Re: List of most highly used matlab functions.

[Jaroslav Hajek]

On Tue, Jul 29, 2008 at 6:14 PM, John W. Eaton <address@hidden> wrote:
> On 28-Jul-2008, David Bateman wrote:
>
> | While looking randomly around I came across the web page
> |
> | http://www.jmatlab.org/benchmark.html
> |
> | that discusses the most used matlab functions as defined by Nick Higham.
> | It seems that the only matlab functions in this list of 111 functions
> | that are not in the core of Octave are
> |
> | gallery, ezplot, bvp4c, fminbnd, fzero
>
> I guess it depends on who the user(s) are.  I've never needed any of
> these functions, and didn't even know that bvp4c existed.
>
> The gallery function might be nice to have, but I don't see it as
> essential.
>
> I supose ezplot would be helpful for some.
>
> It seems to me that bvp4c is rather specialized.  I imagine there are
> whole swarming masses of users who never need to solve multipoint
> boundary value problems, or even know what they are.
>
> We do see a number of messages looking for fminbnd and fzero though.
>
> | As for fminbnd and fzero, they are both in octave-forge and appear to be
> | reasonably complete so perhaps we should port these to Octave itself.
>
> Unless it is necessary for these functions to use (approximately) the
> same algorithms as are used by the Matlab versions of these functions,
> then I think fminbnd could be written as a wrapper around Octave's sqp
> function, and fzero could be written as a wrapper around fsolve.  The
> same could be done for fmincon and fminunc (I even have minimal
> versions of fmincon and fminunc if someone would like to take them as
> a starting point).
>
> Oh, now I see that fminbnd and fzero are defined to work scalar
> functions.  In that case, I really don't understand why these
> functions even exist.  Why is it necessary to have special funtions
> for scalars?  And why would we want separate (weak?) algorithms for
> these functions instead of just wrapping the general purpose
> optimization functions?
>
> jwe
>

The reason is that special algorithms exist for 1d functions that
perform reportedly better than the general N-d algorithms. I am
already working on improving the optim toolbox.
My aim is to have the 1d functions (fzero, fminbnd) as well as the
equation solvers
(fzero, lsqnonlin - these problems are actually very similar)
implemented in core octave.
I'm not yet sure whether to add a general optimization function like fmins.

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