Re: minpack code for least squares?

[Jaroslav Hajek]

On Thu, Feb 14, 2008 at 7:32 PM, Jonathan Stickel <address@hidden> wrote:
> Date: Thu, 14 Feb 2008 06:46:24 -0500
>  From: Ben Abbott <address@hidden>
>  Subject: Re: minpack code for least squares?
>  To: Jaroslav Hajek <address@hidden>
>  Cc: address@hidden
>
>
>  On Feb 14, 2008, at 3:36 AM, Jaroslav Hajek wrote:
>  >> hello,
>  >>
>  >> I've noticed that Octave includes the Powell's hybrid method from
>  >> MINPACK for its fsolve function. Since MINPACK also features
>  >> nonlinear least squares (via trust-region approach), I suggest
>  >> including also that relevant part of MINPACK (lmder,lmdif etc) and
>  >> wrap it in a counterpart to fsolve (with similar interface) to do
>  >> nonlinear least squares. I'd volunteer for the project.
>  >>
>  >> What do you think?
>
>  > I'm all for that. I'd like to have more optimization options in
>  > Octave.
>
>  > Any chance minpack includes a minimax method as well?
>
>  > Ben
>
>
>  There is an existing nonlinear least squares function (and other
>  optimization functions) in the octave-forge package "optim".  I don't
>  know how it compares to the MINPACK routine.
>
>  Jonathan
>
That's true. It does not, however, seem to implement (after quick
scan) a trust-region
concept, so there's at least a notable algorithmic difference (and
I've noticed at least
one more).
Further, MINPACK's LMDER and LMDIF are likely the most widely used
nonlinear least-squares codes in history, so they're sort of "proven
quality".
But anyway, thanks for pointing me to the optim package, I'll probably
try to implement
my idea into this package first and check how it compares.

regards

-- 
RNDr. Jaroslav Hajek
computing expert
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz