1
$\begingroup$

what is your preferred test set to test quality of non linear solver?

this could be set of data, model and results obtained with some benchmark solver, or simply a panel of test functions that could be used to generate synthetic data.

thanks!

$\endgroup$
3
$\begingroup$

It depends a lot on which nonlinear problem you want to solve.
http://plato.asu.edu/bench.html
is a good place to start for local constrained optimization,
http://www.mat.univie.ac.at/~dferi/testenv.html
for global constrained optimization, and
http://archimedes.cheme.cmu.edu/?q=dfocomp
http://coco.gforge.inria.fr/doku.php?id=bbob-2012
for unconstrained black box optimization.

I am not aware of special test sets for nonlinear least squares.

If you know the kind of fitting problem you want to apply it to, you can generate from any known model simulated data and then compare the solutions obtained with the input for the simulation.

But if you want to have a general-purpose constrained least square solver, then you can take anyway hardly any advantage of the least squares structure. Thus it would be sensible to use (or develop) a general purpose nonlinear program solver, and test it on the test sets from one of the above sites.

$\endgroup$
  • $\begingroup$ i should say more precisely that i am testing an excel like solver: an interpreted nl model that has to be fitted to some data (so one nl model with one or more Xi variables, one single equation to inverse), with bounds constraints only in a first step, then possibly with linear and nl constraints. thanks !! $\endgroup$ – kiriloff Oct 3 '12 at 8:19
  • $\begingroup$ @fonjibe: see the addition to my answer $\endgroup$ – Arnold Neumaier Oct 3 '12 at 9:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.