During playouts, I generate local responses first, then captures of groups of size 3 or more anywhere, then random moves. The probability of generating a capture move is 5% times the size of the group.
Most of the experiments when I enabled full board captures, the win rate went down, but in the end with the final combination of other heuristics it gained about 2% win rate over Gnugo. In local responses I check for capture or escape of the last move played and that makes a big difference, but I don't read ladders. David > -----Original Message----- > From: [email protected] [mailto:computer-go- > [email protected]] On Behalf Of Mark Boon > > David Fotland said he has a low probability on capture, but I don't > think he ever gave specific numbers. > > Mark > _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
