groupby/apply behaves different if a column is timezone-aware or not

[jordansamuels]

Code Sample, a copy-pastable example if possible

import pandas as pd

dates = ['2001-01-01'] * 2 + ['2001-01-02'] * 2 + ['2001-01-03'] * 2
index_no_tz = pd.DatetimeIndex(dates)
index_tz = pd.DatetimeIndex(dates, tz='UTC')
df1 = pd.DataFrame({'x': list(range(2))*3, 'y': range(6), 't': index_no_tz})
df2 = pd.DataFrame({'x': list(range(2))*3, 'y': range(6), 't': index_tz})

result1 = df1.groupby('x', group_keys=False).apply(lambda df: df[['x','y']].copy())
result2 = df2.groupby('x', group_keys=False).apply(lambda df: df[['x','y']].copy())

try:
    pd.testing.assert_frame_equal(result1, result2)
except:
    print("does not work")

Problem description

It appears that groupby/apply appears to sort differently for the two dataframes in the example; the only difference between them is the timezone-awareness of the t column.

(Note: I posted this in the pydata/pandas gitter, and @jorisvandenbossche suggested I open an issue here for it.)

Expected Output

We would expect the assertion to pass, i.e. the does not work message would not appear.

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.6.8.final.0
python-bits: 64
OS: Linux
OS-release: 4.18.0-24-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8

pandas: 0.24.2
pytest: 4.6.3
pip: 19.1.1
setuptools: 41.0.1
Cython: None
numpy: 1.16.4
scipy: 1.3.0
pyarrow: 0.13.0
xarray: None
IPython: 7.5.0
sphinx: None
patsy: 0.5.1
dateutil: 2.8.0
pytz: 2019.1
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: 3.1.0
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml.etree: 4.3.4
bs4: None
html5lib: None
sqlalchemy: 1.3.5
pymysql: None
psycopg2: None
jinja2: 2.10.1
s3fs: 0.2.1
fastparquet: 0.3.1
pandas_gbq: None
pandas_datareader: None
gcsfs: None

[mroeschke]

Looks specific to when copy is called:

In [8]: df1.groupby('x', group_keys=False).apply(lambda df: df[['x','y']].copy())
Out[8]:
   x  y
0  0  0
2  0  2
4  0  4
1  1  1
3  1  3
5  1  5

In [9]: df2.groupby('x', group_keys=False).apply(lambda df: df[['x','y']].copy())
Out[9]:
   x  y
0  0  0
1  1  1
2  0  2
3  1  3
4  0  4
5  1  5

In [10]: df1.groupby('x', group_keys=False).apply(lambda df: df[['x','y']])
Out[10]:
   x  y
0  0  0
1  1  1
2  0  2
3  1  3
4  0  4
5  1  5

In [11]: df2.groupby('x', group_keys=False).apply(lambda df: df[['x','y']])
Out[11]:
   x  y
0  0  0
1  1  1
2  0  2
3  1  3
4  0  4
5  1  5

Might be shades of #14927

[jordansamuels]

@mroeschke agree, seems related. There is also the observation that group_keys seems to be respected only when the t column doesn't have a timezone:

In [2]: df1.groupby('x', group_keys=True).apply(lambda df: df.copy())
Out[2]:
     x  y          t
x
0 0  0  0 2001-01-01
  2  0  2 2001-01-02
  4  0  4 2001-01-03
1 1  1  1 2001-01-01
  3  1  3 2001-01-02
  5  1  5 2001-01-03

In [3]: df2.groupby('x', group_keys=True).apply(lambda df: df.copy())
Out[3]:
   x  y                         t
0  0  0 2001-01-01 00:00:00+00:00
1  1  1 2001-01-01 00:00:00+00:00
2  0  2 2001-01-02 00:00:00+00:00
3  1  3 2001-01-02 00:00:00+00:00
4  0  4 2001-01-03 00:00:00+00:00
5  1  5 2001-01-03 00:00:00+00:00

Should I mark that as a separate issue?

I'm mentioning this assuming that calling copy is indeed the "right" way. #19175 seems to imply it is.

[mroeschke]

Usually there isn't a need to call copy in apply. You can leave it here in this issue.

[mroeschke]

This looks fixed on master. Could use a test

In [155]: import pandas as pd
     ...:
     ...: dates = ['2001-01-01'] * 2 + ['2001-01-02'] * 2 + ['2001-01-03'] * 2
     ...: index_no_tz = pd.DatetimeIndex(dates)
     ...: index_tz = pd.DatetimeIndex(dates, tz='UTC')
     ...: df1 = pd.DataFrame({'x': list(range(2))*3, 'y': range(6), 't': index_no_tz})
     ...: df2 = pd.DataFrame({'x': list(range(2))*3, 'y': range(6), 't': index_tz})
     ...:
     ...: result1 = df1.groupby('x', group_keys=False).apply(lambda df: df[['x','y']].copy())
     ...: result2 = df2.groupby('x', group_keys=False).apply(lambda df: df[['x','y']].copy())

In [156]: result1
Out[156]:
   x  y
0  0  0
1  1  1
2  0  2
3  1  3
4  0  4
5  1  5

In [157]: result2
Out[157]:
   x  y
0  0  0
1  1  1
2  0  2
3  1  3
4  0  4
5  1  5