| Re: Simple? problem [message #30077 is a reply to message #30073] |
Tue, 09 April 2002 10:23   |
James Kuyper
Messages: 425
Registered: March 2000
|
Senior Member |
|
|
Ivan Valtchanov wrote:
> Hi,
>
> I have a small problem concerning a good programming techniques, so here it is:
> ----------------------
> pro test,nx,ny,image
>
> ; Make some random X and Y arrays
> x = randomu(s,1000)
> y = randomu(s,1000)
>
> ; now give arbitrary weights for each point
> w = randomu(s,1000)/100.0
> ; take the square
> w2=2.0*w*w
>
> ; I want to construct a 2-D image with a specified dimension
> image = fltarr(nx,ny)
>
> ; I want to sum up the contribution of each point as Gaussian
> ; with width=w to the image pixels
>
> for i=0, nx-1 do begin
> xgi = i/float(nx)
> dx = x-xgi
> for j=0, ny-1 do begin
> ygi = j/float(ny)
> dy = y-ygi
> dr2 = dx*dx+dy*dy
> arg = -dr2/w2 > (-20.0) ; to avoid overflows
The only thing that statement prevents is underflows. It prevents arg
from ever being smaller than -20.0. exp(-20.0) is an extremely small
number, not an extremely large one. Ordinarily, underflows aren't much
of a problem.
> image[i,j] = total(exp(arg))
'arg' is a scalar, so exp(arg) is also a scalar. Therefore, the total()
doesn't do anything useful. You could just as well say exp(arg).
> endfor
> endfor
>
> return
> end
> ---------------------------
>
> This is obviously quite unoptimised - two cycles etc. Do you have any ideas, references or do you know if it is already solved in IDL?
>
> I have looked for something similar in David Fanning pages and IDL astronomical libraries but I couldn't find something to adapt, maybe I have missed it?
image = exp((x-(indgen(nx)#replicate(1.0/float(nx),ny))^2 - $
(y-replicate(1.0/float(ny),nx)#indgen(ny))^2))/w2)
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
