[computer-go] Re: Scaling monte carlo up to 19x19

[Nick Apperson]

I believe this to be a good idea, but I couldn't get around some minor
problems.  Essentially, because the local searches are coupled to one
another, it ends up exploding as you consider this coupling (scaling to
larger regions).  You then have to trade accuracy or have more computing
power than I have access to...  There are cases though where you can find
local solutions that are independent in some sense from the rest of the
board and the program I am working on now does that.  I'm sure someone
smarter than me will be able to figure out a better way though.  There are
certain times when this technique is highly useful.  For a simple example,
imagine a board with two walls down the middle bordering on each other.  In
some sense, as long as both those walls live, there are 2 subgames taking
place.  This type of situation is where I see your approach as being the
most useful.  Just my thoughts.

- Nick

On 1/30/07, Weston Markham <weston.markham at gmail.com> wrote:
>
> I have an idea in the back of my mind that is an extreme version of
> this:  Divide the board into 361 separate local searches, then use
> information from these to guide a global search.  The local searches
> would be done on the full board, but would only search for strategies
> that will capture or defend individual intersections.  I suspect that
> this first phase could potentially benefit from parallelization for a
> significant portion of the game.  Eventually this parallelism will
> break down because there will only be a limited number of local
> battles, and the eventual status of points that are in the same chain
> will almost always be the same.  Any practical program would need to
> deal with this gracefully, of course, rather than duplicate its effort
> many times.  Also, I only have a vague idea of how to take advantage
> of the information gained from the local searches, when performing the
> global search.
>
> Weston
>
> On 1/30/07, Dave Dyer <ddyer at real-me.net> wrote:
> > The idea isn't more than lightly toasted (less than half baked), but
> > the kernal is turn the full board search into set of searches on
> > much smaller boards, using the overlapping strips as boundary
> > conditions, then do some unifying final step to pick the move.
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://computer-go.org/pipermail/computer-go/attachments/20070130/ef4fbbb1/attachment.htm