pandas groupby resample unexpected result

piRSquared Published at Dev

piRSquared

Setup

import pandas as pd

df = pd.DataFrame({'grp': [1, 2] * 2, 'value': range(4)},
                  index=pd.Index(pd.date_range('2016-03-01', periods=7)[::2], name='Date')
                 ).sort_values('grp')

I wanted to group by 'grp' and resample my index daily, forward filling missing values. I expected this to work:

print df.groupby('grp').resample('D').ffill()

            grp  value
Date                  
2016-03-01    1      0
2016-03-05    1      2
2016-03-03    2      1
2016-03-07    2      3

It did not. So I tried this:

print df.groupby('grp', group_keys=False).apply(lambda df: df.resample('D').ffill())

            grp  value
Date                  
2016-03-01    1      0
2016-03-02    1      0
2016-03-03    1      0
2016-03-04    1      0
2016-03-05    1      2
2016-03-03    2      1
2016-03-04    2      1
2016-03-05    2      1
2016-03-06    2      1
2016-03-07    2      3

It did work. Shouldn't these two methods have produced the same output? What am I missing?

Response to ayhan's comment

print sys.version
print pd.__version__

2.7.11 |Anaconda custom (x86_64)| (default, Dec  6 2015, 18:57:58) 
[GCC 4.2.1 (Apple Inc. build 5577)]
0.18.0

ayhan showed that the results looked the same on python 3, pandas 18.1

After updating pandas to 18.1

2.7.11 |Anaconda custom (x86_64)| (default, Dec  6 2015, 18:57:58) 
[GCC 4.2.1 (Apple Inc. build 5577)]
0.18.1

The issue has been resolved.

ayhan

It looks like one of the issues due to the changes in resample API in version 0.18.0.

It works as expected in 0.18.1:

df.groupby('grp').resample('D').ffill()
Out[2]: 
                grp  value
grp Date                  
1   2016-03-01    1      0
    2016-03-02    1      0
    2016-03-03    1      0
    2016-03-04    1      0
    2016-03-05    1      2
2   2016-03-03    2      1
    2016-03-04    2      1
    2016-03-05    2      1
    2016-03-06    2      1
    2016-03-07    2      3

Collected from the Internet

Please contact [email protected] to delete if infringement.