[Octave-bug-tracker] [bug #46045] [optim] "trust-region-reflective" implementation for lsqnonlin

[Olaf Till]

Follow-up Comment #2, bug #46045 (project octave):

It will always be so that certain optimization problems can be successfully
handled by some algorithms, but not by others. So principally it can be useful
to have several algorithms available.

But the algorithm you mention is, if I remember right, conceptionally similar
to the one in 'nonlin_residmin'. And even if we implement it, we won't get
exactly the same algorithm as Matlab.

So even if you had success with your problem in Matlab, I would not suggest
trying to implement just this algorithm in the hope of a benefit, without
having further information.

If you have a certain kind of problems which systematically are difficult to
solve by 'nonlin_residmin', the right way is probably to provide a test
problem, try to examine the reason for failure, and take an informed decision
on enhancing the current algorithm or implementing a new one.

If you are determined to help in such a way, it's probably best done on the
maintainers mailing list, not here (failures on optimization problems are not
necessarily bugs).

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?46045>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/