Re: computer-go: Pattern matching

[David Fotland <fotland@xxxxxxxxxxxxxxx>]

At 03:56 PM 11/6/99 -0500, Gary  Boos wrote:
>MFG has a "...seperate database..."  for the opening.
>MFG got outplayed in the opening by SmartGo at the congress.
>Can you comment on the code (i.e. not the database) that is
>used to help play the opening (which seems to be a weak
>area for MFG) ?
>        Gary

The opening is a weak area for Many Faces, mostly because I
rely too much on rote knowledge rather than detailed
opening patterns.  MF has a big full board opening database
collected from thousands of games.  I use it to bias the
evaluation toward moves that are more popular.  It also has
a big joseki database.  Beyond that there is very little
opening code.  If there is a joseki move available in a corner,
it will not consider non-joseki moves in that corner.  Sometimes
this causes big problems when sequences in one corner spill
over toward the next corner.  This rule sometimes prevents
the program from making the correct extension.

I force a move in an empty corner if one is available.

I have some code to evaluate moves in empty corners to encourage
connected 4-4 corners or shusaku style spiraling 3-4 points.

I try to prevent giving opponent two shimari.

I have some code to evaluate edge extensions and invasions, but it
is not very good.  I wrote it when I was about 10 kyu, and haven't
worked on it much since.

I try to make or prevent double wing formations.

David