Sent from my iPhone
On Jul 2, 2008, at 12:37 PM, Jason Galbraith <[EMAIL PROTECTED]> wrote:
I've been looking at RAVE (Rapid Action Value Estimate), which MoGo
uses. The
score of states during simulation is stored in state-action pairs,
which are
all updated with the playouts, rather than just those states visited
in the
tree. How would you store these scores? The number of potential
states
visited seems prohibitively large.
Jason Galbraith
Orego research group
I think it's more like "for each state (in the UCT tree), scores are
tracked for every action"
When I asked in the past, RAVE values for actions taken in a game are
applied to all parent nodes. I keep separate counters for RAVE and non-
RAVE simulations.
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
