Hi!
On Mon, Dec 14, 2009 at 07:46:54PM +0100, Rémi Coulom wrote:
> Petr Baudis wrote:
> > How do you (e.g. CrazyStone) solve the issue? Or do you perform
> >explicit unpruning by sorting the nodes instead of biasing them?
>
> I bias them too. I use move probability, not move gamma, which
> normalizes the value between 0 and 1. I multiply this probability by
> a constant, divide by the number of playouts, and add it to the move
> score.
I hope I'm not asking too many questions :-) - I'm currently
implementing your way of pattern matching in Pachi, and some things are
unclear.
(If you or anyone are curious about the code, it's at
http://repo.or.cz/w/pachi.git/blob/HEAD:/HACKING#l152
http://repo.or.cz/w/pachi.git/blob/HEAD:/pattern.h
but no worries, I'm not asking you to debug my code for you. ;) Also
anyone looking at it please note that it's awfully slow for now, and
probably has many bugs.)
I wonder now, do you use separate set of gammas for simulations and node
biasing? Since I've found that the performance seems very bad if I don't
include some time-expensive features, since the gammas are then very
off; I will probably simply generate two gamma sets, but perhaps it's
enough to do some trick like merging features by computing weighted
(geometric?) averages?
You say you consider patterns that appear at least 625 times, does that
mean that they are _played_ at least 625 times, or that they are among
the move candidates at least 625 times? I'm using similar filter based
on the former, but judging by the numbers it is likely you do the latter?
For your 9x9 tests, were you using the gammas you harvested from 19x19
games? I'm finding that especially border distance gammas seem very
unsuitable for 9x9.
P.S.: It's fascinating that you've managed so great results with plain
UCT, I've always thought in the past that you are using the patterns
with UCT+RAVE. It's strange so few independent implementations of your
method exist... (I now wish I'd use the endless hours I've spent tuning
my rule-based heavy playouts to rather implement the general pattern
matcher.)
--
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
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
