Re: [computer-go] 2nd KGS Computer Go Tournament

[Nick Wedd <nick@xxxxxxxxxxxxx>]

In message <20050509160836.417212FD98@xxxxxxxxxxxxxxxxx>, Paul 
Tabor <paul@xxxxxxxxxxxxxxxxx> writes
> > On 5/9/05, Nick Wedd <nick@xxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > >Sure you can; if there is a dispute regarding a seki (one side 
> > > >claiming the
> > > >other is dead) you play it out. Now the proper playout is: pass; pass.
> > > >in which case, since neither side has made any non-pass moves,
> > > >the position is (correctly) scored as-is. And if one side does make moves,
> > > >then it's their loss:)
(I didn't write that, John Tromp did.)

There are two things that we might be talking about here, and calling 
"disputes".

There are the situations that we saw on Sunday in the KGS tournament, in 
which one bot repeatedly marks dead stones as dead while its opponent 
repeatedly marks them as alive again.  These were not stones about which 
any sane 30k would have any doubt, they were clearly dead.  Now this is 
not a problem.  The Tournament Director can recognise such dead stones, 
count the game, and assign a result (with the consent of the bots' 
owners, if necessary).  It would be nice if bots would admit that their 
clearly-dead stones are dead, but I know how hard it is to persuade 
programmers to act sensibly (see
http://www.britgo.org/gopcres/padvice.html), and I am not going to 
insist on something that really does not matter.

Or, we may be referring to "real" disputes, where the true status of a 
group is difficult (for a 20-kyu) to assess, and after both bots have 
passed, there is a genuine disagreement about the status of a group. 
Now I have never heard of this happening.  Bots tend to defend their 
groups that might be weak, and to attack their opponent's groups that 
might be weak, and one way or another the situation gets resolved.  Even 
with Japanese rules in force I have never heard of a "real" dispute 
happening;  with the Chinese rules that we use for KGS bot tournaments, 
it is far less likely.
    So I am not going to ask programmers to implement a 
dispute-resolution play-it-out protocol (which some will probably get 
wrong) for something that is extremely unlikely to happen anyway.

> I understand that you can apply a rule on life and death that says 
> player a's  group is alive if either player b accepts it is alive or 
> player b can kill it if he moves first once both players have passed. 
> And at least this means stones in seki don't get removed from the board 
> from a player who can argue his case.
> I'm not sure it's all that simple though. Please forgive the notation 
> below, which I've used to get round the proportional spacing you get 
> with some characters like # and *.
> a and b represent the two players. e represents empty points.
>
> Assuming Japanese rules for now, but I'm sure it's possible to set up 
> examples with other rule sets.
< diagram replaced by more legible version: >

 O . # # O . . . .
 . O # # O . . . .
 # # . # O . . . .
 # # # O O . . . .
 O O O O O . . . .
 . . . . . . . . .
 . . . . . . . . .
 . . . . . . . . .
 . . . . . . . . .

> In the local position, there should be no points in this.
Using Japanese rules, there are no points in this.

Using Chinese rules, of the 25 points in the corner, O has 13 and # has 
12 - one of the vacant points belongs entirely to #, two of them are 
shared equally.

> However, using this approach, if I've understooded it correctly, b 
> would get one point within the seki, because the programs wouldn't 
> explicitly understand it as a seki, but instead as 2 live groups - one 
> surrounding no points and the other surrounding one point.
I don't know what strange "understandings" bots may have about this 
position.  In any case, most of them are unable to tell me, and those 
that are able have no incentive to do so honestly.
    Fortunately, I don't care either.  _I_ know how to score this corner 
once they have both passed, and if they disagree about the game result, 
I plan to score it correctly.

Yes, if we wanted bots to play "properly", we should get them to 
understand and implement all the details of the game end procedure. But 
as I doubt 1% of the world's dan-rated players understand these details, 
I don't see it as a sensible option.

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