How to use pandas to group pivot table results by week?

Question

Below is a snippet of my pivot table output in .csv format after using pandas pivot_table function:

Sub-Product     11/1/12 11/2/12 11/3/12 11/4/12 11/5/12 1         


        

Answer 1

The tool you need is resample, which implicitly uses groupby over a time period/frequency and applies a function like mean or sum.

Read data.

In [2]: df
Out[2]: 
      Sub-Product  11/1/12  11/2/12  11/3/12  11/4/12  11/5/12  11/6/12
GP   Acquisitions      164      168       54       72      203      167
GP   Applications      190      207       65       91      227      200
GPF  Acquisitions     1124     1142      992     1053     1467     1198
GPF  Applications     1391     1430     1269     1357     1855     1510

Set up a MultiIndex.

In [4]: df = df.reset_index().set_index(['index', 'Sub-Product'])

In [5]: df
Out[5]: 
                    11/1/12  11/2/12  11/3/12  11/4/12  11/5/12  11/6/12
index Sub-Product                                                       
GP    Acquisitions      164      168       54       72      203      167
      Applications      190      207       65       91      227      200
GPF   Acquisitions     1124     1142      992     1053     1467     1198
      Applications     1391     1430     1269     1357     1855     1510

     Parse the columns as proper datetimes. (They come in as strings.)

In [6]: df.columns = pd.to_datetime(df.columns)

In [7]: df
Out[7]: 
                    2012-11-01  2012-11-02  2012-11-03  2012-11-04  \
index Sub-Product                                                    
GP    Acquisitions         164         168          54          72   
      Applications         190         207          65          91   
GPF   Acquisitions        1124        1142         992        1053   
      Applications        1391        1430        1269        1357   

                    2012-11-05  2012-11-06  
index Sub-Product                           
GP    Acquisitions         203         167  
      Applications         227         200  
GPF   Acquisitions        1467        1198  
      Applications        1855        1510  

Resample the columns (axis=1) weekly ('w'), summing by week. (how='sum' or how=np.sum are both valid options here.)

In [10]: df.resample('w', how='sum', axis=1)
Out[10]: 
                    2012-11-04  2012-11-11
index Sub-Product                         
GP    Acquisitions         458         370
      Applications         553         427
GPF   Acquisitions        4311        2665
      Applications        5447        3365