[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [computer-go] Groups, liberties, and such

You could skip the liberty count too, but 
> liberty counts 
> are good for presorting move candidates to improve performance of 
> alpha-beta, for short-circuiting branches that are guaranteed 
> to exceed 
> preset limits (e.g. you can't kill a 4-liberty group in less 
> than 7 plies), 
> etc.

You should also short circuit the final evaluation as well.  If a string has
only one liberty and it's opponent's move, you don't need to make the move.
You already know the answer.  So this should be 6 plies.  Often you can save
one more ply since the string has no way to get more liberties on its
previous move.

Many Faces doesn't have a fixed depth limit for tactical searches, so I
don't use this technique.  Instead, I look at the number of new liberties
available at each liberty, and try to prove that the liberty count must
exceed the limit.  For example, assume the tactician will say string is safe
if it gets 4 liberties.  Then if a 3 liberty group that has two places where
it can get two additional liberties, it can't be captured.  Calculating the
number of new liberties available at each point is really useful for move
generation and evaluation.  I also do this incrementally.


computer-go mailing list