LabVIEW
curve fitting
mjaugustin wrote:
thank you for remembering me i have doe the same way you said inow i had two arrays one theortical and one from experiment in this i have to find best fit for beta value
Well, your program is about 10x too complicated, but I cannot help much without having also the datafile.
- don't use "delete from array" to get rows or colums, use "index array".
- none of your sequence frames have any purpose, delete them.
- Why do you convert 1D array (size, min index, etc) to 2D arrays, for the only purpose to index out the first element?
- All your code should fit on the size of a postcard, you are doing things way too complicated
So please attach one of your typcal datafiles. Also include the missing subVI and I see what I can do.
What are the adjustable parameters and what inputs should be kept constant?
I see you are using LabVIEW 7.1. This will make things quite a bit more complicated, because all the fitting tools have been significantly improved in LabVIEW 8.0. Do you have access to a newer version?
I wrote some custom tools for earlier versions, but I no longer have access to 7.1, so I won't be easily able to make an example for you.
As a first step, you should calculate the theoretical curve for the exactly same x-values as your data. This means you should use a FOR loop and simply feed it the X-array as an autoindexing input. Now you have corresponding y values for each x and can easly compare y(real) vs. y(fit) in a 1-1 fashion, e.g. to calculate chisquared. There is no need generate a seperate x-range. Since the x-values are evenly spaced (I think) you don't even need an xy-graph, simply use a waveform graph and adjust x0 and dx according to the x-array.
You need at least two parameters, it seems beta adjust the amplitude and e.g. W0 (or the wavelength, they are fully correlated here) adjusts the width.
Attached is a very quick rewrite, downconverted to LabVIEW 7.1. The file IO part did not downconvert correctly, so for simplicity, I added your data as a diagram constants. Also the square function was not available. As predicted, the entire code (and more) can be fit on a postcard, and you don't even need the subVI.
To use, just run it and manually adjust the W0 and Beta with the sliders and see how the function changes. ( I disabled the other controls, because they currently only get read at program start anyway. If course you would want to implement an event structure, so recalculations only occur if needed. Change them in edit mode if needed.
So far it does not do any fitting, simply because I no longer have 7.1 so I cannot do this for you, but a simple 2-paramter fit should be easy to implement. I don't recommend the LabVIEW.7.1 version of the levenberg marquardt fit, it has too many limitations. In this old post is an example that calls the model by reference using a home-made variation. It should be trivial to adapt it to your situation (which is even simpler!). All you need to do is wrap the model in a subVI as shown.
For comparison, I show you also an image of the FP and diagram of the simulator in LabVIEW 8.5.1. As you can see, the file IO is much more functional in newer versions.
