes that either bots can score properly (games end 
> after 2 passes) or cannot (so there's no point in asking the bot a 
> second/third/nth time about status, we already lost all faith in the 
> bot when first disagreement happened...). now, since it is assumed 
> that one of the bots cannot score, the bots are asked to remove 
> everything they wish to be considered dead - because there's no other 
> way to determine it (once we believe one of the bots cannot score 
> properly). that does _not_ mean fill all libs...
No, the protocol does not explicitly state that all liberties down to 
the last two eyes MUST be filled, but I claim that this is the result 
we will get. If any stone or string of stones left on the board is 
alive and we play under Chinese rules, then plopping a stone on any 
intersection results in getting a point even if there is absolutely no 
chance of that stone or group growing into a two eye group. My claim is 
that that this behavior is strongly encouraged by the protocol because 
if the opportunity for getting points is there then programs will do 
it.

I am impressed when a program recognizes that there is nothing more to 
be gained and passes. The other program should not be allowed to pass 
and thus force its stones to be considered alive.

I think you state the real problem very well when you say
> we already lost all faith in the bot when first disagreement 
> happened...). now, since it is assumed that one of the bots cannot 
> score ...
If one of the bots cannot score properly then let the programs know 
which groups are subject to question. A bot that does not specify the 
group it claims is alive and wrongly accused by the other program of 
being dead should have no right to claim that the other bot is wrong. 
My point is that we will eliminate a large majority of these 
disagreements if the lists of disputed stones are sent to the other 
program when we say "you may resolve this dispute by playing it out." 
More mature programs will likely agree to kill a specific dead group. I 
see a big loophole in my request as well: the bot can just send the 
whole board over as disputed: All my color are alive, all your color 
are dead. But if we are trying to get these programs to actually play 
better rather than just win tournaments by finding loopholes in the 
rules, this behavior won't last long.

But there will still be unresolved disputes and a TD will have to 
decide. Go is just that kind of a game.

And I do think that the subject of resolution of disputes and clock 
time needs to be made clear.

Cheers,
David

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