[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Fuseki and Joseki Database with Neural Network




On Fri, 24 Jul 1998, Juergen Kahnert wrote:

> MX> One question: how long should the GA string be in order to encode
> MX> all the responses in it?
> 
> Good question. :)
> I don't know.
> 

	Let's do some calculation.
	1. If we use one string to represent the whole board (one big
pattern), you know how many strings we need: roughly 3^361, which is
bigger than the number of atoms in the universe. So, let's forget it.
	2. If we use multiple strings (1 pattern/string), then the
question would be how many strings do we want? Or how long a string should
be? The shortest string is 1, but this is obviously useless. Suppose we
use 19 strings (patterns) to match a board, then each string would be 19
chars long. Then we need 3^18 ~= 10^9 patterns (strings), still a bit
longer than we can handle. Suppose we want the number of GA strings to
be in the 10^7 range, let x be the string length, then 3^x = 10 ^7 =>
x~=15. i.e., each string would be about 15 chars long, ~ 4 x 4 range on
the board (too small?).
	Then how could you weight the 361/15 ~= 24 strings (patterns)?
	
	Thanks.

-- Mousheng