[SciPy-user] nonlinear fit with non uniform error?

massimo sandal massimo.sandal@unibo...
Thu Jun 21 05:31:55 CDT 2007


Matthieu Brucher ha scritto:

>     In which sense not Gaussian? In the sense that for each point, the
>     uncertainity is not Gaussian distributed? It should at least with good
>     approximation be. If it is in another sense, please explain...
> 
> 
> If the error is not Gaussian (normally distributed, ...), least squares 
> is not the "most likely" optimization (maximizing likelyhood on gaussian 
> data is the same as least squares), you should use more robust cost 
> functions.

I guess you refer to the distribution of the error for *each single 
point*, not the distribution of the average error in the dataset for 
different points. In this case yes, it is Gaussian, so there should be 
no problem.

My question was different, anyway: each point can have a different error 
size (i.e. sigma of gaussian distribution) respect to its neighbours.

m.

-- 
Massimo Sandal
University of Bologna
Department of Biochemistry "G.Moruzzi"

snail mail:
Via Irnerio 48, 40126 Bologna, Italy

email:
massimo.sandal@unibo.it

tel: +39-051-2094388
fax: +39-051-2094387
-------------- next part --------------
A non-text attachment was scrubbed...
Name: massimo.sandal.vcf
Type: text/x-vcard
Size: 274 bytes
Desc: not available
Url : http://projects.scipy.org/pipermail/scipy-user/attachments/20070621/6ff2223e/attachment.vcf 


More information about the SciPy-user mailing list