It's possible that the success of MC approach is in large part due to the 
particular topological structure of the Go game space. And this topological 
structure depends on the simulation polcy, because a specialized simulation 
policy ( in contrast to the random play)  takes out a subset of the complete Go 
game sapce. Even though certain simulation policy win more games aganst the 
random play, the subset of the Go game space taken out by this certain policy 
may not be suitable for the MC approach.

DL

>I spent significant amounts of time on this topic, with relatively very
>few success. The only thing I managed to show is that the relation
>between the playouts predictive power for the game result is not well
>correlated to playing strength of the resulting program (this was
>implied in a Mogo paper).


_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to