pandas 1.4.0 release causes a small error

[fealho]

The new pandas release no longer coerces null-like values in value_counts and mode to a NaN value. This change causes an error in the following test:

RDT/tests/unit/transformers/test_categorical.py

Line 145 in 40d2667

def test__get_intervals_nans(self):

That happens because the test assumes both None and np.nan will be treated the same way. However, the categorical transformer uses value_counts, which no longer has that behaviour.

RDT/rdt/transformers/categorical.py

Line 92 in 40d2667

frequencies = data.value_counts(dropna=False)

This error can be fixed by simply converting the None to a np.nan in the following line:

RDT/tests/unit/transformers/test_categorical.py

Line 159 in 40d2667

data = pd.Series(['foo', np.nan, None, 'foo', 'foo', 'tar'])

Note: This issue only shows up on python 3.8. Neither py3.7 nor py3.9 cause this to happen, and I have no clue why.