scikit-lego
scikit-lego copied to clipboard
Un-pandas `GroupedTimeSeriesSplit`
Description
In #604 there was a tiny hint at this 😂
In this PR I tried to remove the use of pandas from GroupedTimeSeriesSplit
by moving on numpy backend. The only method that still needs a "dataframe"-like object is .summary()
(which it's even a nice to have).
In a way this is an alternative way to achieve #597 when a dataframe is just a nice abstraction but not mandatory. The only function that has to work to proceed forward and run .split()
method is indexable
(from sklearn.utils.validation).