This is the mail archive of the
gsl-discuss@sourceware.org
mailing list for the GSL project.
Re: let nmsimplex handle non-finite function values properly
- From: Brian Gough <bjg at network-theory dot co dot uk>
- To: "Alxneit-Kamber Ivo" <ivo dot alxneit at psi dot ch>
- Cc: <gsl-discuss at sources dot redhat dot com>
- Date: Wed, 25 Apr 2007 23:02:32 +0100
- Subject: Re: let nmsimplex handle non-finite function values properly
- References: <1177493004.4321.11.camel@pc446.psi.ch>
At Wed, 25 Apr 2007 11:23:24 +0200,
Alxneit-Kamber Ivo wrote:
> the logic behind the patch is that we can generally treat non-finite
> function values (POSINF, NEGINF, NAN) as "no improvment reached" in
> 'gsl_fminimizer_iterate()'. only when we contract the whole simplex
> with respect to its lowest point we have to abort in case we encounter
> non-finite function values. the same is true in 'gsl_fminimizer_set()'.
> this ensures that the simplex never contains non-finite values.
>
> also included are two minor clean ups.
> leak_plug.patch: plugs memory leak if 'gsl_fminimizer_alloc()' fails.
> dim_test.patch: checks for a correct number of parameters in
> 'gsl_fminimizer_alloc()' and for the correct sizes of the various vector
> arguments passed around.
That looks good, I will apply the patches. Thanks!
--
Brian Gough