Hi everyone,
It has been a while since I last posted, I hope you are all doing well!
I have some experimental data that looks like a cubic function with three approximately linear part; the head and the tail has a steep slope and the middle part always has the smallest slope. The linear equation to the middle part is what I am interested in. The problem is: in some data the may only be the first two part, or the first two part look very similar and the third part is missing.
I already wrote a program in MC15 to find the slope of the middle part and the following is my strategy:
divide the curve into say 6 to 7 or more sections and use polyfitc 1st degree to fit each section
find the section with a minimum slope
find all other section within 1 stdev of the minimum slope, which is given by polyfitc, combine all these section and do another polyfitc on the combined section.
Since the curve looks like continueous and the three region have different slope, so I did not constraint only adjecent section can be combined.
But once in a while, if the three parts look indistinguishable (or the last part is missing), I have a linear fit that is parallel to my data instead of on it...
Then I make a second condition: if the linear fit using all data points can fit those combined section better than just using a portion of the data, this mean my entire data set looks very linear, then I use a linear line that fit all data available.
I would like to learn what is other strategy to do what I want. I tried to identify the transition region between different quasi-linear part, but I do not know how to do that. Now my program needs about 2 minutes to go through 95 data set with 1700 data point each. Is there a more efficient way ?
Thanks
Henry
