No this is not doable. Especially not in the endgame-subgame sense,
because there are no independent sub-games in 99% of a normal gogame.
This is the reason non-MCTS old school programs got stuck. Search must
be global. I am not saying that local search cannot be helpful. It
just has to be integrated efficiently into global search.
Sorry for sounding so negative. But reinventing old square shaped
wheels will not bring us forward.
Magnus
Quoting Adriaan van Kessel <[email protected]>:
[email protected] wrote on 11-06-2009 23:00:56:
I think one approach to scaling playouts is to subdivide the board,
stipulate the outcome in a fixed part, and do "norman" playout in
the interesting part. The trick is to do a reasonable subdivision
in the first place. Imagine an advanced UCT model that conceptually
behaves like human evaluation, where you try to divine the outcome
of some local fight (ignoring everything else) as a first pass.
Exactly.
This is more or less the Berlenkamp-approach: divide the board
into subgames, and estimate the "temperature" for each of them.
After that, you can select a plan on how (in which order) you want
to handle the subgames. Which is mostly about ko and sente, but that
is the nature of the game.
IIRC Berlenkamp's approach only applied to endgames, where the whole board
is already occupied, and no more territory can be gained. Extending it
towards
the beginning of the game with "take moyo" or "create new groups" goals
_could_
be doable. (what is the "temperature" of an isolated stone ?)
AvK
Disclaimer RIVM
--
Magnus Persson
Berlin, Germany
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
