Re: [computer-go] Search = Bad!

["Vlad Dumitrescu" <vlad_dumitrescu@xxxxxxxxxxx>]

----- Original Message ----- 
From: "Frank de Groot" <frank@xxxxxxxxxxxxxxxxx>
> When you would plot a triangle with a color printer of all possible Go
> games, and every pixel would be a position-result (white when it's a win
> with perfect play for white, black when it's a win with perfect play for
> black, gray when it's undecided yet), you will get a nice-looking, regular
> result, with more and more detail at the edges and increases complexity
> towards the edge like a fractal.

Hi,

A thought: black/white/gray isn't really enough. One should either have such a
diagram for every komi value, or instead store a value (between -361 and +361)
for the best result of the game from that point on.

Another thought: I think the fractal analogy is interesting as a basis for
finding new approaches, but I wonder if it should be taken literally. The game
tree is a tree - how would you map it onto a surface so that you get the fractal
analogy to work? Given such a mapped point, how do we find its successors in the
tree?

A question: what do you mean above by "undecided yet"? If this fractal describes
the game, then all points are decided... otherwise it's just an approximation of
the game, and when reaching such a point one needs to find out its value - by
doing more iterations on the fractal equation, which in my book is, well,
search...

Completely eliminating search means solving the game. Navigating a completely
mapped space is straightforward, if the map is complete (i.e. no gray points),
but this assumes the map can be generated. Otherwise, if one only has a limited
horizon, one needs to search... Or am I missing something?

regards,
Vlad
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/