Re: [gnugo-devel] How to teach gnugo wonderful tesujis?

[bump]

> Hmm. This is might not a majority opinion on this list, but I don't like
> pattern driven approaches too much if there is an alternative.
> And I would call it "going around _systematically_" :-) What I imagined
> is to go around the border of big regions of territory, and try
> disconnections there.

Whatever the approach, maybe it's time to start a file
called regression/tesuji.tst for tesuji's that GNU misses.

The term "tesuji" may be too inclusive for this
discussion. It groups together moves that can actually
be classified. The classification which is best for
humans may or may not be best for GNU Go, however.

The book Tesuji Jiten by Go Seigen and Kensaku Segoe
tries to classify tesujis, and their classification is
based on the shape of the key move. For example, all
tesujis involving a clamp (hasami tsuke) go into one
chapter. Tesuji tend to be moves that occupy a vital
point, and it might even be useful to go around the
board and make a list of moves that look like vital
points. For example vital points might get preference
in move ordering.

This *would* be a pattern driven exercise, for example
the eye-stealing tesuji refers to this shape:

??*
X.?
OX?

But a classification based on shape might not be
appropriate for GNU Go.

The two examples given in the list (by Paul and Arend)
are disconnection tesujis, which might be amenable to
a common approach, such as the one Arend is
suggesting. Other tesujis involve goals other than
disconnection, such as capturing. The atari_atari
module can find a few of these, misses a lot more.

Dan