import pandas as pd
import numpy as np
dates = pd.date_range('20141229',periods=14, name='Day')
df = pd.DataFrame({'Sum1': [1667, 1229, 1360, 9232, 8866, 4083, 3671, 10085, 10005, 8730, 10056, 10176, 3792, 3518],
'Sum2': [91, 75, 75, 254, 239, 108, 99, 259, 395, 355, 332, 386, 96, 111],
'Sum3': [365.95, 398.97, 285.12, 992.17, 1116.57, 512.11, 504.47, 1190.96, 1753.6, 1646.25, 1344.05, 1582.67, 560.95, 736.44],
'Sum4': [5, 5, 1, 5, 8, 8, 2, 10, 12, 16, 16, 6, 6, 3]},index=dates); print(df)
The df produced looks like this:
Sum1 Sum2 Sum3 Sum4
Day
2014-12-29 1667 91 365.95 5
2014-12-30 1229 75 398.97 5
2014-12-31 1360 75 285.12 1
2015-01-01 9232 254 992.17 5
2015-01-02 8866 239 1116.57 8
2015-01-03 4083 108 512.11 8
2015-01-04 3671 99 504.47 2
2015-01-05 10085 259 1190.96 10
2015-01-06 10005 395 1753.60 12
2015-01-07 8730 355 1646.25 16
2015-01-08 10056 332 1344.05 16
2015-01-09 10176 386 1582.67 6
2015-01-10 3792 96 560.95 6
2015-01-11 3518 111 736.44 3
Let's say I resample the Dataframe to try and sum the daily data into weekly rows:
df_resampled = df.resample('W', how='sum', label='left'); print(df_resampled)
This produces the following:
Sum1 Sum2 Sum3 Sum4
Day
2014-12-28 30108 941 4175.36 34
2015-01-04 56362 1934 8814.92 69
Question 1: my definition of a week is Mon - Sun. Since my data starts on 2014-12-29 (a Monday), I want my Day label to also start on that day. How would I make the Day index label be the date of every Monday instead of every Sunday?
Desired Output:
Sum1 Sum2 Sum3 Sum4
Day
2014-12-29 30108 941 4175.36 34
2015-01-05 56362 1934 8814.92 69
What have I tried regarding Question 1?
I changed 'W' to 'W-MON' but it produced 3 rows by counting 2014-12-29 in 2014-12-22 row which is not what I want:
Sum1 Sum2 Sum3 Sum4
Day
2014-12-22 1667 91 365.95 5
2014-12-29 38526 1109 5000.37 39
2015-01-05 46277 1675 7623.96 59
Question 2: how would I format the Day index label to look like a range? Ex:
Sum1 Sum2 Sum3 Sum4
Day
2014-12-29 - 2015-01-04 30108 941 4175.36 34
2015-01-05 - 2015-01-11 56362 1934 8814.92 69
In case anyone else was not aware, it turns out that the weekly Anchored Offsets are based on the end date. So, just resampling 'W' (which is the same as 'W-SUN') is by default a Monday to Sunday sample. The date listed is the end date. See this old bug report wherein neither the documentation nor the API got updated.
Given that you specified label='left' in the resample parameters, you must have realized that fact. It's also why using 'W-MON' does not have the desired effect. What is confusing is that the left bound is not actually in the interval. As such the
The approach for getting the right label and data in the interval has changed somewhat over different revisions of pandas. The most recent (and hopefully current) approach is listed first for quick access with earlier version for historical reference. All should yield the results listed in the Original Approach at the end of this answer.
Update 1.1.0
The most succinct approach is now using the handy closed parameter that controls which side of the interval is closed/included (see documentation):
df_resampled = df.resample('W-MON', label='left', closed='left').sum()
Thanks to xiaotong xu for noting this simpler form.
An alternate approach since the handy loffset argument is deprecated as of version 1.1.0 is shifting after the resample. In this particular case that would mean:
from pandas.tseries.frequencies import to_offset
df_resampled = df.resample('W', label='left').sum()
df_resampled.index = df_resampled.index + to_offset(pd.DateOffset(days=1))
Update
There is now was a loffset argument to resample() that allows you to shift the label offset. So, instead of modifying the index, you simple add the loffset argument like so:
df.resample('W', how='sum', label='left', loffset=pd.DateOffset(days=1))
Also of note how=sum is now deprecated in favor of using .sum() on the Resampler object that .resample() returns. So, the fully updated call would be:
df_resampled = df.resample('W', label='left', loffset=pd.DateOffset(days=1)).sum()
Original approach
So, to display the start date for the period instead of the end date, you may add a day to the index. That would mean you would do:
df_resampled.index = df_resampled.index + pd.DateOffset(days=1)
For completeness, here is your original data with another day (a Sunday) added on the beginning to show the grouping really is Monday to Sunday:
import pandas as pd
import numpy as np
dates = pd.date_range('20141228',periods=15, name='Day')
df = pd.DataFrame({'Sum1': [10000, 1667, 1229, 1360, 9232, 8866, 4083, 3671, 10085, 10005, 8730, 10056, 10176, 3792, 3518],
'Sum2': [10000, 91, 75, 75, 254, 239, 108, 99, 259, 395, 355, 332, 386, 96, 111],
'Sum3': [10000, 365.95, 398.97, 285.12, 992.17, 1116.57, 512.11, 504.47, 1190.96, 1753.6, 1646.25, 1344.05, 1582.67, 560.95, 736.44],
'Sum4': [10000, 5, 5, 1, 5, 8, 8, 2, 10, 12, 16, 16, 6, 6, 3]},index=dates);
print(df)
df_resampled = df.resample('W', how='sum', label='left')
df_resampled.index = df_resampled.index - pd.DateOffset(days=1)
print(df_resampled)
This outputs:
Sum1 Sum2 Sum3 Sum4
Day
2014-12-28 10000 10000 10000.00 10000
2014-12-29 1667 91 365.95 5
2014-12-30 1229 75 398.97 5
2014-12-31 1360 75 285.12 1
2015-01-01 9232 254 992.17 5
2015-01-02 8866 239 1116.57 8
2015-01-03 4083 108 512.11 8
2015-01-04 3671 99 504.47 2
2015-01-05 10085 259 1190.96 10
2015-01-06 10005 395 1753.60 12
2015-01-07 8730 355 1646.25 16
2015-01-08 10056 332 1344.05 16
2015-01-09 10176 386 1582.67 6
2015-01-10 3792 96 560.95 6
2015-01-11 3518 111 736.44 3
Sum1 Sum2 Sum3 Sum4
Day
2014-12-22 10000 10000 10000.00 10000
2014-12-29 30108 941 4175.36 34
2015-01-05 56362 1934 8814.92 69
I believe that is what you wanted for Question 1.