I have been playing with pure MC players, thanks to Lukas Lew's library. I noticed that they tend to play silly moves, like putting the opponent into atari, even if the move is a self-atari as well. Any reasonable player can see that it won't work. But as long as there are enough alternative moves, it seems to be a matter of chance who gets to capture first, and thus the MC player will see this as a reasonably good move.
Of course, such moves can be avoided at the root level, by adding heuristics to prevent them. But they keep happening inside the play-outs, where applying all sort of go-specific rules may slow things down too much. I looked around on cgos, but can't see which players are pure MC, and which use other tricks (like uct, or extra go knowledge). Wasn't Anchorman some version of MC? I have a program playing modified MC (Halgo-1.1-500k) and will set up a traditional MC as well (Halgo-1.0-500k) as soon as I iron out a silly bug. I think my variant plays more reasonable-looking games, but I want to see how it survives the test of many opponents. - Heikki -- Heikki Levanto "In Murphy We Turst" heikki (at) lsd (dot) dk _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
