I wonder if it's common for both gnugo and MFGO to agree on a wrong score? In computer chess it's not uncommon for 2 program to want to play the same weak move for the same (incorrect) reasons.
Don On Mon, Apr 26, 2010 at 12:24 PM, David Fotland <[email protected]>wrote: > This is a change of topic, since the final score is done by Many Faces’ > engine, not the MC playouts. We were talking about MC support for Japanese > rules in order to play well in a game that is being scored by Japanese > rules. > > > > Final score accuracy is not perfect under either Japanese or Chinese > rules. Correct scoring by computer is either case is about the same > difficulty. MF is comparable accuracy to gnugo. In test games, the two > programs disagree on the score about 1 in 200 games. > > > > David > > > > *From:* [email protected] [mailto: > [email protected]] *On Behalf Of *Don Dailey > *Sent:* Monday, April 26, 2010 6:43 AM > > *To:* [email protected] > *Subject:* Re: [Computer-go] Many Faces and Japanese rules > > > > David, > > Do you have a good sense of the success rate of your scoring algorithm? > Is it better than Gnugo for instance? Does it get 99% of the scores correct > or is it higher, or lower? > > It's my understand that correct Japanese scoring by computer is almost > impossible to get right in every case, although one can come relatively > close. > > Don > > > On Mon, Apr 26, 2010 at 2:00 AM, David Fotland <[email protected]> > wrote: > > It is a little tricky to make it work. > > > > -----Original Message----- > > From: [email protected] [mailto:computer-go- > > > [email protected]] On Behalf Of Martin Mueller > > Sent: Sunday, April 25, 2010 1:52 PM > > To: [email protected] > > > Subject: [Computer-go] Many Faces and Japanese rules > > > > Hi David, > > > > I kind of see how to do it in the tree, but can you handle Japanese > > > rules even in the playout phase? It seems very tricky to me. Fuego will > > happily play extra defensive moves in its territory, as long as it is > > beyond the search tree, since all the playouts are scored with Tromp- > > Taylor rules and are still wins. > > > > See e.g. > > > http://fuego.svn.sourceforge.net/viewvc/fuego/trunk/regression/cleanup_ > > japanese_rules.tst?revision=1105&view=markup > > http://fuego.svn.sourceforge.net/viewvc/fuego/trunk/regression/sgf/game > > s/2010/LittleGolem/ > > > > > Martin > > > > > The pure MC part of MFGO handles Japanese rules correctly. The only > > point > > > I was making is that there is nothing inherent in the MC algorithm > > that > > > makes it not suited to Japanese rules. > > > > > _______________________________________________ > > Computer-go mailing list > > [email protected] > > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go > > _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go > > > > _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >
_______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
