I'm not sure if everyone does this already, but I realized that at the end of a random playout there are no territories bigger than one intersection, so you don't really need to call your area score method in most cases.
Let D = blackStonesOnBoard - whiteStonesOnBoard + whiteStonesCaptured - blackStonesCaptured - komi. If abs(D) > K for some small threshold K (I use size/2 = 4 for a 9x9 board, but you can use size if you want to be super safe) then D>0 means black wins, white wins otherwise. The only problem is when D is close to 0, since the real area score also counts each eye as one point and this is not reflected in D. So depending on how many eyes each side has, the result might go one way or another. We can either keep track of the number of eyes and update D accordingly, or we can just call the area score method for these borderline cases. If we do the latter we are still saving a lot of time in most cases if we choose K small enough. This gave me a nice speedup of 10% or so if I remember correctly.
_______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
