On Thu, 2008-09-25 at 18:55 +0200, Gian-Carlo Pascutto wrote: > Don Dailey wrote: > > > The only thing I know to check is to see if I am sending the proper komi > > to the programs. The only other possible glitch is that the version > > of leela I am using is ignoring the komi I send - but I don't think this > > is the case. > > The problem was that Leela reset the komi on a boardsize command, which > is against the GTP spec.
I worked around that problem by issuing the komi command last. The correct komi from Leela's point of view (at the level I'm testing) appears to be 2.0. At 1.5 komi black wins 10 out of 10. At 2.5 komi black wins only 121 out of 470 which is 25.74% At some point I'll build a tree with statistics from these games, to see if black is making some common mistake which would indicate a higher komi. - Don
Description: This is a digitally signed message part
_______________________________________________ computer-go mailing list firstname.lastname@example.org http://www.computer-go.org/mailman/listinfo/computer-go/