BUG: SeriesGroupby with sort=False behaves inconsistently with .quantiles(...).unstack()

[ArneLeijon]

Pandas version checks

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

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

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd

ind = pd.MultiIndex.from_tuples([(0, 'a', 'B'),
                                 (0, 'a', 'A'),
                                 (0, 'b', 'B'),
                                 (0, 'b', 'A'),
                                 (1, 'a', 'B'),
                                 (1, 'a', 'A'),
                                 (1, 'b', 'B'),
                                 (1, 'b', 'A'),
                                 ], names=['sample', 'cat0', 'cat1'])
ds = pd.Series(range(8), index=ind)
print(ds)

s01 = ds.groupby(level=['cat0', 'cat1'], sort=False).quantile([0.2, 0.8])
print('\n(cat0, cat1) Quantiles As Series: Result as expected:\n', s01)
d01 = s01.unstack()
print('\nQuantiles Unstacked: Result as expected:\n', d01)

s1 = ds.groupby(level='cat1', sort=False).quantile([0.2, 0.8])
print('\ncat1 Quantiles As Series: Result as expected:\n', s1)
d1 = s1.unstack()
print('\nQuantiles Unstacked: NOT as expected; cat1 index SORTED alphabetically:\n', d1)

print('\nQuantiles Unstacked, by iterrows:')
for (cat, q) in d1.iterrows():
    print(f'{cat}: {q.values}')

Issue Description

I need to preserve the pre-defined order of MultiIndex labels in both 'cat0' and 'cat1', when calculating quantiles. The problem is that the result is as expected when the index 'cat1' in the example variable d01 is the second of two groupby levels, but does NOT give a similar expected result when 'cat1' is the single groupby level, as in the variable d1 in the example.

Expected Behavior

In my code example, I expected the unstacked DataFrame d1 to retain the index order ['B', 'A'], same as in the Series variable s1, and in analogy to the similarly unstack-ed DataFrame variable d01.

Installed Versions

INSTALLED VERSIONS

commit : 37ea63d
python : 3.10.7.final.0
python-bits : 64
OS : Darwin
OS-release : 22.4.0
Version : Darwin Kernel Version 22.4.0: Mon Mar 6 21:01:02 PST 2023; root:xnu-8796.101.5~3/RELEASE_ARM64_T8112
machine : arm64
processor : arm
byteorder : little
LC_ALL : None
LANG : None
LOCALE : None.UTF-8

pandas : 2.0.1
numpy : 1.23.3
pytz : 2022.2.1
dateutil : 2.8.2
setuptools : 63.2.0
pip : 23.1.2
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.1.2
IPython : None
pandas_datareader: None
bs4 : None
bottleneck : None
brotli : None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : 3.6.2
numba : None
numexpr : None
odfpy : None
openpyxl : 3.0.10
pandas_gbq : None
pyarrow : None
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : 1.9.1
snappy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
zstandard : None
tzdata : 2023.3
qtpy : None
pyqt5 : None

[rhshadrach]

Thanks for the report. Can include the expected and current output - this can change overtime and it is helpful to people who look into the issue in the future.

unstack is using .index.levels to determine the result, and this appears inconsistent in the result of groupby:

ind = pd.MultiIndex.from_tuples([(0, 'a', 'B'),
                                 (0, 'a', 'A'),
                                 (0, 'b', 'B'),
                                 (0, 'b', 'A'),
                                 (1, 'a', 'B'),
                                 (1, 'a', 'A'),
                                 (1, 'b', 'B'),
                                 (1, 'b', 'A'),
                                 ], names=['sample', 'cat0', 'cat1'])
ds = pd.Series(range(8), index=ind)

s01 = ds.groupby(level=['cat0', 'cat1'], sort=False).quantile([0.2, 0.8])
print(s01.index.levels)
# [['a', 'b'], ['B', 'A'], [0.2, 0.8]]

s1 = ds.groupby(level='cat1', sort=False).quantile([0.2, 0.8])
print(s1.index.levels)
# [['A', 'B'], [0.2, 0.8]]

Further investigations and PRs to fix are welcome!

[Charlie-XIAO]

pandas/pandas/core/groupby/groupby.py

Line 4320 in 4f14b45

mi = MultiIndex.from_product([idx, qs])

Previously if idx is not MultiIndex, we will call MultiIndex.from_product, which calls factorize_from_iterables resulting in levels to be sorted. I think we can do something similar for Index as for MultiIndex, that is, setting levels and codes manually. For instance,

def _insert_quantile_level(idx: Index, qs: npt.NDArray[np.float64]) -> MultiIndex:
    nqs = len(qs)

    if idx._is_multi:
        idx = cast(MultiIndex, idx)
        lev_codes, lev = Index(qs).factorize()
        levels = list(idx.levels) + [lev]
        codes = [np.repeat(x, nqs) for x in idx.codes] + [np.tile(lev_codes, len(idx))]
        mi = MultiIndex(levels=levels, codes=codes, names=idx.names + [None])
    else:  # modified
        lev_codes, lev = Index(qs).factorize()
        levels = [idx, lev]
        codes = [np.repeat(range(len(idx)), nqs), np.tile(lev_codes, len(idx))]
        mi = MultiIndex(levels=levels, codes=codes, names=[idx.name, None])
    return mi

I will open a PR that implements this solution.