BUG: tz-aware datetime with column-wise comparisions failing with np.minmum/maximum

[adbull]

Code Sample, a copy-pastable example if possible

>>> import numpy as np
>>> import pandas as pd

>>> naive = pd.to_datetime(['now'])
>>> utc = naive.tz_localize('UTC')

>>> np.minimum(naive, naive)
>>> np.minimum(utc, utc)
>>> np.minimum(pd.Series(naive), pd.Series(naive))
>>> np.minimum(pd.Series(utc), pd.Series(utc))

  File "bug.py", line 10, in <module>
    np.minimum(pd.Series(utc), pd.Series(utc))
  File "~/anaconda3/lib/python3.5/site-packages/pandas/core/series.py", line 498, in __array_prepare__
    op=context[0].__name__))
TypeError: Series with dtype datetime64[ns, UTC] cannot perform the numpy op minimum

Problem description

When a tz-aware datetime is placed in a Series, the numpy operations fmin/fmax/minimum/maximum throw an error, even though these operations work fine on a tz-aware DatetimeIndex.

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.5.2.final.0
python-bits: 64
OS: Linux
OS-release: 4.9.8-100.fc24.x86_64
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: C
LANG: C
LOCALE: None.None

pandas: 0.19.2
nose: 1.3.7
pip: 9.0.1
setuptools: 27.2.0
Cython: 0.25.2
numpy: 1.11.3
scipy: 0.18.1
statsmodels: 0.8.0
xarray: 0.9.1
IPython: 4.2.0
sphinx: 1.5.1
patsy: 0.4.1
dateutil: 2.6.0
pytz: 2016.10
blosc: None
bottleneck: 1.2.0
tables: 3.3.0
numexpr: 2.6.2
matplotlib: 2.0.0
openpyxl: 2.4.1
xlrd: 1.0.0
xlwt: 1.2.0
xlsxwriter: 0.9.6
lxml: 3.7.2
bs4: 4.5.3
html5lib: 0.999
httplib2: None
apiclient: None
sqlalchemy: 1.1.5
pymysql: None
psycopg2: None
jinja2: 2.9.4
boto: 2.45.0
pandas_datareader: None

[jreback]

this is related to this: #15553

but that said, there is only so much that can be done when passing things directly to numpy arrays like this (its actually not the passing, but returning, some numpy functions are friendly and some are not).

I suppose this could be made to work, below is a much more idiomatic way to do this. See the 2nd part for the actual issue.

naive

In [61]: s = Series(pd.date_range('20130101', periods=4))

In [62]: s2 = s[::-1]

In [63]: df = DataFrame({'A':s, 'B':s2.values})

In [64]: df
Out[64]: 
           A          B
0 2013-01-01 2013-01-04
1 2013-01-02 2013-01-03
2 2013-01-03 2013-01-02
3 2013-01-04 2013-01-01

In [65]: df.max(axis=1)
Out[65]: 
0   2013-01-04
1   2013-01-03
2   2013-01-03
3   2013-01-04
dtype: datetime64[ns]

doesn't raise, but incorrect results for tz-aware

In [67]: s = Series(pd.date_range('20130101', periods=4, tz='US/Eastern'))

In [68]: s2 = s[::-1]

In [69]: df = DataFrame({'A':s, 'B':s2.values})

In [70]: df
Out[70]: 
                          A                   B
0 2013-01-01 00:00:00-05:00 2013-01-04 05:00:00
1 2013-01-02 00:00:00-05:00 2013-01-03 05:00:00
2 2013-01-03 00:00:00-05:00 2013-01-02 05:00:00
3 2013-01-04 00:00:00-05:00 2013-01-01 05:00:00

In [71]: df.max(axis=1)
Out[71]: 
0   NaN
1   NaN
2   NaN
3   NaN
dtype: float64

[jreback]

and @adbull

When a tz-aware datetime is placed in a Series, the numpy operations fmin/fmax/minimum/maximum throw an error, even though these operations work fine on a tz-aware DatetimeIndex.

this is not true at all, they naively look like they are working, but because of the same issue above (numpy has no clue about timezones, and forget about missing values), these are completely wrong (they are tz shifted incorrectly)

In [78]: i = pd.DatetimeIndex(df.A)
Out[78]: 
0   2013-01-01 00:00:00-05:00
1   2013-01-02 00:00:00-05:00
2   2013-01-03 00:00:00-05:00
3   2013-01-04 00:00:00-05:00
Name: A, dtype: datetime64[ns, US/Eastern]

In [79]: np.maximum(i, i)
Out[79]: DatetimeIndex(['2013-01-01 05:00:00-05:00', '2013-01-02 05:00:00-05:00', '2013-01-03 05:00:00-05:00', '2013-01-04 05:00:00-05:00'], dtype='datetime64[ns, US/Eastern]', name='A', freq='D')

[mroeschke]

This work on master now. Could use a test as always.