Welcome on the list, Trevoke. :-)
On Sun, Nov 01, 2009 at 08:20:15PM -0500, Aldric Giacomoni wrote:
> terry mcintyre wrote:
> >
> > In general, the time taken to run 1000-game tests hinders research.
> > That's one avenue to explore better solutions, perhaps.
> >
> > Warm Regards,
> > Terry McIntyre
> >
>
> Rene, Terry,
>
> Thanks for the warm welcome and the information!
> Terry, I'll have to explore that, but it seems to be a corollary of The Way
> Things Are more than anything else, isn't it?
Well, it would be interesting to find a metric that is quicker to
compute and shown to be strongly correlated to game tests in wide
variety of conditions... It would be of immense practical value
especially since it would enable much quicker and thus better tuning
of various constants and such. However, the task itself seems quite hard
to me and kind of "janitorial" in the field.
> Someone did send me a link to the Go Library - there is now tons of stuff to
> read in there.
> Most of the programs out there now are Neural Networks, it seems. Are there
> any
> who tried to play with knowledge hard-wired in there, such as Smart Go
> (http://www.smart-games.com/knowpap.txt) ? And if so, what is that knowledge?
GNUGO is a prime example of popular program with hard-wired knowledge,
though it's quite a few stones weaker than the strongest MCTS programs
nowadays. I think more work on integrating MCTS in GNUGO in a smart way
could make for a very interesting project.
> On the topic of academia.. While I would love to actually do graduate studies
> around the game of go, I may have to do something boring, like solving a
> real-world problem that would make people's lives better (gasp). Does anyone
> have an idea of real-world problems which could be correlated to go as far as
> AI
> goes (besides developing my own brew of Psychohistory) ?
That depends on the particular method you choose to research, I guess.
MCTS is generally usable for wide variety of planning tasks with
difficult-to-create evaluation function and very wide search space, and
MCTS Go research is often sold as a precursor of such. I'd be very
interested in some concrete application examples as well, though.
> Does anybody here use Ruby at all for coding? Or is everyone using lower-level
> languages like C++ ?
Java is somewhat popular, but most people use some variant of C I think,
mainly since MCTS is performance-critical task. If you'd choose a
different approach than MCTS, different language choices might make
sense as well.
--
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/
