Re: lsqnonlin and nonlin_residmin

[Olaf Till]

After reading some Matlab online help, back to your original question:

On Wed, Aug 05, 2015 at 12:25:36PM +0100, Asma Afzal wrote:
> In Matlab, for instance, the following calls generate the same object:
> 
> opts = optim.options.Lsqnonlin  and
> opts = optimoptions('lsqnonlin')
> 
> and according to the help, this is the hierarchy for the class
> optim.options.Lsqnonlin
> 
> optim.options.Lsqnonlin < optim.options.Lsqncommon <
> optim.options.MultiAlgorithm...
> optim.options.Lsqcurvefit < optim.options.Lsqncommon <
> optim.options.MultiAlgorithm...

'optim.options.' only seems to state the package namespace and
subnamespace (even if we wanted, we can't imitate this in current
Octave; and I wasn't aware of this notation). 'Lsqnonlin' is the name
of a class, and therefore the function 'Lsqnonlin' is its constructor
and returns and object of that class. (Your later posts indicate you
thought 'Lsqnonlin' were a method of a class 'options' -- I don't
think that's so.)

As Nir, I think you are free to choose suitable classes and maybe
class-hierarchies, not necessarily adhering to Matlabs
class-hierarchies, and you are free to choose between traditional
classes (documented in Octave, but maybe not so much in current
Matlab) and "classdef" (not yet documented in Octave). But you have to
work out an initial design yourself.

Note that it is difficult to say if Octaves current classdef
implementation is already sufficient for what you need.

Something different: You seem to fight at several frontiers at the
same time. This creates some "task switching overhead", which possibly
could be reduced. Would it not be better first to pursue one topic
until a function is ready to be included into the optim package?
That's only a suggestion.

Olaf

-- 
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

signature.asc
Description: Digital signature