BUG: groupby on empty dataframe returns pandas type

[cristianmtr]

System information

DISTRIB_ID=Pop
DISTRIB_RELEASE=22.04
DISTRIB_CODENAME=jammy
DISTRIB_DESCRIPTION="Pop!_OS 22.04 LTS"

modin==0.15.2

Python 3.8.13

    import modin.pandas as md_pd
    import pandas

    df_modin = md_pd.DataFrame(
        [], columns=["smth", "answer_uuid", "item_id", "item_level"]
    )
    pandas_df = df_modin.groupby("answer_uuid")
    assert type(pandas_df) == pandas.core.groupby.generic.DataFrameGroupBy

Describe the problem

Groupby returns of pandas type. Shouldn't it return of modin type?

[mvashishtha]

@cristianmtr thank you for reporting this bug. I can reproduce it at version c9fc326. The bug only applies to empty dataframes. For empty dataframes we default to pandas and end up using the pandas groupby result.

You can work around this bug by

1. doing your groupby on the pandas dataframe
2. doing a groupby operation like sum
3. converting the result to a modin series or dataframe

e.g for your snippet above you can do md_pd.DataFrame(pandas_df.sum())

I don't see an easy fix for this bug. Probably the right thing is to not default to pandas on the groupby call, but instead default to pandas for the operations on the groupby object, e.g. groupby.sum. The default to pandas would have to convert the original dataframe to pandas, then do the operation on that pandas object.