Le Lundi 10 Avril 2006 00:39, Gunnar Farneb�ck a écrit : > Today's KGS tournament was 9x9 with very long thinking time, 28 > minutes per player. As it's difficult to get GNU Go to spend all that > time in a meaningful and robust way I decided to make an experiment. > > Attached is a patch to GNU Go 3.7.9, used in the tournament, which > adds a Monte-Carlo evaluation on top of the ordinary evaluation. More > precisely it first makes an ordinary move generation, then takes the > top moves, defined as valued at least one third of the best move, and > reevaluates those by an alpha-beta Monte-Carlo search. (Exception, > fuseki database moves are played immediately to save time.)
Thanks. This is very interesting and looks promising. Congrats. Alain _______________________________________________ gnugo-devel mailing list gnugo-devel@gnu.org http://lists.gnu.org/mailman/listinfo/gnugo-devel
