From: Amber Jain
Date: Fri, 14 Jun 2013 14:54:58 -0500
Honestly, I do agree that least square fitting will be a departure from
what LAPACK provides.
Thanks for the suggestion of building my own website for it, and
considering it a decent implementation :)
On 06/14/2013 02:44 PM, Langou, Julien wrote:
( I am going to speak for the LAPACK developers here, without consulting
( I assume that if one of us disagree he/she will just step-up and
contradicts me. )
Thanks for your email. I do not think we are ready for such a departure of
what LAPACK is. LAPACK has a focus on dense linear algebra (included
banded matrices), we do not even support sparse matrices or iterative
methods (for solving linear system of equations or computing eigenvalues).
Incorporating the Levenberg-Marquardt method would really be awkward for
us. This said: there is no question that people out there needs this
method in a plain format (as your implementation is), and there is no
question that your code is a reasonable implementation. So maybe why don't
you release this on your own? Just need a copyright and a license file
and a website and you are good to go. I think at the LAPACK level, we are
already quite busy with our package and this is where our expertise is. It
would be unwise to depart from that model.
Thanks for asking! It is great that you felt comfortable enough for
asking! It is great for us to have these kind of opportunities for
incorporating new items in LAPACK. Sometimes yes, sometimes no.
I see that you are using LAPACK DSYSV in your code and that is great!
On 6/13/13 12:54 PM, "Amber Jain" <ajain5@Domain.Removed> wrote:
I could not figure out any open source code in Fortran to do a
non-linear least square code. Hence I wrote one of my own to do it
(using Levenberg-Marquardt method).
I would be proud if it is any use to LAPACK. I have attached the codes.
It is decently commented. I can of-course modify it according to the
needs of the LAPACK programming style.