________________________________
From: Mark Boon <[email protected]>
>The option that I'd have researched is:
> - spot semeai
> - solve them with a dedicated solver
> - if a move needs to be played to decide the semeai, favour that move
> in playouts (i.e. play first with a certain chance).
> - if a move is played in the semeai that requires a reply to keep
> winning, favour that reply in playouts.
> I don't say it would work any better. Just that it would have been
> first on my list to try.
To really "solve" a semeai (or any tsumego problem ) would be to spot
relationships such as "w wins with correct play,
where correct play is defined as a particular set of move-response pairs."
Those move-response pairs could then, as Mark Boon suggests, be favored in
playouts. They'd be re-calculated when a move is played which affects the
semeai. A thorough solution would be a sort of tree - when move X happens,
there
are responses a, b, c; after that, the solution is reduced to a simpler set of
move-response pairs.
In most cases, such responses would be heavily favored.
The solution should also note when a ko is involved.
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
