How to use Pandas groupby week when the week number spans more than one year

user_3466313 0
Viewed 784

I need to groupby week, however a week like this one (the first week of the year) spans two years, 2018 and 2019.

Typically I would do the following: 

df.groupby([df.DATE.dt.year,df.DATE.dt.week]).sum()

which results in the single week characterized as two separate weeks in the output. I am sure I can brute force with IF statements, however I am wondering if there is a more clean way to group by week during these year transitions.

pandas-groupby pandas python
2 Answers

You can convert the dates to pandas Period objects, and then group on them.

df = pd.DataFrame(
    {'Date': pd.DatetimeIndex(start='2018-12-24', end='2019-01-05', freq='d'),
     'Data': [1] * 8 + [2] * 5})
>>> df
         Date  Data
0  2018-12-24     1
1  2018-12-25     1
2  2018-12-26     1
3  2018-12-27     1
4  2018-12-28     1
5  2018-12-29     1
6  2018-12-30     1
7  2018-12-31     1
8  2019-01-01     2
9  2019-01-02     2
10 2019-01-03     2
11 2019-01-04     2
12 2019-01-05     2

>>> (df
     .assign(period=pd.PeriodIndex(df['Date'], freq='W-Sun'))  # Weekly periods ending Sundays.
     .groupby('period')['Data'].mean())
period
2018-12-24/2018-12-30    1.000000
2018-12-31/2019-01-06    1.833333  # (1 * 1 + 2 * 5) / 6 = 1.833 
Freq: W-SUN, Name: Data, dtype: float64

Note that there are only six days in the final period in the example above.

user_24118020

Or just using resample

df.set_index('Date').resample('W-SUN').Data.mean()
Date
2018-12-30    1.000000
2019-01-06    1.833333
Freq: W-SUN, Name: Data, dtype: float64
user_79645270
Related