BUG: median() not correctly handling non-float null values (fixes #10…

[mortada]

closes #10040

[mortada]

Note that the existing unit test is actually wrong

In [1]: from pandas import Series, Timestamp

In [2]: from datetime import timedelta

In [3]: s = Series([Timestamp('20130101') + timedelta(seconds=i*i) for i in range(10)])

In [4]: td = s.diff()

In [5]: td
Out[5]:
0        NaT
1   00:00:01
2   00:00:03
3   00:00:05
4   00:00:07
5   00:00:09
6   00:00:11
7   00:00:13
8   00:00:15
9   00:00:17
dtype: timedelta64[ns]

td.median() should really be 00:00:09 instead of 00:00:08 as it is currently asserting.

[jreback]

pls add a release note for 0.16.1

[mortada]

@jreback it's updated

[jreback]

ty!

side issue. I briefly looked it and it appears to work, but are there cases where precision drops are an issue? e.g. we have int64 converted to float64 which drops precision, but it DOES seem to work. if you can experiment and see (I don't have a good) example. If you find somethign pls file an issue.

[mortada]

sure I'll play with this and see if I can find examples where precisions become an issue. It does seem like it could be a problem.