Re: computer-go: 5x5 Go is solved?

["Andrew Derrick Balsa" <andrebalsa@xxxxxxxxxxxxxxxxxx>]

John,

That's a very interesting suggestion. Basically what you are suggesting
is a way to compress this database.

And in fact just **attempting** to compress this database would teach us
a lot. For example:

1) Try to teach a NN to "walk" the database. Do the weights converge?
After how many games? Try with one and two hidden layers, varying the
number of nodes in the hidden layers. Etc...

2) Try to compress the database using fractal compression or some other
method. Is it at all possible, or is the solution space too chaotic?

3) Try to find a set of rules (heuristics) that completely describes the
solution space. How large or small is this set of rules? Compare these
rules to the explanation of a professional Go player when he "walks" the
database.

And so on and so forth.
--
Andrew D. Balsa
andrebalsa@xxxxxxxxxxxxxxxxx



On Sun, 20 Oct 2002 22:44:26 +0200, "John Tromp" <John.Tromp@xxxxxxxxxxxxxxxxx>
said:
> > More I think about it, more I am convinced, that something like this
> > should already be possible for 5x5 Go with todays hardware-possibilities
> > and that it would be a very funny thing for every Go player to
> > "surf around" in such a database.
> 
> such a database could be made quite small in fact.
> fix a standard go program, like gnugo.
> assume that the program can order all legal moves (plus pass)
> in a position from 0, 1, ... up to k-1, supposedly from best to worst.
> then if move 0 really is an optimal move, we don't store this position in
> the database. if the first optimal move is number i, then we add
> an entry for this position mapping it to i.
> the stronger the program, the smaller the database.
> 
> figuring out the result of a position is now just a matter of playing it
> out with help of the database, until there are 2 consecutive passes.
> 
> actually, gnugo is not the most suitable for this purpose. maybe some
> custom program can be made to do much better...
> 
> regards,
> 
> %!PS                       %  -John Tromp (http://www.cwi.nl/~tromp/)
> 42 42 scale 7 9 translate .07 setlinewidth .5 setgray/c{arc clip fill
> setgray}def 1 0 0 42 1 0 c 0 1 1{0 3 3 90 270 arc 0 0 6 0 -3 3 90 270
> arcn 270 90 c -2 2 4{-6 moveto 0 12 rlineto}for -5 2 5{-3 exch moveto
> 9 0 rlineto}for stroke 0 0 3 1 1 0 c 180 rotate initclip}for showpage
> 
-- 
http://fastmail.fm - One of many happy users:
  http://www.fastmail.fm/docs/quotes.html