ScientificPython - LeastSquareFit diverges

[Konrad Hinsen]

On Jul 19, 2006, at 16:53, Harold Fellermann wrote:

> :-) Anyway, I had
> the impression that the leastSquaresFit in Scientific Python is an
> implementation of
> the Levenberg Marquardt algorithm as it is presented in the Numerical
> Recipes.

True.

> Accoring
> to reviews, this algorithm is not famous for its stability
> (e.g. http://www.stanford.edu/class/cme302/wnnr/nr.html). Better
> implementations
> are out there (e.g. http://www.ics.forth.gr/~lourakis/levmar/). Are
> there any plans to
> improve the SciPy algorithm? Would it be a welcome contribution to
> SciPy to work
> this part out?

Yes, definitely. And no, I have no plans to do it myself any time  
soon. The current implementation has always been sufficient for my  
needs, and time is scarce...

BTW, ScientificPython (http://dirac.cnrs-orleans.fr/ 
ScientificPython/) is not the same thing as SciPy (http:// 
www.scipy.org/). Both are scientific libraries for Python, but their  
focus is different: ScientificPython aims at providing pythonic  
modules for scientific computing, whereas SciPy's objective is to  
provide Python interfaces to the large pool of scientific libraries  
from the Fortran/C/C++ world. Scientific users of Python should  
probably have both of them installed.

Konrad.
--
---------------------------------------------------------------------
Konrad Hinsen
Centre de Biophysique Moléculaire, CNRS Orléans
Synchrotron Soleil - Division Expériences
Saint Aubin - BP 48
91192 Gif sur Yvette Cedex, France
Tel. +33-1 69 35 97 15
E-Mail: hinsen at cnrs-orleans.fr
---------------------------------------------------------------------