ENH: support .strftime for datetimelikes

[jreback]

I think Series.dt.strftime() might be useful
This is essentially just calling .format() on a datetimelike lindex (which handles NaT and such)

In [1]: dates = pd.to_datetime(pd.Series(['20010101', '20010331']), format = '%Y%m%d')

In [2]: dates
Out[2]: 
0   2001-01-01
1   2001-03-31
dtype: datetime64[ns]

In [4]: pd.DatetimeIndex(dates.values).format()
Out[4]: ['2001-01-01', '2001-03-31']

In [5]: pd.DatetimeIndex(dates.values).format(date_format='%Y%m%d')
Out[5]: ['20010101', '20010331']

[jorisvandenbossche]

Why not just exposing the already existing format function through the dt accessor instead of adding a new method?

[jreback]

strftime is the 'expected' name here (expected meaning what you use to format datetime.datetime (as well as Timestamp)

[mortada]

One problem I have with format() is that it is not clear the format string should be passed via date_format=. I'd always try to do

In [9]: pd.DatetimeIndex(dates.values).format('%Y/%m/%d')
Out[9]: ['', '2001-01-01', '2001-03-31']

and realize that it doesn't work. (also not sure why there's an empty string in front in this case, seems like a bug)

Maybe this should be opened as a separate issue/PR?

[jreback]

.format() is an internal method (only really used by the to_csv, and takes a bunch of arguments). That's why I am suggesting strftime