I did this, stopping search if other moves mathematically couldn't catch up. I found that the savings in percentage of total nodes depended on how many playouts the program did. The larger the number of playouts, the larger the savings. I also made a rule where after a certain percentage of the total playouts, if one node is far ahead but theoretically another could still catch up, I'd still terminate, which doubled the savings. I don't have the details here but I remember this way I got equal winning-rate at 32K playouts with these rules vs. 64K playouts without.
Mark _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
