This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
simplex minimization
- From: Tim F <fenn at agora dot dhs dot org>
- To: gsl-discuss at sources dot redhat dot com
- Date: Mon, 30 Dec 2002 16:50:15 -0500 (EST)
- Subject: simplex minimization
I've been using gsl (1.3) to minimize a seven parameter function against
some sample data I have, and it seems to give me good results based on fdf
minimization (any of the fr, pr or bfgs algorithms yield similar
numbers):
iteration: 6
params:
-0.00131 gradient: -5610116.24709
0.02920 gradient: -8297325.73283
-0.05581 gradient: -184748.86988
-0.00000 gradient: 0.00000
0.00000 gradient: 0.00000
0.00000 gradient: 0.00000
0.20521 gradient: 48147814.97896
f(x): 1491335.000000 tot. gradient: 34055624.129164
these numbers are around what I expect (based on comparison with other
programs that carry out a similar form of minimization) even though the
gradients may seem large (the 4th-6th parameters are forced to zero in the
example I'm testing). Out of curiousity, I tried to compare these values
with those determined by the simplex algorithm. However, the minimization
always stops and claims to have succeeded before performing any iterations
(with a size stopping point of 0.01):
iteration: 1
params:
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.20521
f(x): 1566284.000000 tot. size: 0.000000000000
and no matter the starting values, this always seems the case - i.e. the
simplex size is always 0.00. Might it have something to do with forcing
some of the parameters in the minimization to zero? If code would be
helpful in nailing this problem down, let me know.
Regards,
Tim F