RE: computer-go: board evaluation

["Magnus Persson" <magnus.persson@xxxxxxxxxx>]

> Chris Fant wrote:
> I would like to know what type of board evaluations people are trying.  The
> most effective, I'm sure, begins with a local tactics searcher trying to
> determine life/death status of groups, but I'm mostly interested in faster,
> static evaluations.  I know there is the radiating influence idea.  What
> else have people tried?  

My new experimental version of Viking (viking4 on NNGS) uses patterns 
to detect connections, eyes, and nakade in combination with influence 
to determine territory and/or potential eyehape. It uses this 
evaluation function with global search 1 ply (2 ply for some local 
obvious moves) for both colors. It has no tactics. (but it will some 
day). Since january I have added about 5000 patterns and I guess I 
have to add another 5000 and debug those that has been entered. 

It plays well in many situations, but if any vital pattern is missing 
it plays bad moves. In a typical middle game position against a 
stronger player it almost always beleives all opponent groups are 
dead because the missing patterns accumlates, when strong players 
lives in complicated ways. If it has the right patterns it can play 
correct in many life & death situations without deep reading. 

The problem now is that it is not robust. Even if it plays several 
moves in a row as it understands the situation, it almost always 
blunders once in each fight and collapse. When tactics and better 
global search is added I hope that the remaining flaws will vanish, 
but currently my plan is get this simple static evaluation strategy 
as far as posssible.

It is currently 24* on NNGS and if you are curious you are of course 
welcome to try it.
--
Magnus Persson
Department of psychology, Uppsala University
Box 1225, SE-751 42, Sweden
018-471 2141 (work), 018-460264 (home)
070-2987879 (cellular)
MAILTO: magnus.persson@xxxxxxxxxxxxxxxxx
URL: http://www.docs.uu.se/~magnuspe