Resampling produces all NaN values for a particular dataset

[brownan]

I ran into a particular dataset that seems to cause Series.resample() to produce incorrect results. Each datapoint is almost but not exactly a minute apart. (This is trimmed down from a larger dataset, but this still reproduces the problem)

>>> import pandas; from pandas import Timestamp
>>> data1 = pandas.Series(
... {
...  Timestamp('2015-03-31 21:48:52.672000'): 2,
...  Timestamp('2015-03-31 21:49:52.739000'): 1,
...  Timestamp('2015-03-31 21:50:52.806000'): 2}
... )
>>> data1
2015-03-31 21:48:52.672000    2
2015-03-31 21:49:52.739000    1
2015-03-31 21:50:52.806000    2
dtype: int64
>>> data1.resample("10S")
2015-03-31 21:48:50   NaN
2015-03-31 21:49:00   NaN
2015-03-31 21:49:10   NaN
2015-03-31 21:49:20   NaN
2015-03-31 21:49:30   NaN
2015-03-31 21:49:40   NaN
2015-03-31 21:49:50   NaN
2015-03-31 21:50:00   NaN
2015-03-31 21:50:10   NaN
2015-03-31 21:50:20   NaN
2015-03-31 21:50:30   NaN
2015-03-31 21:50:40   NaN
2015-03-31 21:50:50   NaN
Freq: 10S, dtype: float64

Expected output:

>>> data1.resample("10S")
2015-03-31 21:48:50   2
2015-03-31 21:49:00   NaN
2015-03-31 21:49:10   NaN
2015-03-31 21:49:20   NaN
2015-03-31 21:49:30   NaN
2015-03-31 21:49:40   NaN
2015-03-31 21:49:50   1
2015-03-31 21:50:00   NaN
2015-03-31 21:50:10   NaN
2015-03-31 21:50:20   NaN
2015-03-31 21:50:30   NaN
2015-03-31 21:50:40   NaN
2015-03-31 21:50:50   2
Freq: 10S, dtype: float64

Changing any one of the timestamps or taking one of the two datapoints out works correctly. Something about this particular dataset causes resample to fail.

I have more examples on this ipython notebook: http://nbviewer.ipython.org/gist/brownan/1000ba324ad7df917e32

INSTALLED VERSIONS
------------------
commit: None
python: 2.7.5.final.0
python-bits: 64
OS: Linux
OS-release: 3.10.0-123.20.1.el7.x86_64
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8

pandas: 0.16.0
nose: 1.3.4
Cython: None
numpy: 1.9.2
scipy: None
statsmodels: None
IPython: 3.0.0
sphinx: 1.3.1
patsy: None
dateutil: 2.4.1
pytz: 2015.2
bottleneck: None
tables: None
numexpr: None
matplotlib: 1.4.3
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
httplib2: None
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: None

edit: I can also reproduce this with the latest version on master: 0.16.0-157-g161f38d

[TomAugspurger]

Looks like we forget to pass along a kwarg, or our inference for downsampling is buggy:

In [19]: data1.resample('10S', how='mean')
Out[19]: 
2015-03-31 21:48:50     2
2015-03-31 21:49:00   NaN
2015-03-31 21:49:10   NaN
2015-03-31 21:49:20   NaN
2015-03-31 21:49:30   NaN
2015-03-31 21:49:40   NaN
2015-03-31 21:49:50     1
2015-03-31 21:50:00   NaN
2015-03-31 21:50:10   NaN
2015-03-31 21:50:20   NaN
2015-03-31 21:50:30   NaN
2015-03-31 21:50:40   NaN
2015-03-31 21:50:50     2
Freq: 10S, dtype: float64

Passing how='mean' explicitly works, but that should be the default here.

[gliptak]

This shows expected result running with pandas: 0.18.0+114.g6c692ae.dirty

[jorisvandenbossche]

@gliptak Do you want to do a PR to explicitly add a test for this to confirm that it is fixed and to close this issue?