Resample yields empty groups

[JonasAbernot]

With some parameters, the last group yield by resample is empty. Example :

import pandas as pd
import numpy as np

df = pd.DataFrame(np.random.normal(size=(10000,4)))
df.index = pd.timedelta_range(start='0s', periods=10000, freq='3906250n')

df.loc['1s':,:].resample('3s',how=lambda x : len(x))

Depending of the 'how' function used, this can lead to surprising bugs.

INSTALLED VERSIONS
------------------
commit: None
python: 2.7.8.final.0
python-bits: 64
OS: Linux
OS-release: 3.13.0-46-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: fr_FR.UTF-8

pandas: 0.16.0-294-g45f69cd
nose: 1.3.6
Cython: 0.20.2
numpy: 1.9.2
scipy: 0.14.0
statsmodels: 0.5.0
IPython: 3.0.0-dev
sphinx: 1.2.2
patsy: 0.3.0
dateutil: 2.4.2
pytz: 2015.4
bottleneck: None
tables: 3.1.1
numexpr: 2.4
matplotlib: 1.4.3
openpyxl: 1.7.0
xlrd: 0.9.2
xlwt: 0.7.5
xlsxwriter: None
lxml: None
bs4: 4.3.2
html5lib: 0.999
httplib2: None
apiclient: None
sqlalchemy: 0.8.2
pymysql: None
psycopg2: 2.5.3 (dt dec mx pq3 ext)

[jreback]

In [7]: pd.set_option('max_rows',12)

In [8]: df.loc['1s':,:]
Out[8]: 
                        0         1         2         3
00:00:01         0.767847 -1.805006  0.513914 -0.533759
00:00:01.003906  1.034297 -0.873930  1.254777 -0.460738
00:00:01.007812 -1.905457 -0.497061 -0.550036 -0.400423
00:00:01.011718  0.526214 -0.569812 -0.817764  1.204511
00:00:01.015625  0.061491  0.939611  0.308094  1.300434
00:00:01.019531 -0.147869  0.971442  1.239615  0.637635
...                   ...       ...       ...       ...
00:00:39.039062  1.664856 -0.821650 -0.551620 -0.442644
00:00:39.042968  1.133944  0.797726 -0.677378 -0.488098
00:00:39.046875 -0.343148 -0.123394 -1.010421  1.476257
00:00:39.050781  0.311632 -0.418035 -1.200112 -1.735927
00:00:39.054687  0.291330 -0.559795 -0.516269  1.088944
00:00:39.058593  0.918740 -0.516714 -0.415188  0.106167

[9744 rows x 4 columns]

In [9]: df.loc['1s':,:].resample('3s',how=lambda x : len(x))
Out[9]: 
            0    1    2    3
00:00:01  768  768  768  768
00:00:04  768  768  768  768
00:00:07  768  768  768  768
00:00:10  768  768  768  768
00:00:13  768  768  768  768
00:00:16  768  768  768  768
...       ...  ...  ...  ...
00:00:25  768  768  768  768
00:00:28  768  768  768  768
00:00:31  768  768  768  768
00:00:34  768  768  768  768
00:00:37  528  528  528  528
00:00:40    0    0    0    0

[14 rows x 4 columns]

looks correct. The last group is just a point as this is evently divisible.

[jreback]

FYI also show the actual data (generated from the code) if reporting a bug, as its then clear by simply looking what's the problem.

[jreback]

any reason you are not using how='count' (its the same result just much faster)

[JonasAbernot]

Yep, ok, I wasn't clear enough. The last group contains no point : The index finishes at 39'05, and the group begins at 40'. For the 'count' task, it is actually not a problem. But for others which can't support 0-length objects this can be annoying.

Example:

df = pd.DataFrame(np.random.normal(size=(10000,4)))
df.index = pd.timedelta_range(start='0s', periods=10000, freq='3906250n')
from scipy import fft

Something that works:

In [25]: df.resample('3s',how=lambda x : max(fft(x)))
Out[25]: 
                  0          1          2          3
00:00:00  55.527131  63.876320  50.189927  60.702282
00:00:03  53.586627  63.214890  55.694863  55.196211
00:00:06  63.159294  51.598472  61.389132  60.393747
00:00:09  73.133776  63.760377  69.555783  64.445265
00:00:12  60.349962  48.913074  50.045405  57.562742
00:00:15  58.858030  49.733304  55.012356  62.641561
...             ...        ...        ...        ...
00:00:24  59.661202  61.519860  49.886808  49.105434
00:00:27  48.506358  55.936740  52.039330  57.650969
00:00:30  52.030271  58.446403  59.234081  64.254844
00:00:33  57.767135  56.672450  52.793359  69.297208
00:00:36  56.431251  64.871565  63.356116  67.926122
00:00:39   5.341347   5.263054   5.745918   4.918816

[14 rows x 4 columns]

Something that doesn't :

In [26]: df.loc['1s':,:].resample('3s',how=lambda x : max(fft(x)))
Out[26]: 
Empty DataFrame
Columns: []
Index: []

Just because of generating a empty group (wich is weird), fft fails, and this leads (silently) to this empty DataFrame.

I hope this is clearer now.

(About not using 'count', the only reason is my lack of culture)

[mroeschke]

This looks to work on master now. Could use a test

In [17]: import pandas as pd
    ...: import numpy as np
    ...:
    ...: df = pd.DataFrame(np.random.normal(size=(10000,4)))
    ...: df.index = pd.timedelta_range(start='0s', periods=10000, freq='3906250n')
    ...:
    ...: df.loc['1s':,:].resample('3s').apply(lambda x: len(x))
Out[17]:
                     0      1      2      3
0 days 00:00:01  768.0  768.0  768.0  768.0
0 days 00:00:04  768.0  768.0  768.0  768.0
0 days 00:00:07  768.0  768.0  768.0  768.0
0 days 00:00:10  768.0  768.0  768.0  768.0
0 days 00:00:13  768.0  768.0  768.0  768.0
0 days 00:00:16  768.0  768.0  768.0  768.0
0 days 00:00:19  768.0  768.0  768.0  768.0
0 days 00:00:22  768.0  768.0  768.0  768.0
0 days 00:00:25  768.0  768.0  768.0  768.0
0 days 00:00:28  768.0  768.0  768.0  768.0
0 days 00:00:31  768.0  768.0  768.0  768.0
0 days 00:00:34  768.0  768.0  768.0  768.0
0 days 00:00:37  528.0  528.0  528.0  528.0