I hate to keep calling this MC-UCT because many programs do not use UCT
so until we come up with better terminology I'm going to call it MC-BFS
for Monte Carlo with Best First Search.
The other aspect of the picture is that MC-BFS is a new technique.
It's not reasonable to believe that we have taken it nearly as far as it
will go. It may be the case for any given individual that they cant' t
Petri Pitkanen wrote:
Well MC-UCT programs (Mogo, CrazyStone and many others) need more CPU.
Doubling the CPU gives a constant raise in playing level. There are
even some threads about the issue. But becoming -say - 6 stones
stronger I guess they would need more then 2^6 times more CPU. Also
there could be a corner point for current algorithms- so maybe they
need new playout algorithms in future?
CPU power is only one aspect of the picture. Having a scalable
algorithm is a vital part of the picture - it means you are not running
up against a hard limit. Also history has proven that enormous
computing power that seems out of reach is really just a few
months/years away.
I personally believe that "corner points" are imagined in many
cases. It is usually a question of credulity - most people take time
to adjust to a new idea. Although some over-adjust, the most typical
reaction is to believe anything new and amazing must be broken. The
human mind trying to justify, or at least ease the transition from
previously held misconceptions.
Good scalable algorithms have no "corner point" - but that doesn't mean
they cannot be improved by things like "new playout algorithms in the
future" just as you say. In fact, I consider that part of the
process. We have this new idea in computer go, it's wonderfully
scalable, but by no means have we exhausted the possibilities.
Even computer chess has seen amazing progress in software in recent
years - totally amazing stuff. But none of it especially innovative,
just finding better ways to do the same basic things. I think
computer Go has major refinements possible in the future that is going
to be worth several stones even without extra hardware or major new
ideas - and new ideas are sure to come.
It's really quite exciting to see the recent progress. Although we
should probably expect incremental progress, the last couple of years
has seen an explosion of progress.
Traditional programs may not even gain on CPU alone (GnuGo, ManyFaces
of Go, GO++). Depends on actual implementation. But for example GnuGo
would hardly be any stronger with more CPU. So it would need new
algorithms and more "expert" rules.
Traditional programs seem to be limited in scalability, but it seems to
me that they do have limited scalability in the sense that the
programmers keep refining them to play better on faster hardware.
Would the latest gnugo run on a 80286 and move in a reasonable amount of
time? Probably not.
2008/7/28 Darren Cook <[EMAIL PROTECTED]>:
My question isn't about how strong programs are now, or what is the
definition of a dan, or what you think will happen in the future. The
question is: what do you need to give your current 19x19 program another
6-ish ranks in strength (or 6+N where N is the distance between your
program and the top programs).
It's a real shame to me we don't have a standard rating system. But
did you notice there has been a real paradigm shift here? We are no
longer talking about what is needed to get to 1 Dan (whatever that
means.) We used to lament the fact that 1 Dan was so
unapproachable. Now we are sad and discouraged that we can seem to
get to 6 Dan.
- Don
Darren
--
Darren Cook, Software Researcher/Developer
http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic
open source dictionary/semantic network)
http://dcook.org/work/ (About me and my work)
http://darrendev.blogspot.com/ (blog on php, flash, i18n, linux, ...)
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
