0
$\begingroup$

How can a non-linear function be approximated by an appropriate amount of linear functions?

In the picture below, it would be quite easy to draw 10-15 linear functions to describe all data points quite accurately, so once the points are separated into sections, I could easily get very accurate results with linear regression.

However, both, the number of required functions and the length of each part varies. So is there an non-manual algorithm / way to separate the function into sections?

Thank you!

Visualization of data points

$\endgroup$
  • $\begingroup$ Do you mean regression splines? $\endgroup$ – Frans Rodenburg Apr 17 at 9:18
0
$\begingroup$

I'm not sure if your data requires local lines or local constants. But, in general, there is Local Regression, in which you can also fit local degree-k polynomials. When $k=1$, this boils down to local linear regression. It's a non-parametric method like Kernel Regression (you don't have to deal with the end-points of your local curves etc.), which you can also use and specify different kernels functions.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.