This is the mail archive of the
gsl-discuss@sourceware.org
mailing list for the GSL project.
Re: alternate multimin api
- From: "Brian Gough" <bjg at network-theory dot co dot uk>
- To: James Bergstra <james dot bergstra at umontreal dot ca>
- Cc: gsl-discuss at sources dot redhat dot com
- Date: Sat, 27 May 2006 16:24:35 +0100
- Subject: Re: alternate multimin api
- References: <20060516005235.GA12658@aphex.iro.umontreal.ca>
James Bergstra writes:
> while (! m->giveup )
> {
> minalgo_suggest(m);
> if (m->want_y)
> {
> ... calculate m->test.y from m->test.x;
> }
This type of API is called 'reverse communication'. It is more
flexible but traditionally hasn't been used as much due to the user
having to provide more information (we have't used it in GSL). There
are a number of ACM TOMS papers which discuss it.
Having access to a reverse communication interface can be useful, as
there are some types of problems which can only be solved that way.
I'm looking at the linesearch in multimin at the moment to improve it
so I'll keep this in mind.
--
Brian Gough