[SciPy-user] Flux conservative rebinning

[Yannick Copin]

Hi,

I'm looking for a "flux-conservative" rebinning scheme, i.e. resampling 
an input vector while consering the total "flux" (i.e. sum).

As for now, I'm aware of basically two techniques, none of them suitable:
1. boxcar-filtering and sub-sampling (e.g. 
http://article.gmane.org/gmane.comp.python.numeric.general/888)
2. interpolation (e.g. http://www.scipy.org/Cookbook/Rebinning)

For example (attachment), given

y = array([0,2,1,3], dtype='d')

with a total sum of 6.0, boxcar-filtering and sub-sampling (x2) would be:

yy = convolve(y, ones(2), mode='valid')[::2]

giving yy = [ 2.  4.]  sum = 6.0. The problem is that I can reduce size 
only by an integer factor (e.g. I cannot rebin y on 3 points), and 
furthermore it looks like a waste of CPU to compute a convolution on 
full input vector and *then* sub-sampling.

Using recipe #3 of http://www.scipy.org/Cookbook/Rebinning allows "any 
dimension sizes" rebinning, but unfortunately does not conserve the flux:

yy = congrid(y, (3,), method='linear', centre=True)

gives yy = [ 0.33333337  1.5         2.66666698]  sum = 4.5.

Finally, what I would have liked is yy = [ 0.66666667  2. 
3.33333333]  sum = 6.0, which is

yy = dot(mat, y)

with mat = array([[1.,1./3.,0.,0.],[0.,2./3.,2./3.,0.],[0.,0.,1./3.,1.]]).

Any way to efficiently construct the 'mat' array?

Cheers.

PS: BTW, method='neighbour' in congrid of example #3 of 
http://www.scipy.org/Cookbook/Rebinning does not work out of the box and 
raises:

NameError: name 'xi' is not defined

Is the code still incomplete?
-- 
    .~.   Yannick COPIN  (o:>*  Doctus cum libro
    /V\   ---===<<<### NOT IN THE OFFICE ###>>>===---
   // \\  Institut de physique nucleaire de Lyon (IN2P3 - France)
  /(   )\ http://snovae.in2p3.fr/ycopin/
   ^`~'^
-------------- next part --------------
A non-text attachment was scrubbed...
Name: rebin.py
Type: text/x-python
Size: 5915 bytes
Desc: not available
URL: <http://mail.scipy.org/pipermail/scipy-user/attachments/20060706/c5a4b696/attachment.py>