Don Dailey wrote:
[EMAIL PROTECTED] wrote:
For those currently coding this up, I think the most important thing
about this playout algorithm is that it is *temporary*. You will
almost certainly be?replacing it with something different and better
just a little bit down the road. So you probably don't want to worry
about hair-splitting tweaks except as an academic exercise.
Yes, I agree. Also my hair brained scheme of pre-generated tables of
list traversal orderings was just an academic exercise as you say.
But the problem is that when you do heavy playouts you have the same
problem except that the probabilities of the legal moves are no longer
equal. Unless, of course, the board system keeps a list of legal moves,
not just empty points and own eyes in playout mode have zero score which
is the same as if they were illegal moves. Am I the only one doing this?
I don't know if the impact is measurable in strength or not. But as the
game advances, the own eyes are found in the root position. When that
happens, the bias applies to the same move in all the playouts. In this
case the program is favoring systematically the exploration of a move
for no reason at all and it is always the same move. Since the number
of simulations is limited, that pays less attention to other moves.
I don't know if it is bad, but at least it is ugly. ;-)
Jacques.
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
