Filtering EDA peaks in long signals

[manasa-kalanadhabhatta]

Describe the solution you'd like When using the kim2004 (and possibly neurokit) method in eda_peaks(), the min amplitude threshold of 10% for filtering SCR peaks is based on the maximum peak amplitude in the entire signal. This makes sense for a signal that is a few seconds long (like the 50s windows in Kim et al.). However, for arbitrarily long signals (> few minutes), this filters out valid peaks if the peak amplitudes change with time or if there are signal artifacts that are recognized as the largest peak.

How could we do it? Filtering peaks should be done based on local maxima, possibly using a sliding window. Or, it could be useful to update the documentation to recommend applying eda_peaks() on smaller chunks when dealing with longer signals.