Mogo is very strong, definitely dan level, at all parts of the game (except the first 10 moves, where it can be a bit fuzzy; Note: I've only been testing it at 9x9). It's understanding of sente and gote in a close endgame is outstanding. I thought I had found weak play with monkey jumps, and with ko, but after playing out many variations I concluded Mogo knew what it was doing and I didn't. It makes good shape, and when it doesn't it always seems to be for the right reasons.
Which makes its one genuine weak point stand out all the more. In the FAQ Sylvain says Mogo sometimes "does not read well the L&D status of a group in a corner." This seems to be nakade (http://senseis.xmp.net/?Nakade). See [1] and [2] for examples of where Mogo is going very wrong. By contrast, Many Faces's static evaluation (takes less than a second) understands these positions. Even when the problem position is fully developed and on the board (rather than deeper in the tree) it does not see it, and makes all its subsequent moves on the assumption the group is alive. I.e. it doesn't try to run or attack the neighbouring groups. Sylvain, can you explain more about why it has this particular weakness? Also, is there a plan to fix it? If fixing it deep inside the tree is too expensive, I wonder if a pattern library of problem positions that is only used at the root (and then, later, the first 2-3 ply) might help. E.g. if a probable nakade situation is found flag the points involved as moves that must not be pruned but instead considered with fairly high priority. Darren [1]: Top-left corner. Mogo says (based on its game win confidence) white is alive, but is actually unsettled. After 1..4 it is dead but Mogo still thinks white is alive. (If white plays x this is actually dead by bent-4, so we really need to see the rest of the board, and know the ruleset, to be sure; but, anyway, Mogo didn't play x.) ...... ..OO## .OO##. .O##. .##.. .#.. .3.2.. 14OO## .OO##. .O##. .##.. .#.. .#.O.. #OOO## .OO##. xO##. .##.. .#.. [2]: White group in upper-left is dead. ...O.#O.. ###O.##O. OOOO..##O .O...##O. O####.OO. #...O.... .###O.O.. ...#O..x. ......... Mogo estimates it has a slight advantage (0.51) and plays x; in fact it is losing by around 10pts. In another very similar position (which I could neither track down, nor reproduce) it was convinced (around 0.85) it had a win, until I started filling in the liberties of its dead group and suddenly it plummeted to 0.4, then resigned the move after. However some very similar (to my eye) corner positions it did seem to understand. I'm not sure what the difference was. >From the above position I played very passive moves for black, and white's confidence rose to 0.6, but no higher. I.e. Mogo is seeing white is dead in some paths, but it is seeing white is alive in more paths. -- Darren Cook http://dcook.org/mlsn/ (English-Japanese-German-Chinese free dictionary) http://dcook.org/work/ (About me and my work) http://dcook.org/work/charts/ (My flash charting demos) _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
