Hello,
On Sun, Apr 06, 2008 at 08:55:26PM -0600, David Silver wrote:
> Here is a draft of the paper, any feedback would be very welcome :-)
>
> http://www.cs.ualberta.ca/~silver/research/publications/files/MoGoNectar.pdf
you are saying that in minimax, opponent moves are selected by
minimizing the lower confidence bound - this seems novel, is that so? I
always got the impression that for the opponent moves, you reverse the
mean value but still use UCB.
Is the FPU heuristics not mentioned in the paper only for space
reasons, or did it prove not to be so useful in practice in the end?
I see three unclear details about the RAVE algorithm: Are only nodes
in the tree considered for the inclusion, or also moves in the following
random playout? And one of the sentences hints that there is a separate
period of playouts purely seeding the RAVE value before the UCT-RAVE
linear combination takes over - is that so? And it does not follow from
the paper that that the UCB formula is used for the RAVE value as well,
while the ICML paper states that.
I am surprised on the bad effect of the grandfather heuristic and the
good effect of the even game heuristic. I assume that the effect of the
heuristics should accumulate when several of them are combined in the
prior value?
The paper looks very nice otherwise.
By the way, has anyone measured how big influence different weight of
playouts has on the various heuristics? For example, I'm still
struggling to get any improvement from RAVE whatsoever myself, but I
don't know if it is because I'm getting it wrong or whether it might
manifest only when making the playouts stronger (I randomly use only
three hardcoded patterns in the playouts).
Thanks,
--
Petr "Pasky" Baudis
Whatever you can do, or dream you can, begin it.
Boldness has genius, power, and magic in it. -- J. W. von Goethe
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
