As my question shows it, I am not a statistician. My problem is that I have too many data points to be used in a nonlinear fit (I have millions of them, automatically acquired). Is there a methodology for sampling the data in order to minimize the loss of information ?
I have a program that I built that can do this. When I had a series of data with minimal change, I was running into multiple infinite calculations, as to say, the program would zero out the denominator. What I had to do in order to get a proper sequence for form fitting was to find an amount of acceptable change. Each set of data will have a different acceptable change rate, but when I did this it forced out all of the infinite calculations and fit the graph properly.
Essentially you are makeing your data points slightly more linear relationally. I had massive series of data where over the course of 500 data points the change would only be maybe 0.3 units of change. When I forced my data to remove minimal changes, I was able to get a nearly perfect fit.
I hope this helps