On Nov 13, 2007 6:40 PM, Chris Fant <[EMAIL PROTECTED]> wrote: > > No need to get turned off on that. In most cases you don't need to > > shake that much. Remember you only need to get the new stone and its > > direct opponent neighbours connected to a liberty. There's plenty of > > tricks for early termination. Last time I tested it I got ~75k uniform > > random playouts on 9x9 (IIRC libego got ~100k on the same machine). > > > > Erik > > That's pretty good. And I'm guessing your code wasn't optimized to > the extent that libego has been.
Maybe not as much as libego, but I did put some effort into the optimization. I'm sure it can still be optimized further, but for that I would consider some specialized assembler instructions (I believe others here have already done that). > Is it fair to say that heavy playouts are a better use of MC execution time? If you have to choose, yes, but in principle one does not exclude the other. It really depends on the kind of features you want from your representation. If, e.g., you frequently need exact liberty counts then a pure bit-board representation may not suffice. > And does a bitboard > approach make it more difficult to incorporate heavier concepts into > the playouts? For some concepts bitboards are actually quite nice, but for others you may want something else. My latest program has a mixed representation with bitboards and other data structures. Erik _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
