Why does adding CDay() to DatetimeIndex include a weekend?

Question

I am on pandas version 1.1.3.

Suppose I have the following index

import pandas as pd
from pandas.tseries.offsets import CDay

idx = pd.bdate_range("2010-02-03","2010-02-12") + pd.Timedelta('22H')

If I then add:

shifted_idx = idx + 0*CDay(offset=pd.Timedelta('4H'))

I would expect the timestamp in idx of 2010-05-05 22:00 to move to 2010-05-08 02:00 since CDay() has weekmask = ‘Mon Tue Wed Thu Fri’ by default.

However, we see in shifted_idx a timestamp 2010-02-06 02:00:00 which is on a Saturday and violates the weekmask?

Answer 1

The offset parameter is applied after the day shift is applied, that's why you are getting these results.

You can see that by taking a look at the source code of the apply method of the CustomBusinessDay class. Note that CDay is just an alias of CustomBusinessDay.

Not sure if this counts as a bug, because you can argue that this is the intended behavior. I couldn't find any documentation to support nor oppose this.

For now, you can work around this by manually applying the offset first, then applying the mask. Something like this:

idx_hours_offset = idx + pd.Timedelta('4H')                                          

shifted_idx = idx_hours_offset + 0*CDay()

Answer 2

By typing shifted_idx = idx + 0*CDay(offset=pd.Timedelta('4H')) you just add 4h to the previous timestamp, the CDay thingy doesn't work there.

It's much simpler to do a normal pd.date_range and add the four hours to that range and finally slice the dataframe using the following code:

    import pandas as pd
    idx = pd.date_range("2010-02-03","2010-02-12") + pd.Timedelta('22H')
    shifted_idx = idx + pd.Timedelta('4H')
    shifted_idx = shifted_idx[shifted_idx.dayofweek < 5]