It's quite different from PN. PN expands a leaf node one ply and backs up values to the root. I play a line as many ply as needed until I get a high confidence evaluation of win or lose. In this sense I am doing something like UCT with nonrandom play outs. PN typically doesn't use move ordering information. I have a good move generator that sorts candidate moves by probability of winning, so I always pick the highest probability unevaluated child to expand.
Say I want to do a life and death search on a group to kill it. The move generator suggests 5 moves. Make the best one, and evaluate. If the status is still unclear, I call the generator to make move to live. I pick the best and make it, etc. Say I go 8 ply deep and find that the group dies. Now I have one line of play that works. To pick the next node to expand I want to try a node that can change to root from win to loss. So I only have to look at opponent moves. I pick the move that has the highest probability of changing the root to a loss (which can be at any depth), and play out another line until I have a stable result. This models the way I read life and death problems. David From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Jason House Sent: Thursday, December 13, 2007 5:50 AM To: computer-go Subject: Re: [computer-go] Non-global UCT On Dec 12, 2007 10:19 PM, David Fotland <[EMAIL PROTECTED]> wrote: Many Faces' life and death search is best first and probability based, but I don't use UCT to select moves. I select the move that has the highest probability of changing the value of the root (from success to fail or vice versa). I don't use MC to evaluate the endpoints. I look forward down one line until the result is clear (alive or dead), but I follow the best move from the move generator, not random moves. Out of curiosity, how does that compare to proof number? Proof number uses a heuristic like "for this outcome to occur, N child nodes must have outcome X". It picks the path with the smallest N.
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
