I'm amazed at how far Lazarus is from evaluating the opening position correctly in 7x7 Go.
When I do a search, even Lazarus believes white has a slight advantage at 5.5 komi although it eventually goes positive (black has the advantage) after several minutes of searching. I've been playing self-test games with Lazarus with a komi of 8.5 which presumably makes all the games a win for black. Still, white is winning 56.6 percent of the games. I ran almost 1200 games with AnchorMan at komi 8.5 and AnchorMan wins about 61 percent of the games, so it appears that the better player is getting a more realistic score, although still far from correct. How clear is it that the correct Komi is 9 and not another value? Has it been proved, or is just strongly believed? And how strongly? - Don On Wed, 2006-10-11 at 08:14 -0400, Don Dailey wrote: > On Tue, 2006-10-10 at 20:28 -0700, David Fotland wrote: > > 7x7 is pretty much solved (by people, not exhaustive search), so it's a much > > easier game than chess. > > Correct komi is 9 points. http://senseis.xmp.net/?7x7BestPlay > > > > David > > > I did a komi test with AnchorMan - a mediocre but not super weak program > at various komi's. > > I used 6.5, 7.5, 8.5 and 9.5. I'm running 500 games but 9.5 has only > played about 200. > > At komi 9.5 White is winning 71.6 percent of the games. > At komi 8.5 White is winning 59.4 percent of the games. > At komi 7.5 White is winning 61.2 percent of the games. > At komi 6.5 White is winning 52.4 percent of the games. > > I suspsect at 5.5 Black will be winning. I'm running this test now for > completeness and black is ahead after 9 games so far. > > I don't have any doubt that the correct komi is 9 but it's interesting > how far off this is. AnchorMan can't win with White even at 6.5 which > is probably because it is a weak program. > > I may try doing the test with Lazarus, which should be playing > significantly stronger. > > I may also try the test with an opening book that gets into the game > with some correct moves. The stat's may change significantly even if I > give white the best 2nd move. > > - Don > > > > > > > > > > > > > -----Original Message----- > > > From: [EMAIL PROTECTED] > > > [mailto:[EMAIL PROTECTED] On Behalf Of Don Dailey > > > Sent: Tuesday, October 10, 2006 10:49 AM > > > To: computer-go > > > Subject: Re: [computer-go] Computer go in 7x7 > > > > > > > > > Yes, it wouldn't be best for building a strong player unless the > > > program assumed a certain komi. But then it gives up if it cannot > > > achieve that komi. > > > > > > Which reminds me - when you run test suites, you have this > > > problem. I have to doctor up problems so that the correct > > > move gives a win and all > > > incorrect choices lose. Or you can specify KOMI for each individual > > > problem. > > > > > > 7x7 is more like Chess in the branching factor, but it's still a much > > > harder problem than Chess. In the beginning, it's worse than most > > > chess positions. > > > > > > If anyone is interesting in trying 7x7 and has a linux > > > machine available, perhaps a spare home computer with a > > > resolvable inet address > > > and web server, we can put up a version of CGOS for 7x7. > > > Just let me > > > know. > > > > > > > > > - Don > > > > > > > > > > > > > > > On Tue, 2006-10-10 at 13:28 -0400, Chris Fant wrote: > > > > But then you have to change your code to favor average > > > terrirory over > > > > win ratio. > > > > > > > > > > > > On 10/10/06, Don Dailey <[EMAIL PROTECTED]> wrote: > > > > > A good way to structure 7x7 matches, is that you forget > > > komi, and > > > > > just play 1 game as white and 1 games as black, adding up > > > the territory in > > > > > both games. Then you don't have to worry about which > > > komi is correct. > > > > > > > > > > - Don > > > > > > > > > > > > > > > On Tue, 2006-10-10 at 13:02 -0400, House, Jason J. wrote: > > > > > > I think Crazy Stone did very well on 7x7 with an > > > opening book. If > > > > > > you go down that road, I recommend that you consider using the > > > > > > attached file to populate the database. > > > > > > > > > > > > -----Original Message----- > > > > > > From: [EMAIL PROTECTED] > > > > > > [mailto:[EMAIL PROTECTED] On Behalf Of > > > > > > [EMAIL PROTECTED] > > > > > > Sent: Tuesday, October 10, 2006 11:48 AM > > > > > > To: [email protected] > > > > > > Subject: [computer-go] Computer go in 7x7 > > > > > > > > > > > > Hello all, > > > > > > > > > > > > I am wondering about Go in 7x7. I know that the game in > > > this size > > > > > > has no real interest in itself. However, I think that > > > the level of > > > > > > computer go programs > > > > > > is much higher in 7x7 than in 9x9, and it could be > > > interesting to see > > > > > > until > > > > > > where we can go in 7x7. > > > > > > > > > > > > There was a discussion on this list recently about the level of > > > > > > CrazyStone in 7x7, and also (perhaps not on this list, > > > only on kgs > > > > > > I don't remember) of > > > > > > Magnus saying that the level of Valkyria was very good > > > on 7x7. I am > > > > > > wondering > > > > > > if we can make a player that beats almost all humans in 7x7? > > > > > > > > > > > > I have tried to put MoGo on kgs in 7x7, but it looses > > > quite a lot > > > > > > of games. Ok, I have put only 5 minutes time, so it plays quite > > > > > > fast, and perhaps could > > > > > > be better with more time, but it is clearly far from a > > > really good > > > > > > player. > > > > > > > > > > > > I have implemented a meta-UCT (that means that instead > > > of playing > > > > > > a random game after the tree, you make MoGo playing against > > > > > > itself) to generate an > > > > > > opening database. I have made this meta-UCT 100% > > > parallel for a cluster > > > > > > (as > > > > > > the evaluation is now so costly, parallelisation is > > > trivial). I can > > > > > > dedicate > > > > > > quite a lot of computers to this task as I have access > > > to a cluster. > > > > > > So I wonder if you think that this meta-UCT can be effective to > > > > > > generate a > > > > > > good opening database, and this way manage to have a really good > > > > > > computer Go > > > > > > player in 7x7 ? > > > > > > Perhaps usual UCT exploits too much for this task, because my > > > > > > experiments show > > > > > > that the beginning of the tree is quite narrow. Here > > > are the first > > > > > > moves: D4 > > > > > > D5 E5 C4 D3 C5. I have no idea if it is very stupid or normal. > > > > > > > > > > > > Remi was saying that the right komi for 7x7 is 9, but I > > > have made > > > > > > my experiments with komi 7.5, and the meta-UCT predicts > > > that black > > > > > > is almost always loosing. Is it because the blacks > > > moves are very > > > > > > bad? > > > > > > > > > > > > To conclude, I think this question of 7x7 is > > > interesting because > > > > > > we can > > > > > > > > > > > > perhaps have a very good player. What do you think? If you have > > > > > > good ideas to make the meta-UCT generate a good database, I can > > > > > > dedicate a lot of CPU to > > > > > > this task, and would interested to see what happen. > > > > > > > > > > > > Sylvain > > > > > > > > > > > > _______________________________________________ > > > > > > computer-go mailing list > > > > > > [email protected] > > > > > > http://www.computer-go.org/mailman/listinfo/computer-go/ > > > > > > _______________________________________________ > > > > > > computer-go mailing list > > > > > > [email protected] > > > > > > http://www.computer-go.org/mailman/listinfo/computer-go/ > > > > > > > > > > _______________________________________________ > > > > > computer-go mailing list > > > > > [email protected] > > > > > http://www.computer-go.org/mailman/listinfo/computer-go/ > > > > > > > > > _______________________________________________ > > > > computer-go mailing list > > > > [email protected] > > > > http://www.computer-go.org/mailman/listinfo/computer-go/ > > > > > > _______________________________________________ > > > computer-go mailing list > > > [email protected] > > > http://www.computer-go.org/mailman/listinfo/computer-go/ > > > > > > > _______________________________________________ > computer-go mailing list > [email protected] > http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
