[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: computer-go: a new fractal algorithm
> > the big problem seems to be how to reduce the board size without losing
> > importnt information. If a reduced group becomes dead (because there isn't
> > enough space to make it alive), then the reduction fails...
This is a good point. The reduced shapes don't follow the same rules as
mere stones. Perhaps the reduction might work somewhat, if there were also
special stones to represent an eye and so on.
By the way: try to imagine a board with 31 distinct living groups
reduced to 5x5 board!!! What use is there from having solved 5x5 go?
But of course the weirdest ideas are usually the best way to good ideas...
:)
What I am doing now is solving a subgame:
Small (4x4 for now) area is surrounded by living white stones. Black is
trying to form a single eye in the area and white is trying to prevent it.
All (43M) positions are evaluated as win for black or white or the one in
turn. Also some wins for white are win for black if he gets 2 turns. (I
left details like ko out here)
Next I will take a patch from real game and use the database to classify
eyes or half eyes and so on. This might be useful for life and death
evaluation. (which sucks in my program ;) )
Has anyone used that approach?
It might be somewhat linked to fractal go, if reduced stones are thought
of as shapes.
--
Tapani Raiko, <tapani.raiko@xxxxxxxxxxxxxxxxx>, 050 5225 750,
http://www.hut.fi/u/praiko