This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: optimisation
- From: Fabrice Rossi <rossi at ufrmd dot dauphine dot fr>
- To: gsl-discuss at sources dot redhat dot com
- Date: Mon, 06 Jan 2003 17:14:44 +0100
- Subject: Re: optimisation
- Organization: Dauphine University
- References: <3E19A624.29393.E26D75@localhost>
Nicholas Cunliffe wrote:
> Just thought Id cast an opinion. Having multi variable optimization
> routines which REQUIRE user supplied derivetives is just LAME.
> How many real problems are there which are so simple that the
> derivatives can be supplied? None of any scientific or engineering
> interest.
A lot of real world statistical estimation problems can be expressed in such
way that derivatives can be easily calculated. Semi parametric models such as
neural networks (which are estimated thanks to gradient based algorithms) can
be used to model almost any regular (in a very broad sense) real world
regression or discrimination problem. That is of course LAME and has NO
scientific or engineering interest.
My 2 euro-cents: if you need something (such has multivariate optimization
without derivatives) do not start by making a fool of yourself with aggressive
posting, but rather ask politely for help.
Fabrice
Disclaimer: yes, quite a long time ago, I wrote the first sketch of the
useless, lame and uninteresting (on a scientific and engineering point of
view) multivariate optimizer that sucks so much that it requires derivatives.