The old, knowledge based programs all used large pattern databases. Performance of the pattern matcher was never an issue. I think it was about 3% of the time in Many Faces. If this is going to work you will need some breakthrough in choosing and evaluating the patterns to use.
David > -----Original Message----- > From: [email protected] [mailto:[email protected]] > On Behalf Of [email protected] > Sent: Monday, June 14, 2010 8:02 AM > To: [email protected] > Subject: [Computer-go] re. : Effective search depth > > > So what can above calculation tell us? According to above calculation it > could estimate > > that the effective search depth of the today's strong Go programs are about > 11, if the playout > > number is one million and assume m=600, f=1.5. If an effective search depth > of 50 > > is required to reach high dan level. Then the playout number needs to > increase by a > > factor of 1.5^39, about 7.4 milliom times. That is 7.4 trillion playouts > needed. > > I suspect that Monte-Carlo isn't the best way to achieve high dan level play, > even though > it is the best Go playing method at the moment. > I'm doing research into board evaluation using pattern recognition, with the > key being an > algorithm that can find patterns on a board without the time required being > proportional > to the number of patterns being searched for. > My aim would then be to only do lookahead as far as the next tenuki play, in > the sense > that for every move examined the likely sequence of moves following it would > be examined > up to the point where the next play at each leaf node was a tenuki, and at > that point the > evaluation of the board at the leaf node would be used to min-max the tree of > moves to > decide best move. The depth would obviously vary, but the total number of > moves to > evaluate would be unlikely to be above a thousand. > _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
