Re: [gnugo-devel] games against 3.3.16

[Gunnar Farneback]

SP Lee wrote:
> You are right. The J pattern set the minimum value of this move to 35.
> I would modify the move valuation in this case, thus if a dragon is in
> critical condition, trust OWL analysis instead of the Joseki moves
> unless OWL analysis doesn't have any result.

I don't think this is a feasible approach. Which dragons should we
care about? How large do they have to be? Joseki moves are usually
joseki for a reason, e.g. sacrificing some stones in exchange for
outside influence.

The proper solution is to revise or remove bogus joseki patterns. It
is valuable to know that the pattern databases have evolved over a
long time and some patterns are remnants from a time when the GNU Go
engine in all ways was much more crude than it is today. While a
pattern like EJ61 may have been necessary at some time I'm fully
convinced that GNU Go today can find and properly evaluate such moves
by other means. After removing EJ61, N18 in this position is valued
-0.84, which seems about right.

/Gunnar