Depending on your implementation, it may be faster to re-derive the
liberty list when you need it. For example, if your playout move
suggester only suggests capturing the last stone played, you don't
need to do all of the work to update liberty counts for any other
chains.
Thanks a lot for everyone's help here. I think I have my ladder-
reader working, although my program was still playing out ladders.
The solution turned out to be something completely different: turn up
the "exploration" coefficient in the UCT formula. If it's too low,
the program commits to a particular move too early, builds up a lot
of playouts through that move, and keeps playing the move even though
it has read a sequence of follow-ups that lead to a bad result.
Peter Drake
http://www.lclark.edu/~drake/
On Aug 2, 2008, at 7:06 AM, Álvaro Begué wrote:
On Sat, Aug 2, 2008 at 9:43 AM, Jason House
<[EMAIL PROTECTED]> wrote:
On Aug 2, 2008, at 4:31 AM, Gunnar Farnebäck
<[EMAIL PROTECTED]> wrote:
It's often a good idea to bias capturing moves in the playouts,
regardless whether it's a ladder or not. This would result in those
stones being captured in most simulations.
What method do people use for finding capture moves in playouts?
Pseudo
liberties can miss simple stuff like open triangles and one-eyed
groups.
Additionally, some literature discusses captures to add group
liberties.
What's the preferred method to detect
that?_______________________________________________
When we wrote dimwit, John Tromp found a fast method that I described
here: http://computer-go.org/pipermail/computer-go/2007-November/
012342.html
However, my current thinking is that it's probably best to just keep a
real liberty count and a list of liberties for each chain. This way
you can also find atari moves, which would be very hard to do if you
only keep pseudo-liberties.
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/