Median returns odd value when applied to timedeltas

[AndreaBravi]

from pandas import DataFrame
from numpy import datetime64
data = ['2015-02-03', '2015-02-07']
data = DataFrame(data, dtype=datetime64)
data.diff().median()

This code returns

0   -53374 days +00:06:21.572612
dtype: timedelta64[ns]

while I would expect it to return the same as data.diff().mean()

0   4 days
dtype: timedelta64[ns]

Here is my setup

INSTALLED VERSIONS
------------------
commit: None
python: 2.7.6.final.0
python-bits: 64
OS: Darwin
OS-release: 14.4.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: None
LANG: en_GB.UTF-8

pandas: 0.16.0
nose: 1.3.4
Cython: None
numpy: 1.9.2
scipy: 0.14.0
statsmodels: None
IPython: 2.3.1
sphinx: None
patsy: None
dateutil: 2.4.2
pytz: 2015.2
bottleneck: None
tables: None
numexpr: None
matplotlib: 1.4.2
openpyxl: 2.1.3
xlrd: 0.9.3
xlwt: 0.7.5
xlsxwriter: None
lxml: None
bs4: 4.3.2
html5lib: None
httplib2: None
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: 2.5.4 (dt dec pq3 ext)

[jreback]

xref #9442. The algo that is doing this is converting to float64 so its losing precision. (as I don't think the underling ops is defined for int64).

this needs special handling for int64 in core/nanops/nanmedian. PR's are welcome

[mortada]

so I looked into this and I think it's actually not a precision issue. It's because the function is not handling pd.NaT correctly. Please see #10072

Basically values are being forced into float and therefore pd.NaT becomes a non-null floating point number. median() therefore computes the median with that invalid value (that value happens to be a large negative number)