Re: computer-go: Data Mining and Rulebase for GO

[Matthew John Saffell <saffell@xxxxxxxxxxx>]

On Sun, 12 Nov 2000, Ray Tayek wrote:

> At 06:26 PM 11/12/2000 -0600, you wrote:
> 
> >  ... I am fighting to get my compute 4 times faster. 360 times faster is not
> >a small number. It may take Intel 16 years to do it.
> 
> the programs will play faster, but not much better.
> 

I think the point is that 3^360 is so big that dividing by 360 is a drop
in the bucket.  For the sake of argument assume that getting rid of
symmetries and throwing out illegal board positions gets the total number
of board positions to be considered down to 3^100 (?).  That reduction is
HUGE - but 3^100 is still roughly 10^47.  Of course dividing by 360 is 90
times better than dividing by 4, but it only reduces the problem to 10^45.

I didn't understand the details of your scheme totally, but if you're
trying to store a bit of information (really 2 bits if you say
"win/lose/draw") for each of your big numbers - then the scheme takes up
10^47 bits of space.  Or roughly 10^34 terabytes.  So irrespective of the
time it takes to produce these bits of information, you still have to
store them somewhere, and I dare say it won't be on planet Earth.  

So there are big numbers, and there are BIG numbers.  

Matt


> we are going to have to get very clever to make any progress with go. i 
> suspect the programs will hit a brick wall around the 4 kyu level (on a 
> 19x19 board) in the sense that most 4 kyu's will easily learn to beat all 
> the programs.
> 
> 
> >----- Original Message -----
> >...
> >At 11:17 PM 11/11/2000 -0800, you wrote:
> > >...
> > >The purpose of finding the smallest large number for a board configuration
> > >is to set a standard board representation,...
> >
> >sure, ...
> >...
> >perhaps, but 360 is small increase. it probably won't make any difference.
> 
> ---
> ray tayek http://home.earthlink.net/~rtayek/
> orange county java users group http://www.ocjug.org/
> want privacy? http://www.freedom.net/
> hate spam? http://samspade.org/ssw/
>