On Wed, 2007-12-05 at 17:52 -0800, terry mcintyre wrote: > I just finished a few 19x19 games with the freely downloadable version > of Mogo, and noticed that Mogo loses a fair number of points in the > endgame.
This is typical of MC engines... If it's winning by 100.5 points, it'll lose 100 points without a worry. If it's winning by 1.5 points, you can be certain the endgame will be far better. > > I'll email an sgf file shortly, but the pattern is fairly easy to > describe: when territories have been fairly well enclosed, Mogo will > just as often play > inside its own territory as fill a "dame" or neutral point which is > claimed by neither player. It will even ignore common yose plays at > the edges. > > I seldom play Chinese rules, so my understanding may be deficient; > please correct me if so. > > My observations seem to match this description: > http://senseis.xmp.net/?TerritoryAndAreaScoring > > Under Japanese rules, it doesn't matter who plays dame; they don't > count as territory. Under Chinese rules, it is important to play dame, > which > effectively expands one's area. I understand area to mean the count of > one's territory plus one's live stones. > > I won a couple games against Mogo by the simple expedient of patiently > filling in dame, connecting any singletons left in atari, and > capturing any singletons - all of which increase my stones on the > board without diminishing my territory - that is, increasing my area. > As I steadily increased my score, Mogo would play inside its own > territory, which neither increases nor decreases area. I'd pass, Mogo > would pass, and the score would show me ahead by a few points. This > was on a dual-core athlon with 1800 seconds total time and two > threads. > > Reviewing the TerritoryAndAreaScoring page, it occurs to me that there > are several conventions regarding the handling of handicap stones with > Chinese scoring; I was using GoGui to compute the area score, and will > look into the conventions. It is possible that Mogo uses a different > convention than GoGui, and may have estimated the score a bit > differently. More study. > > If yose handling were improved, Mogo ( and perhaps other similar > programs ) would be a few stones stronger. This looks like > "low-hanging fruit," but I could be wrong. > > Terry McIntyre <[EMAIL PROTECTED]> > They mean to govern well; but they mean to govern. They promise to be > kind masters; but they mean to be masters. -- Daniel Webster > > > > > ______________________________________________________________________ > Looking for last minute shopping deals? Find them fast with Yahoo! > Search. > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
