This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
multifit/Levenberg-Marquardt
- From: Sanjay Bhatnagar <sbhatnag at aoc dot nrao dot edu>
- To: gsl-discuss at sources dot redhat dot com
- Date: Fri, 19 Sep 2003 08:59:27 -0600
- Subject: multifit/Levenberg-Marquardt
- Reply-to: sbhatnag at aoc dot nrao dot edu
Hi,
I have been using GSL for my work on image deconvolution. I need to
use the Levenberg-Marquardt algorithm for Non-linear minimization.
However the problem I am solving involves large data size as well as a
large no. of parameters.
If I use the GSL implementation, I will have to allocated the
Jacobian which is of the size NxP were N is the data size and P the
no. of parameters. For me N=1024x1024 or more and P few hundred.
Hence, holding the entire Jacobian in the memory is impractical.
Also, it's most efficient for me to allocate and manage the data
array in the user code.
Is there a GSL implementation which works with lesser memory
requirement? E.g. one which needs the full derivatives (the integral
of dChisq/dParam over all data) rather than needing an array (the
Jacobian) which holds the derivatives evaluated at each point. The
size of the array in the latter case is NxP! I feel that in the
current form of GSL implementation, it's un-usable for large
problems. Which is disappointing.
Regards,
sanjay