Quoting Matt Gokey <[EMAIL PROTECTED]>:
similar multi-level or hierarchical/partitioned search
concepts have been suggested by several people here over the years,
myself included many times.
Certainly it's the obvious approach to try. And then, as you say,
"how" becomes the difficutly!
... but haven't yet figured out how to make it work in practice. I
tried some things years before MC/UCT popped up without any real
success.
I don't understand the follow-up to your post claiming that you can't
do this for these kinds of games because they are not forcing move
sequences. We're talking about the play-out part of the search used to
sample the game tree. Anything goes, right? Of course, whether any
particular play-out method helps or not is another question.
That was a response to someone else's response, if I understand you.
I'm thinking along the lines of weighted search. Move order matters in
a game, but not always, and with the first random playouts it's very
likely that whatever order we've tried was either unimportant or less
than optimal.
So we'd keep track of which stones, of which color, were on the board
each time we'd tried a particular move, and whether this had been for
White or for Black, and whether the outcome was favorable or
otherwise. As we went along we'd develop a set of features associated
with successful outcomes plus a different set of features (probably
overlapping!) associated with unsuccessful outcomes, and it seems
potentially useful to weight the likelihood of trying a move with the
square of the number of positive features, minus the square of the
negative features. I'd expect this to lead to a tendency to both use
moves from successful local sequences and play them in order, without
explicitly calling for that.
Mainly I'm thinking of ways a program could generalize from this kind
of information. The value of one specific point in relation to the
stones that had been in place in a different game would almost
certainly change, but perhaps a different point would have a similar
relationship to stones in the new position.
Every time I actually start coding, some new idea or some new
objection comes along...
Forrest Curo
----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
