Like most of the UCT programs (I believe), Orego adds one tree node per
Monte Carlo run. At present, this node includes data from the run that
created it. Thus, after the first run, my tree looks like this:
ROOT: 1/1 wins
CHILD A: 1/1 wins
Ignoring the other children, I eventually do another run through that child,
getting this:
ROOT: 1/2 wins
CHILD A: 1/2 wins
GRANDCHILD B: 0/1 wins
My concern here is that that there have been two runs recorded in node A and
only one in node B. Does this cause a problem for the UCT formula, which
assumes that the number of runs through a node is the sum of the number of
runs through the children?
I think the effect would only be a problem if you are using the
memory-saving technique of not expanding a leaf until you have been
through it many times (100, for example).
I'm running a test of this now. Regular UCT vs UCT with a
modification: subtracting 100 from the parent node visit count in the
UCT formula (actually 99 to avoid taking the log of zero). Both
agents suppress node expansion until the leaf reaches 100 visits.
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
