Week Start/End Resamping: WS/WE frequencies

[quasiben]

There are a number of offset aliases listed here:
http://pandas.pydata.org/pandas-docs/dev/timeseries.html#offset-aliases

I'm currently trying to resample a timeseries where the buckets are return with the beginning and/or end of the week. There are a number of Start Month/End Month frequency aliases. Am I missing something or should we include Week End/Week Start resampling aliases as well?

[jreback]

look at the next section
u can specify w-mon for example

[quasiben]

Yes, the anchored offsets are nice. But these will change the values:

In [1]: import pandas as pd

In [2]: import datetime as dt

In [3]: daily_daterange = pd.date_range(start=dt.datetime(2013, 1, 31),
   ...:                                 end=dt.datetime(2013,2,28),
   ...:                                 freq='B')

In [4]: daily_dataset = pd.DataFrame({'value' : range(len(daily_daterange))}, index=daily_daterange)

In [5]: daily_dataset.resample('W-MON')
Out[5]: 
            value
2013-02-04      1
2013-02-11      5
2013-02-18     10
2013-02-25     15
2013-03-04     19

In [6]: daily_dataset.resample('W-SUN')
Out[6]: 
            value
2013-02-03    0.5
2013-02-10    4.0
2013-02-17    9.0
2013-02-24   14.0
2013-03-03   18.5

I want the values to stay the same and the buckets returned to be the beginning of the week.

[jreback]

ok

so give an example then of what u want

[quasiben]

Thu  2013-01-31  val_1
Fri  2013-02-01  val_2
Sat  2013-02-02  val_3
Sun  2013-02-03  val_4

In [4]: daily_dataset.resample('W-THU')
Out[4]: 
            value
2013-01-31      0
2013-02-07      3
2013-02-14      8
2013-02-21     13
2013-02-28     18

In [5]: daily_dataset.resample('W-MON')
Out[5]: 
            value
2013-02-04      1
2013-02-11      5
2013-02-18     10
2013-02-25     15
2013-03-04     19

In [6]: daily_dataset.resample('W-SUN')
Out[6]: 
            value
2013-02-03    0.5
2013-02-10    4.0
2013-02-17    9.0
2013-02-24   14.0
2013-03-03   18.5

I want to resample weekly but the bucket returned should be the weeks beginning:

daily_dataset.resample('WBEGIN') 
            value
2013-02-03    0
2013-02-10    3
2013-02-17    8
2013-02-24   13
2013-03-03   18

Essentially, adding a timedelta to the index.

[jreback]

how is this diffferent from W-SUN ? (I mean your values are), but conceptually

[quasiben]

Conceptually, nothing is different but the resampling of the values is significant. I believe month end and month start frequency don't change the values but it does return the a bucket at the beginning of the month or the end of month

You could do something like:

daily_dataset.resample('7D')

Then fix the index with an appropriate offset. Maybe I should submit a PR and not waste too much more time :)

[jreback]

why wouldn't you reindex to a daily freq, then simply divide by 7 starting on the day you want?

[aiguofer]

+1 on this. I keep running into issues, and none of the solutions are nice.... a simple WS would make my life easier.

Take this for example. Feb in 2016 started on a MON, and the last week of march 31 was on THURS.

I want to group by week (starting on MON and ending on SUN), count the number of days in the week, but use the first day of the week as the label. That means that for the month of Feb in 2016, my expected output is:

2016-02-01    7
2016-02-08    7
2016-02-15    7
2016-02-22    7
2016-02-29    1

I can do this a few different ways, but they all seem cumbersome and not very intuitive.

import pandas as pd
df = pd.DataFrame(pd.date_range('2016-02-01', '2016-02-29').rename('dt'))
df['val'] = 1

# option 1
a = df.groupby(pd.Grouper(key='dt',freq='W-MON', closed='left', label='left'))['val'].sum()

# option 2
a = df.groupby(pd.Grouper(key='dt',freq='W'))['val'].sum()
a.index = a.index + pd.DateOffset(weekday=0, days=-7)

# option 3
a = df.groupby(pd.Grouper(key='dt',freq='W-MON', closed='left'))['val'].sum()
a.index = a.index + pd.DateOffset(days=-7)

# option 4
a = df.groupby(pd.Grouper(key='dt',freq='W', label='left'))['val'].sum()
a.index = a.index + pd.DateOffset(days=1)

# ideal solution
a = df.groupby(pd.Grouper(key='dt',freq='WS'))['val'].sum()