DataFrame.any(axis={0, 1}) returns inconsistent values for non-zero timedelta

[nmusolino]

Steps to reproduce

In [1]: import pandas

In [2]: df = pandas.DataFrame({'a': pandas.Series([0, 0]),
   ...:                        't': pandas.Series([pandas.to_timedelta(0, 's'), pandas.to_timedelta(1, 'ms')])})

In [3]: df
Out[3]:
   a               t
0  0        00:00:00
1  0 00:00:00.001000

In [4]: df.dtypes
Out[4]:
a              int64
t    timedelta64[ns]
dtype: object

In [5]: df.any(axis=0)
Out[5]:
a    False
t     True
dtype: bool

In [6]: df.any(axis=1)
Out[6]:
0    False
1    False
dtype: bool

Problem description

pandas.DataFrame.any() returns "whether any element is True over the selected axis".

In the example above, the results of df.any(axis=ax) is clearly inconsistent for the two axis directions (cells [5] and [6]).

If an element in column "t" leads to a result of True in df.any(axis=0), then the same element should lead to a result of True in df.any(axis=1).

Expected output

As a user, I expected DataFrame.any() to return True for rows or columns with a non-zero timedelta value. That is, I expected a timedelta column to have the same semantics as integers: zero values are treated as False, while other values are treated as True.

In other words, this is the output I expected:

In [6]: df.any(axis=1)
Out[6]:
0    False
1    True
dtype: bool

Output of pd.show_versions()

In [7]: pandas.show_versions()

INSTALLED VERSIONS
------------------
commit: None
python: 3.4.5.final.0
python-bits: 64
OS: Windows
OS-release: 7
machine: AMD64
processor: Intel64 Family 6 Model 60 Stepping 3, GenuineIntel
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: None.None

pandas: 0.19.1
nose: 1.3.7
pip: 9.0.1
setuptools: 27.2.0
Cython: 0.24.1
numpy: 1.11.2
scipy: 0.18.1
statsmodels: 0.6.1
xarray: 0.8.2
IPython: 5.1.0
sphinx: 1.4.8
patsy: 0.4.1
dateutil: 2.6.0
pytz: 2016.7
blosc: 1.5.0
bottleneck: 1.2.0
tables: 3.2.2
numexpr: 2.6.1
matplotlib: 1.5.1
openpyxl: 2.4.0
xlrd: 1.0.0
xlwt: 1.1.2
xlsxwriter: 0.9.3
lxml: 3.6.4
bs4: 4.5.3
html5lib: 0.999
httplib2: 0.9.2
apiclient: None
sqlalchemy: 1.1.3
pymysql: None
psycopg2: 2.6.2 (dt dec pq3 ext lo64)
jinja2: 2.8
boto: 2.43.0
pandas_datareader: None

[louispotok]

I agree this behavior seems odd. I was surprised to discover that the code is explicitly checking for this situation. This is the relevant code: https://github.com/pandas-dev/pandas/blob/v0.20.3/pandas/core/frame.py#L5055

if axis == 1 and self._is_mixed_type and self._is_datelike_mixed_type:
    numeric_only = True

This check was added here, though I don't know why.

[jreback]

DataFrame._reduce is quite complex because it handles many things. This should actually work block-by-block and dispatch based on the uniform dtype. see how something like isna works on the internals. This would be very similar.

[mroeschke]

This looks fixed on master. Could use a test.

In [47]: df.any(axis=1)
Out[47]:
0    False
1     True
dtype: bool

In [48]: pd.__version__
Out[48]: '0.26.0.dev0+533.gd8f9be7e3'