Thank you for your theoretical ideas. I will look them and see if I can improve my algorithm.
However, I fixed a bug from the algorithm so that SGF records now are working properly. And I improved the kou recognition ability. It still cannot fill the kou, but at least it can try to avoid taking it right back, although there are some minor problems with it. Now playouts are as they should be around 110 moves in length at 9×9. I also enhanced the efficiency of algorithm, that it breaks the loops if they found what they were looking for. Now it is 100 times faster than the original version. This means that playout takes perhaps only about 10 milliseconds in 9×9, which is enough for a while, that it is worthy to try to make an evaluation function. As the algorithm is now fast, board size can be chosen arbitrarily. Tessa version 2.0 at 19×19 board http://valkonen.kapsi.fi/tessa.php?board=19 –Jouni On 9 December 2011 14:54, Álvaro Begué <[email protected]> wrote: > You can accelerate the detection of whether a chain ran out of > liberties or not dramatically by using additional data structures. For > instance, you can keep an array of the same size as the board > indicating which stone is the leader of the chain. It really doesn't > matter which stone you make the leader, but just make sure that all > the stones in the chain have the same leader (when you join two chains > you change who the leader is for one of them, to make them match). You > can then have another array with the size of the board which is > indexed by leader and tells you how many "pseudo-liberties" each chain > has. A pseudo-liberty is an adjacency to an empty point, but you count > the point multiple times if there are multiple stones in the chain > that touch it. This number is much easier to update incrementally. > > In dimwit we also have an array of 16-bit integers which tell us the > contents of the 8 neighbors, so we can determine if something looks > like an eye with a single lookup into a precomputed table. > > I think these ideas can make your program much, much faster. > _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
