Joseph Piche wrote: > I believe the semeai and owl code breaks down pretty bad with this > pattern. It's obvious the best move for black is B2, but when white > responds with C1, gnugo gives up. The response W-C2 B-C1 W-B3 B-A3 is > found with default options. But when you turn the level up, W-C1 is > found. Adding S15b S15c would help there, but W-C1 B-A3 W-B1 B-C2 and > white has options at W-B3 B-A2 and W-A2 which black doesn't know how > to deal with. (Of course the correct response is B-A1 W-B3 B-D1 W-C1 > B-A2.)
That's why there are many tests for followup positions in seki.tst. Please see http://trac.gnugo.org/gnugo/attachment/ticket/175/seki_tests_7_12.1 for a bunch of followup tests in this case. They are not very easy to follow by hand (due to play/undo commands) but can be comfortably viewed with view.pike. The results don't look to bad, except for the failure to find the response to B2-C1. Actually the semeai reading does find A3 but find_moves_to_make_seki() in semeai.c fails to call the semeai reading. /Gunnar _______________________________________________ gnugo-devel mailing list gnugo-devel@gnu.org http://lists.gnu.org/mailman/listinfo/gnugo-devel
