groupby_agg ingores exceptions

[dchigarev]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Windows 10 1809
• Modin version (modin.__version__): 0.7.3+41.gd617985
• Python version: Python 3.7.5
• Code we can use to reproduce:

if __name__ == "__main__":
    import pandas
    import modin.pandas as pd

    data = {
        "col1": [0, 1, 2, 3],
        "col2": [4, 5, 6, 7],
    }

    md_df, pd_df = (
        pd.DataFrame(data).astype({"col1": "category"}),
        pandas.DataFrame(data).astype({"col1": "category"}),
    )
    by = [1, 2, 1, 2]

    md_grp, pd_grp = md_df.groupby(by=by), pd_df.groupby(by=by)

    print(md_grp.quantile())  # prints an empty DataFrame
    print(pd_grp.quantile())  # throws TypeError: No matching signature found

Describe the problem

Some of groupby aggregation operations in pandas may throw TypeError exception, just like quantile in related code. However modin in that case returns an empty DataFrame, which is different behavior from pandas.

Source code / logs

The problem is probably here: modin/backends/pandas/query_compiler.py: def groupby_agg()

...
grouped_df = df.groupby(by=by, axis=axis, **groupby_args)
try:
    result = agg_func(grouped_df, **agg_args)
except (DataError, TypeError):
    result = pandas.DataFrame(index=grouped_df.size().index)
return result
...

when our agg_func raises TypeError we catch it and try to handle that situation, but it seems that this approach sometimes gives us different from pandas behavior

[devin-petersohn]

Thanks @dchigarev this will be somewhat difficult because sometimes errors are thrown asynchronously.