Hi!
On Thu, Nov 19, 2009 at 12:35:09AM -0800, Seth Pellegrino wrote:
> I'm working on an implementation of Rémi Coulom's Elo-based move
> prediction algorithm[1], and I seem to be running into efficiency
> issues. With my current implementation, I'm running at around 8 or 10
> playouts per second. I can't imagine that the playouts would be so
> accurate that I would be able to accurately evaluate my next move at
> that speed. It seems that simply visiting each vacant point on the
> board and checking whether that move would be a legal move for the
> current player (without even computing which gammas are present at a
> point) takes me down to about 60-75 playouts per second.
What language are you using for your implementation? 60-75 playouts
per second is still *extremely* slow, it seems you are in for much
optimizing yet. Common speed of heavy playouts is at least _thousands_
of playouts per second (per core of modern CPU), in case of light
playouts it should be getting into tens of thousands.
I'm not sure how much slowdown in general you get by implementing ELO
patterns, the improvement has not been implemented commonly yet in
public codebases (though it does seem very promising, no doubt); I plan
to implement it in Pachi later, but I want to try out few different
things first yet.
--
Petr "Pasky" Baudis
A lot of people have my books on their bookshelves.
That's the problem, they need to read them. -- Don Knuth
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
