BUG: ValueError on groupby with categoricals

[smithto1]

• closes BUG: ValueError on groupby with categoricals #34951
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

Within DataFrameGroupBy._cython_agg_blocks, if it is aggregating a one-column DataFrame, it creates a SeriresGroupBy, calls the function on that and takes the returned values. But the SeriesGroupBy also does the missing-categories reindexing. The DataFrameGroupBy ends up with values that contain the missing categories, and an index that does not. When they are passed into a BlockManager it raises a ValueError stating that their lengths don't match.

Solutions is to have _cython_agg_blocks create a SeriesGroupBy with observed=True so it doesn't do any reindexing. The reindexing is left to the calling DataFrameGroupBy

This also explains why error only occurred in DataFrameGroupBy but not SeriesGroupBy.

[pep8speaks]

Hello @smithto1! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2020-07-13 20:18:46 UTC

[smithto1]

@jreback comments addressed and all checks passing. Can you do a re-review?

[jreback]

very nice @smithto1 keep em coming!