>If I use AMAF and the Pebbles rule (play the move with the most wins), >I'll only need to store the number of wins for each node, right?
No, also trials. You need to explore moves based upon UCT (or whatever home-brew solution works better for you). The "Pebbles rule" (I should have published something better than this first, so that 'Pebbles Rule' could refer to something really cool) concerns solely the question of which move to select when your search is over. The conventional choice is to play the move that has the most *trials*, and Pebbles version is to play the move that has the most *wins*. You can verify that Pebbles rule is a no-risk proposition, because if the search continued then the move that Pebbles selects (most wins) would eventually also have the most trials. But there should be very little difference between them, and from experimentation on CGOS I can confirm that there is very little difference. >This wouldn't work without AMAF, because a move might never recover >from a few bad runs. With AMAF, it could. Well, maybe. But that would be the Orego rule. :-) >The question is, what about passing? In Pebbles, passing is initialized with a few dozen losing AMAF trials. This is enough to prevent it from being considered as best until alternatives look really bleak. Seems to work OK. Pebbles still has a lot of bugs in seki handling, but it seems that one isn't a problem. I am finding that seki handling is among the most difficult domain- specific issues. You will need a lot of patterns and logic to handle seki well. Pebbles is nowhere near accurate. Best, Brian _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
