RE: [computer-go] Interesting?/Strange-Game of Many Faces

["Persson, Magnus" <persson@xxxxxxxxxxxxxxxxxx>]

Your cut at move R17 is difficult move to read correctly.
My program (Viking) for example also plays wrong here for a simple
reason.
Since white has two separate groups after your cut, the life and death
reader is called once for both groups. It finds two different moves. R18
that (probably) saves the small group, and S17 that saves the large one.
My program would therefore play S17. The problem here is that R18 saves
both groups and is correct.

I would guess that something similar happens to ManyFaces, but I think
MF always try to find all moves that work and plays the one that is most
profitable. My guess here is that it fails to read out R18 because of a
bug deeper in the tree the finally allows you to kill the stones.

Ignoring your move at S19 at move 201 is the big mistake here, and I can
think of many reasons what the bug might be.

a) MF thinks all its stones are alive because: 1) the upper group has an
eye (but this rule also requires that the semeai has a shared liberty)
or 2) The black group is dead because the lower group wins the semeai,
and thus it falsely believes that the upper group is alive.

b) The white stones are all dead, because the lower group has to play 2
shared liberties which is bad in a semeai (but in this case white has
enough liberties.

So my conclusion is that most go programs has to analyze the life and
death of one group at a time. But in a situation with three groups that
will not work well. In this case the search tree involves some very
difficult move generation issues, so after the cut I would guess many
programs (and humans) would have trouble understanding whats going on.
So sacrificing the small group immediately is excusable.

Hope this makes things clear!
MP

--
Magnus Persson
Center for Adaptive Behavior and Cognition


> -----Original Message-----
> From: chrilly [mailto:chrilly@xxxxxxxxxxxxxxxxx]
> Sent: Mittwoch, 14. Juli 2004 12:02
> To: drd@xxxxxxxxxxxxxxxxx; computer-go
> Subject: [computer-go] Interesting?/Strange-Game of Many Faces
> 
>  Dear All,
> 
> Could a better Go-player and/or David Fotland please comment on the
last
> moves (from black-195 on) of the enclosed game: Chrilly-black, Many
Faces
> white.
> 
> I wanted already to resign and played just for curiosity Black-195. It
> confuses me, that I could easily turn the position and win the game by
a
> wide margin. In my feeling is Many Faces play completly strange. But I
am
> not strong enough to judge this. Maybe it is an interesting position
also
> for other programs.
> 
> Best Regards
> Chrilly
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/