Re: GSoC 2015: Optimization Package: Non-linear and constrained least squares lsqcurvefit, lsqlin, lsqnonlin

[Olaf Till]

On Wed, Apr 15, 2015 at 03:30:41PM -0700, AsmaA wrote:
> <snip>
>
> Thank-you for highlighting leasqr.
> 
> I have added a note in the GSoC project plan on the wiki for a 'nlinfit'
> implementation around 'leasqr'.

I don't think it's good to mention leasqr in that note. It uses
exactly the same optimization algorithm -- indeed within the same
subfunction -- as nonlin_residmin/nonlin_curvefit. The latter
functions should be prefered over leasqr since they provide a more
reasonable user-interface with additional capabilities and can be
extended to use different backends.

If there is any question which function to wrap, then if it's
preferable to directly use the algorithm backend function(s) instead
of nonlin_residmin/nonlin_curvefit. But directly wrapping the backends
duplicates some effort which has gone into the existing frontends, so
I wouldn't recommend it.

Olaf

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

signature.asc
Description: Digital signature