BUG: groupby __iter__ on pandas 1.1.x not propagating _metadata on DataFrame subclasses

[ryandvmartin]

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Code Sample, a copy-pastable example

In [1]: import pandas as pd
   ...:
   ...: class SubclassedDataFrame2(pd.DataFrame):
   ...:
   ...:     # temporary properties
   ...:     _internal_names = pd.DataFrame._internal_names + ['internal_cache']
   ...:     _internal_names_set = set(_internal_names)
   ...:
   ...:     # normal properties
   ...:     _metadata = ['added_property']
   ...:
   ...:     @property
   ...:     def _constructor(self):
   ...:         return SubclassedDataFrame2
   ...:

In [2]: df = SubclassedDataFrame2({'A': [1, 2, 3], 'B': [1, 1, 2], 'C': [7, 8, 9]})
   ...: df.added_property = "hello"
   ...: for i, d in df.groupby("B"):
   ...:     assert d.added_property == "hello"
   ...:
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-2-bf64fc505324> in <module>
      2 df.added_property = "hello"
      3 for i, d in df.groupby("B"):
----> 4     assert d.added_property == "hello"
      5

~\RMS-Python\lib\site-packages\pandas\core\generic.py in __getattr__(self, name)
   5133             or name in self._accessors
   5134         ):
-> 5135             return object.__getattribute__(self, name)
   5136         else:
   5137             if self._info_axis._can_hold_identifiers_and_holds_name(name):

AttributeError: 'SubclassedDataFrame2' object has no attribute 'added_property'

Problem description

Under pandas 1.0.x the above is working, on 1.1.x, the _metadata is not propagated to the split frames. I think this is a known issue with groupby and __finalize__, but this specific behavior with __iter__ is a regression from 1.0.x behavior on DataFrame subclasses.

Expected Output

Assert statements above are passing, with metadata propagated to the split dataframe

Output of pd.show_versions()

INSTALLED VERSIONS

commit : db08276
python : 3.7.7.final.0
python-bits : 64
OS : Windows
OS-release : 10
Version : 10.0.19041
machine : AMD64
processor : AMD64 Family 23 Model 113 Stepping 0, AuthenticAMD
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.None

pandas : 1.1.3
numpy : 1.18.1
pytz : 2020.1
dateutil : 2.8.1
pip : 20.0.2
setuptools : 46.1.3.post20200330
Cython : 0.29.17
pytest : 5.4.1
hypothesis : 5.36.1
sphinx : 2.4.4
blosc : None
feather : None
xlsxwriter : 1.2.8
lxml.etree : 4.5.0
html5lib : 1.0.1
pymysql : None
psycopg2 : None
jinja2 : 2.11.2
IPython : 7.13.0
pandas_datareader: None
bs4 : 4.9.0
bottleneck : 1.3.2
fsspec : 0.8.0
fastparquet : None
gcsfs : None
matplotlib : 3.1.3
numexpr : 2.7.1
odfpy : None
openpyxl : 3.0.3
pandas_gbq : None
pyarrow : None
pytables : None
pyxlsb : None
s3fs : None
scipy : 1.4.1
sqlalchemy : 1.3.16
tables : 3.6.1
tabulate : None
xarray : None
xlrd : 1.2.0
xlwt : 1.3.0
numba : 0.49.0

[ryandvmartin]

related comment and existing possible fix: #28283 (comment)

[jorisvandenbossche]

Thanks for the report!
This was not yet fixed by #35688 ?

[ryandvmartin]

I don't think it was fixed there.

I think #37461 is on the right track - I have tracked our specific issue to _chop() in the context of the groupby iterable on 1.0.x in pandas/core/groupby/ops.py:973

in FrameSplitter._chop:

modifying:

    def _chop(self, sdata: DataFrame, slice_obj: slice) -> DataFrame:
        # Fastpath equivalent to:
        # if self.axis == 0:
        #     return sdata.iloc[slice_obj]
        # else:
        #     return sdata.iloc[:, slice_obj]
        mgr = sdata._mgr.get_slice(slice_obj, axis=1 - self.axis)
        return type(sdata)(mgr)

to

    def _chop(self, sdata: DataFrame, slice_obj: slice) -> DataFrame:
        # Fastpath equivalent to:
        # if self.axis == 0:
        #     return sdata.iloc[slice_obj]
        # else:
        #     return sdata.iloc[:, slice_obj]
        mgr = sdata._mgr.get_slice(slice_obj, axis=1 - self.axis)
        return type(sdata)(mgr).__finalize__(sdata)

Propogates the metdata, but as I understand it there are places to and to not use __finalize__ (#28283)

[simonjayhawkins]

I think #37461 is on the right track

the code sample passes with that PR in it's current state. I'll comment there about adding a test for this issue.

However, that PR still needs review and 1.1.4 is scheduled for tomorrow, so moving off 1.1.4 milestone for now.

[simonjayhawkins]

Under pandas 1.0.x the above is working, on 1.1.x, the _metadata is not propagated to the split frames. I think this is a known issue with groupby and __finalize__, but this specific behavior with __iter__ is a regression from 1.0.x behavior on DataFrame subclasses.

first bad commit: [6f065b6] PERF: faster indexing for non-fastpath groupby ops (#34214) cc @jbrockmendel

[Japanuspus]

I think #37461 is on the right track - I have tracked our specific issue to _chop() in the context of the groupby iterable on 1.0.x in pandas/core/groupby/ops.py:973

I hope so -- my own code broke on the 1.1 update as well :)