Re: computer-go: Searches

[julianr@xxxxxxxxxxxx]

David Fotland wrote:
> 
> At 11:50 AM 11/20/99 +0100, you wrote:
> >
> >I can think of a few different ("local") look-ahead procedures that 
> >should exist in most programs:
> >  - ladder reader: read as deep as it takes, but not very wide (terminate
> >    when the running group has three liberties.
> 
> Many Faces reads until the group gets 5 liberties, two eyes, seki, or
> is captured for each string/block.  Each is read 3 times to distinguish
> unconditional vs kos, and interactions with nearby friendly groups.  
> Results are cached to avoid rereading.  Search is deep but not wide.
> Up to 80 ply deep.  Beyond the first few ply, only 1 or 2 cantidate
> moves are tried, so the move generator is quite complex.  Each seach is
> limited to about 120 unforced nodes.  I use alpha-beta depth first search
> with a small transposition table per search.
> 
Is this in order to determine connectivity of groups, or is it actually
used to determine moves? I am thinking of the proverb "only amateurs
play clever moves". Playing a move because it starts a ladder which
bounces around the board and works when read to 80 ply would seem like
a clever move. Of course, computers may not have much choice BUT to make
clever moves. 

Best wishes,

Julian

--
Julian Richardson                         http://www.dai.ed.ac.uk/~julianr/
Mathematical Reasoning Group, Edinburgh University, Phone: +44 131 650 2720