Re: computer-go: perfect play : a new fractal algorithm

[Matt Gokey <mgokey@xxxxxxxxxx>]

Serge Boisse wrote:

>>Since cute and intelligent algorithms for computer go tend to be very rare ;-),
>>may I Propose you a new one ? You'll see that it is definitly completly
>>different from anything else.
>>...
>>I call this algorithm "fractal go".
>>...

Well, I found your idea very interesting!   As I have said multiple times, I believe
go-representation - especially multiple level representation - and reasoning using
higher level abstractions is the key to dealing with the complexity of Go.  There
have been several different approaches introduced on this list that differ greatly
in the details, but have this common thread.  Your idea is another one.  I believe
that despite the apparent problems with the Fractal-Go reduction approach losing too
much information, that it may have some merit.  The local searches for each reduced
board give this search its potential.  An advantage of this idea is that it is
relatively simple.  It would be interesting if someone were to implement this
algorithm and see how well it might play.  Of course, there is still the issues of
searching locally even at the reduced branching factor, life and death analysis, and
whole board evaluation, etc.  If one of us with a full functioning go program added
this Fractal Go "heuristic" to their search for selection/pruning I wonder what the
results would be?   Personally, I tend to think that the changes in the strategic
nature of the game (as the board size shrinks to 11x11 and lower) due to the edges
and corners and proximity of stones, not to mention shape changes, will skew the
strategy on the larger boards, resulting in a lot of mistakes.  However, it would be
interesting to actually find out...

Matt