LabVIEW

cancel
Showing results for 
Search instead for 
Did you mean: 

Basic Spectra in Levenberg Marquardt

Hi Christian!

This is really the way to go! and your programming is very transparent and challenges me to change my own 'style'. Indeed only a very small time investment was needed to alter the fitfunction - I put a formula box in the sub.vi instead of the operating nodes. What I also like very much is the automatic update panel that changes the curve as one is trying to first-guess. Very nice & useful.

I have not yet been able to fit with the more complex fit model, but it is already clear to me that too many parameters and to bad (very bad) first guesses make the LM algirithm go bananas. But still I have to convert some proper data with the troublesome scattering propeties that I want to fit before I can state that I'm saved, but this is likely to be the case.

Thanks a lot!!

Olav
0 Kudos
Message 11 of 14
(903 Views)
It is not clear how stable the fit for the more complex model will be. It will most likely require very good initial guesses for abcd (if these are fitting paramters). You can easily see that there will be multiple minima, for example once you find a best fit, there will be an equally good fit if you switch the values for of a-c and b-d.
 
As a sidenote, LabVIEW 8.0 received a significant overhaul of everything related to fitting and implements some of the ideas shown (e.g. calling the model via reference node for easy modularity). In addition, you have much more detailed control over termination conditions, etc. 🙂
 
0 Kudos
Message 12 of 14
(896 Views)
Hi,

About the minima: It would be nice to be able to set boundaries for a,b,c and d, which are fitparameters indeed. We just got Labview 8 in a few weeks ago so I should check that version and try to constrain abcd to reasonable values in the new L8 regime instead of using the good old arctan trick to limit the effect of a parameter in a certain region to compress its domain.

thanks again

Olav
0 Kudos
Message 13 of 14
(894 Views)

Constraining the parameters is not supported directly, but yes, you can reparametrize the fitting parameters to get the desired effect.

Several possibilities are described in this link. 🙂

0 Kudos
Message 14 of 14
(886 Views)