On Fri, 2008-05-09 at 16:04 -0700, Carter Cheng wrote: > Hi, > > I have been lurking around in this group for sometime and recently have > become interested in perhaps doing some coding and data gathering for > constructing a simple go bot. I have a few basic questions I was wondering if > people in the group could help me answer-
I recommend trying to find someone to join forces with. Writing a bot can take a lot of time, and it's more fun to get cool things working than a really basic bot. That advice applies even if you don't join my development effort ;) > > 1) How typically do UCT bots score simulations quickly? I am not too familiar > with Chinese scoring rules. My understanding is that UCT bots typically play until there's two passes in a row and random players don't pass until they're out of legal moves. From there, it's easy to count points of territory and stones on the board. All territory and stones have clear owners. Just compare stones+territory for each player (and add in komi). > 2) How do machines define eyes analytically (mathematically) and are there > approximations which are faster to calculate for more traditional or UCT > designs (is this information even used for UCT bots?). An eye is any empty point where the 4 neighbors (up down left right) and have no more than n diagonal points belonging to the enemy. n=0 for corners/edges and n=1 for the middle of the board. > 3) What sort of algorithm is used to match patterns once you have mined them > from some data source? I'm not there yet, but I think I'm just about ready to start this part. I look forward to seeing what other people say as the answer to this question. I think I've heard that most use only 3x3 patterns, which I expect to fit in nicely with eye detecting patterns. > 4) And lastly how does UCT cope with ladders? Another one I have not tackled yet. I know Crazy stone will read out ladders and use them in its decisions, but that may just apply to the search tree rather than the playouts. I guess that's another one I look forward to seeing other responses. _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
