2D fit: one question to Mr altenbach

[gnappo]

Dear altenbach,

I started to look at all your posts about fit 2d (the oldest is in 2004) and I downloaded all your files 2Dquadratic planefit, 2Dscatterfit2,3Dpolyfit, XuFit_004. Also I read this https://decibel.ni.com/content/docs/DOC-30394

Ok...I am completely lost. 

I would like to ask you from which of your files I should start. I would like to fit with the function z=a+bx+cx**2 + dy+ey**2+fxy (second order polynamial) a set of experimental data (x,y,z are experimental data, all of them are 1D vectors with the same dimension, furthermore they are very large vectors about 100000 entries each). 

I did not understand:

1) what is the difference between gridded data and scattered data. 

2) if you suggest the polynomial fit (building the H matrix) or the Lev Mar fit. 

I also ask you if there are more recent vi's to start from, because the old ones are very very hard to understand. 

Thank you in advance

[altenbach · Accepted solution]

Use the newer one. Some of the old examples use homebuilt, pre LabVIEW 8.0 subroutines.  The fitting tools received a major overhaul with LabVIEW 8.0 and should be used out of the box.

---

@gnappo wrote:

I would like to ask you from which of your files I should start. I would like to fit with the function z=a+bx+cx**2 + dy+ey**2+fxy (second order polynamial) a set of experimental data (x,y,z are experimental data, all of them are 1D vectors with the same dimension, furthermore they are very large vectors about 100000 entries each). 

---

Since your problem is linear in the coefficients, you should use the general linear fit, constructing the H matrix from the terms. No need for iterative procedures such a lev-mar. Create your H matrix with the various terms for each xy pair.

 Gridded data mans that the x and y values form a rectangular grid with no missing values, i.e. the 2D data can be represented as a simple 2D array. Scattered data means that there is a list of (x,y,z) points and x and/or y are not evenly spaced. You can run this in the two modes to immediately see the difference.

[gnappo]

Thank you altenbach. I going to look at. In case of problems/needs of explanations I will ask you.

[gnappo]

May I ask you 2 questions?

1) z=f(x,y). Do you know how should I calculate the statistical error to be associated to the best fit output vector?

2) I would like to use as weight the statistical errors not only for the experimental z values, but also for x and y. Do you know if there is a way to do that?

Thank you