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

[Olaf Till]

On Sat, Feb 28, 2015 at 06:58:45PM +0100, Olaf Till wrote:
> On Fri, Feb 27, 2015 at 01:00:50PM -0800, AsmaA wrote:
> > Hi,
> > 
> > Thank-you very much for the discussion on the project.
> > 
> > I am following up on details in the background.
> > 
> > 
> > Olaf Till-2 wrote
> > > If that'll be so, the project could indeed be to implement these
> > > functions as wrappers:
> > > 
> > > 1. 'lsqnonlin' wraps 'nonlin_residmin',
> > > 
> > > 2. 'lsqcurvefit' wraps either 'nonlin_curvefit', 'nonlin_residmin', or
> > >    even 'lsqnonlin',
> > > 
> > > 3. 'fmincon' wraps 'nonlin_min',
> > > 
> > > and maybe
> > > 
> > > 4. 'lsqlin' wraps ... (?).
> > > 
> > > Olaf
> > 
> > 'lsqlin' should wrap around Octave's quadratic program 'qp' [2].
> > 
> > 'lsqlin' problem is a special case of 'qp'. <snip>
> 
> Since 'lsqlin' seems to do linear programming, you could indeed see
> this as a special case of quadratic programming. But shouldn't a
> dedicated linear programming algorithm be better?
> 
> I havn't studied linear programming algorithms yet, nor the available
> functions in Octave. But there are e.g. 'linprog' in the optim package
> and 'glpk' in core Octave. Maybe there are even some algorithms in the
> linear algebra package applicable to some large problems (have not
> checked).

Sorry, I probably didn't really think about my last post, it contains
some nonsense, seemingly disregarded the sum of squares ...

Correction:

I didn't look at your reference [1], but indeed, if the problem for
lsqlin is "minimize sumsq (Ax - d), with constraints", you could
probably write it as "minimize .5*x.'(2*A.'A)x + (-2d.'A)x with
constraints" and call 'qp' with H = 2A.'A and q = -2d.'A.

So I'd say you're right.

Olaf

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

signature.asc
Description: Digital signature