BUG: DataFrame([Series with different CategoricalIndexes]) results in non-Categorical columns

[mroeschke]

In [1]: import pandas as pd

In [4]: s1 = pd.Series([1, 2], index=pd.CategoricalIndex(["a", "b"]))

In [5]: s2 = pd.Series([3, 4], index=pd.CategoricalIndex(["a", "c"]))

In [6]: pd.DataFrame([s1, s2]).columns
Out[6]: Index(['a', 'b', 'c'], dtype='object')

I would have expected CategoricalIndex(['a', 'b', 'c'])

[jmarintur]

Hi @mroeschke, can I take this issue? If the answer is yes, what is the expected behavior when you pass a list series with and without categorical indices? Shouldn't we check the whole list? something like this:

all_categorical = all(isinstance(s.index, pandas.CategoricalIndex) for s in data)

and later in the code, change the columns to the right type:

if all_categorical:
    columns = pandas.CategoricalIndex(columns)

[mroeschke]

Sure feel free to work on it (although I am not sure this will be an easy issue to fix). If the indexes are of mixed types the result should be Index[object]

[jmarintur]

I've just created the PR. Please let me know if the changes make sense. As it is, columns will return a Index[object] if the indices are mixed, and a CategoricalIndex[category], if all the series had categorical indices.

I do think it also makes sense to add a test to check this scenario.