On Mon, Nov 12, 2007 at 12:10:03PM -0800, Christoph Birk wrote:
> My (not very optimized) C-code plays 12k games per seconds from the
> opening position. The average game length is about 109 moves using
> an early-termination rule when a "big" group gets captured that leaves
> "most" stones beeing from one color.
This looks like a good technique I should implement too. What "big"
values are popular? I'm thinking size*size/3, but maybe that is too
conservative?
> I strongly recommend using CGOS for tests as you get a lot of games
> against a great variety of (computer) opponents. It allows you in
> particular to verify your (basic) algorithms:
>
> pure MC 10k 1050 ELO
> 50k 1350
> AMAF 10k 1450
> 50k 1450
> UCT 10k 1300
> 50k 1550
>
> All algorithms above are "basic" playouts. No Go knowledge, except
> the "dont-fill-your-1-pt-eye" rule.
Thanks a lot! I'm doing that now and while the ranks are not yet stable,
they are all only slightly above 1050 now already. :-( (Even the
variants with extra domain-specific knowledge.) I guess I still have
some bugs there.
By the way, when I make significant change to the algorithm (even a
bugfix that will significantly improve it), is it considered better
practice to have the changed bot playing from scratch, or introduce the
change to existing bots and see how the rating changes?
P.S.: By the way, so that google picks this up: If you are getting
can't read "mv": no such variable
TCL errors from the CGOS client, that means you are hitting a bug in it
- it does not properly ignore empty lines from the GTP client (or,
heavens forbid, comments). Make sure that you do not emit any. It took
me several 10-minutes waits and time losses of my bot to debug this.
--
Petr "Pasky" Baudis
We don't know who it was that discovered water, but we're pretty sure
that it wasn't a fish. -- Marshall McLuhan
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
