Type issue in empty groupby from DataFrame with categorical

[xflr6]

In a DataFrame without a categorical, the following comparisons work as expected:

df = pd.DataFrame({'id': [None] * 3, 'spam': [None] * 3})
df['spam'] == 'spam'
df.groupby('id').first()['spam'] == 'spam'

However, when a column is Categorical, a groupby on the all-null column behaves unexpected:

df['spam'] = df['spam'].astype('category')
df['spam'] == 'spam'  # works as expected
df.groupby('id').first()['spam'] == 'spam'  # raises TypeError: invalid type comparison

Looks like the groupby converts all types in the group to float64:

>>> df.groupby('id').first().info()
<class 'pandas.core.frame.DataFrame'>
Int64Index: 0 entries
Data columns (total 1 columns):
spam    0 non-null float64
dtypes: float64(1)
memory usage: 0.0 bytes

[jreback]

will mark this as a bug; I suspect that the resulting series is simply not of categorical type (which would make the comparison valid).

pull-requests welcome!

[jorisvandenbossche]

Still failing with 0.23.2

[mroeschke]

Looks to work on master. Could use a test.

In [11]: df.groupby('id').first()['spam'] == 'spam'
Out[11]: Series([], Name: spam, dtype: bool)

In [12]: pd.__version__
Out[12]: '0.26.0.dev0+627.gef77b5700'