Exception when grouping on a modified category

[phil65]

Code Sample, a copy-pastable example if possible

import pandas as pd

ds = pd.DataFrame({"a": [1, 2, 3], "b": ["a", "b", "c"]}) 
ds["b"] = ds["b"].astype("category")
ds['b'].cat.remove_categories(['a'], inplace=True)

# works
ds.groupby("b").groups

# gives ValueError: Categorial categories cannot be null 
ds.groupby("b", observed=True).groups

Problem description

I wonder if that exception should be raised there?
Using 0.24 RC1

Thanks to all working on pandas!

[jorisvandenbossche]

@phil65 Thanks for the report!
Can you provide a more easily reproducible example? Ideally you would construct a small test dataframe with code instead of relying on an external package (eg for adding this test case in the test suite, we don't want to use an external package)

[phil65]

Sure. Done.

[jorisvandenbossche]

Thanks!
I can confirm the error on master (existed on 0.23.4 already as well). It also seems that things like a reduction with ds.groupby("b", observed=True).sum() do work (not error). Although they include a NaN in the grouped keys, which doesn't seem to be correct behaviour (and certainly related to the error)

[jorisvandenbossche]

This is certainly related: #21151