[Pachi] Fuseki Joseki

Petr Baudis pasky at ucw.cz
Sat Jan 21 17:56:03 CET 2012


On Sat, Jan 21, 2012 at 10:38:44AM +0000, jonathan chetwynd wrote:
> on peepo.com pachi v8.99 and previous...
> Pachi never plays corner, and this is considered weak by most.
> Pachi in free H9 game plays faster than 12 sec. in opening and not so well...
> can looks much improved, better variety, again more evident in mid game, rather than opening

  With large-scale pattern, Pachi plays much nicer in opening, and also
acquires some (limited) appreciation for corners.

  But with H9, there is always only so much you can do with MCTS without
major effort. With dynkomi=linear:dynkomi_mask=3, Pachi might play quite
prettier with H9, but also lose *a lot* more, I suspect.

> less important but relevant:
> as per previous mail, score-slope in H9 game needs to be biased so that
> changes in score are expected in mid to end game

  Perhaps. But high handicaps just aren't intrinsically interesting
for me personally to spend a lot of time on them. You are invited to
tweak various dynamic komi behaviour. :-) All the code is quite well
localized to uct/dynkomi.c.

>  is it perhaps the lack of cores and 12 second limit?

  pachiichi plays on KGS with similar version, time settings and also
single core; it's ranked as a very strong 3 kyu.

				Petr "Pasky" Baudis
	The goal of Computer Science is to build something that will
	last at least until we've finished building it.

More information about the Pachi mailing list