[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[gnugo-devel] giant-sized go
From: |
Joe Corneli |
Subject: |
[gnugo-devel] giant-sized go |
Date: |
Thu, 20 May 2004 02:49:45 -0500 |
> 2. has anyone tried making gnugo play on large-sized boards (say 100
> X 100) and if so with what results?
That large, no, never. I'm not aware of any game ever played
by anyone on such a large board. Even 23x23 is substantially
longer and different from 19x19.
Hm -- well maybe its my background in continuous mathematics that's
been nudging me to think that this might be an interesting problem
to look at. Not that 100 X 100 is continuous by any means, but it's
a far sight closer (in a certain sense) than 19 X 19.
And luckily the computer wouldn't get bored, though with circa
10,000 moves possible and who knows how many to consider??, it might
take a rather long time to run the computation.
But in addition to having what I bet would be very aesthetically
interesting output, I imagine that one could might possibly learn
some interesting things about the smaller "human-sized" game from
observing properties and patterns that appear in strictly
"supercomputer-sized" play. But who knows what one might see in a
game that big!
I think that the real question would be: how to evaluate moves?
Almost any individual move that I can easily conceptualize would be
so "small" in the giant-sized game that it seems hard to assign a
reasonable weight. To begin with, we might inherit weights from the
normal game, and expect to see a lot of play around the edges at
first... at least, I guess so. But later it could be that one would
see some degree of self similarity appearing in the game -- so that
having a live group roughly 4/19ths of the way from each corner
would be a huge boon ;). Well, once again, who knows!!
If anyone is interested in looking into getting some supercomputer
time & want to play around with me on this stuff, please tell me!