curve fit

[CHAssoc]

Can someone offer an explanation of the different algorithms used in the General Polynomial Fit.vi?

I am fitting a third order to sets of data with anything from 5 points up to about 15 points, and the algorithm choices (SVD, Givens, Cholesky etc. ) all deliver exactly the same solutions. I don't understand under what circumstances they would differ.

Chris

[altenbach]

The results will only differ if your data is near pathological. In this case, some of the algorithms may fail. 😉

[CHAssoc]

Thanks Altenbach. I kind of expected this answer. Our data are well ordered, being the resistance / temperature measurements of a thermistor sensor. I am fitting the Steinhart-Hart equation 1/T = A + B*ln(R) + C*ln(R)^3 to the measurements and get consistent results. The reason for asking the question is that my answers for the coefficients differ from those calculated by the supplier using the same data, and I would have expected ALL mse techniques to give the same output.

[altenbach]

---

@CHAssoc wrote:
The reason for asking the question is that my answers for the coefficients differ from those calculated by the supplier using the same data, and I would have expected ALL mse techniques to give the same output.

---

Well, do you know how the supplier solves for his solution??? 🙂

If you fit 1/T = A + B*ln(R) + C*ln(R)^3 using General LS fit, you'll probably get a slightly different result compare do solving for the direct values with: T(R)  = 1/(A + B*ln(R) + C*ln(R)^3) using e.g. Levenberg Marquardt, because your inversion causes a skewing of the weights of each data point depending on B or T. You are doing a nonlinear transformation! The results will only be identical if the data has no noise at all!

How different are the results?

Do you have an example data set and the result of the supplier?