search

laszukdawid / PyEMD

Python implementation of Empirical Mode Decompoisition (EMD) method
https://pyemd.readthedocs.io/
Apache License 2.0
857 stars 222 forks source link

Working with missing data #139

Closed devillegaso closed 6 months ago

devillegaso commented 1 year ago

I am new to EMD and I was wondering if the algorithm would have any problem while working with a time series containing missing data. Until now, I'm removing nans before performing an EMD analysis, but by doing so I'm giving as input a T array of non periodic positions in time. Am I introducing an error? In that case, what would be the correct procedure?

Thanks a lot.

laszukdawid commented 1 year ago

Hey, unfortunately missing data is a problem in general. EMD is an empirical method and assumes uniform sampling so you something needs to be present in your time series. There are likely some best practises and suggestions whether and how to interpolate/impute missing data. However, I don't think that's reasonable. EMD is very sensitive to tiny deflections. Maybe other ensemble methods, like EEMD, would be a bit more robust but I don't have knowledge about it.