Review parameters values of Savitzky-Golay-filters

Description

The Savitzky-Golay-Filter scipy.signal.savgol_filter is a useful tool to smooth noisy data. It often works better then for example a rolling mean.

Problem

The arguments for the savgol_filter are window_length and polyorder which were mostly chosen by eye to make the smoothened data fit best. There was no strict rule on how to chose the parameters or any validation about the drawbacks of certain combinations of parameters.

To do

• where are drawbacks from certain parameter settings?
• is there Literature describing problems with savgol_filter?
• is there a solid way to chose the parameters (fitting for least squares ot to minimize outliers?)