nonlinear fit and scaling
- To: mathgroup at smc.vnet.net
- Subject: [mg29668] nonlinear fit and scaling
- From: David Lloret <david at cvc.uab.es>
- Date: Tue, 3 Jul 2001 04:40:22 -0400 (EDT)
- Organization: CESCA - Servidor de news de l'Anella Cientifica
- Sender: owner-wri-mathgroup at wolfram.com
Dear all,
I am using the nonlinearfit package with mathematica 4.0. It works
fine with the data, but the resulting condition number is very high,
900. Many papers suggest on scaling the parameters which involves
modifying the algorithm to work with JH instead of J alone. Since that
looks quite a job, I modified the equations describing the model to deal
with the variables substituted by their scaled versions. That dropped
the condition number very well, down to 9.
The only problem is that the two solutions, normal and scaled, are
actually the same, so I am not sure whether someone is lying.
I appreciate any help.
David
david at cvc.uab.es