Nonlinear curve-fitting difficult functions
- To: mathgroup at smc.vnet.net
- Subject: [mg6579] Nonlinear curve-fitting difficult functions
- From: Jeff Hammock <hammockj at email.uah.edu>
- Date: Fri, 4 Apr 1997 02:11:31 -0500 (EST)
- Organization: The University of Alabama in Huntsville
- Sender: owner-wri-mathgroup at wolfram.com
Hi,
I frequently use the MMA package Statistics`NonlinearFit` to perform
fits to data using equations such as:
f[x_,y_,]:= x Sin[y]
The package uses the Levenberg-Marquardt method to itteratively find the
best-fit solution. This works fine if I have a nice closed-form
solution with well-behaved derivatives.
I WANT to be able to use model functions like:
f[x_,y_]:= NIntegrate[g[x,i],{i,0,y}]
or ANY other function that I can generate that does not have a
closed-form solution.
Does anyone know a trick to force the existing MMA package into using
numerically calculated derivatives in its optimization?
Or, has anyone written a nonlinear fitting routine that handles this
type of functions?
Thanks in advance for any insights....
Matt
***************************************
Matthew Smith, Ph.D.
Research Scientist
The University of Alabama in Huntsville
smithmh at email.uah.edu
***************************************