Levenberg Marquardt Issue

[TheInfamousYano]

Hi,

I'm using a 1D nonlinear Lev-Mar fitting routine, and I find that my program does not fit accurately unless I upscale my y-data values (for example from .16 up to 16). I've read other posts related to this topic and one of them said to change the standard deviation, but that does not seem to affect the fit at all. The Lev-Mar routine just seems to converge right away instead of going through any iterations.

I've attached the 2 relevant VIs and a set of sample data. "TestVApprox6" sends the data to "VoigtApprox6" which then does the nonlinear fit according to the Fitting Equation on the front panel (I modified the formula node inside the Lev-Mar VI with that equation). There are subVIs which perform other calculations but right now I am just concerned with the fitting.

Any ideas?
Thanks for your time
-Stefano

[altenbach]

Please save your toplevel VI as application distribution, so it includes all required subVIs.

[altenbach]

You seem to use a modified Lev-Mar routine that you have not attached, so it is not clear what is going on. My best guess from the limited data is the following problem:

You wire an array of "1s" to the Stdev input, yet your Y data is in the range of 1e-4. This means that the fit is considered converged if it is within a stdev of 1. This is not good enough! Try replacing the diagram constant (near upper left corner in diagram) which is currenty "1" with e.g. "1E-6" and see if things improve. 🙂

[altenbach]

I cannot open your llb... 😞

Message Edited by altenbach on 01-03-2006 02:21 PM

[TheInfamousYano]

Maybe this will work?

[altenbach]

Sorry, try again.

My mistake, please save as Development Distribution and make sure NOT to include VI.lib files.

[TheInfamousYano]

ok hopefully this one will work