Don Dailey wrote:
For instance, in a set of master games what feedback do I have about
each move other than that it was chosen? How do I get the opinion of
the master player concerning the moves played and not played?
Just an example: Search corner joseki sequences played often enough.
Once you know where the player can play some alternative move (because
you only learn "popular" sequences found in many games) or play tenuki
(in this context, tenuki is a move outside the corner we are studying)
_then_ use UCT simulations (measuring territory differences, not % of
wins) to determine the temperature of each move by comparing the
simulation with the move and the simulation with a pass instead.
(Of course, not for passing, but for playing elsewhere). Your joseki
database will have: the moves, their alternatives and the price you play
for abandoning the sequence at a given point.
Good style: In the case of good style, its is not really important
to understand why good style is good. You will only study the move
before the others, you won't necessarily play it. With limited time
resources (i.e. always) studying good style moves before should give
the program better style, but as the result of search (i.e. online
knowledge). Again we integrate "human style" with what a computer
understands.
Far better, if you want to involve human players, is some kind of human
assisted learning where games are played and learning takes place by
trial and error and direct interaction with the "teacher." But this
isn't very practical for machine learning which likes thousands of
examples to work from.
Here and in the rest of your post, I agree. We have to learn from
strong players but we do not talk the same language ( unless they write
programs, of course ;-) ). Their brain directs them to the good moves
only. They filter bad moves so automatically, they don't realize that
there are 250 stupid legal moves or more on a board and what consequences
this has. Handcrafted databases are not a good idea for statistical
analysis, but of course they have other good uses. E.g. not filling
your own eyes is easily implemented as a handcrafted database.
Jacques.
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
