RE: computer-go: Live or Die

[Tim Klinger <klinger@xxxxxxxxxx>]

On Tue, 7 Dec 1999, Darren Cook wrote:

> >We have about 500 or so eye patterns right now.  In the tests that we've
> >run most of the errors were due to missing patterns, and only a few were
> >caused by incorrectly entered patterns.
>
> Can you show a few examples of those eye patterns. E.g. the smallest and
> largest patterns. Do you just have empty/black/white? Is there a maximum
> size for the pattern?

I'm not at my desk right now but I'll post a few when I get a chance.  In
the meantime you can get a fair idea of the pattern language by looking at
section 2 of "An Architecture for Computer Go"  at
http://cns.nyu.edu/~mechner
 
> How many of those patterns came directly from you (i.e. things you
> think it ought to know like three-in-a-row is unsettled, four is
> alive), and how many were a response to a problem it got wrong?

Well, I don't know exactly how many we started with but I think most have
been added in response to specific problems.  What tends to happen is that
we notice it's missing a whole class of patterns and add them all at the
same time.  Then test, refine, repeat.

> Or did I misunderstand and
> by training you mean it learns all patterns automatically?

No, we enter the patterns by hand in response to bad play by the program.

Tim