ENH: Convert datetimelike index to strings with astype(str)

[jorisvandenbossche]

xref #10110

• Fix .astype(str) on DatetimeIndex to use call .format() (Index method)

- [ ] Add Series.dt.format() accessor to call .format() (#10110 does this with .strftime)

• Change Series.astype(str) to call equiv of Series.dt.format() for datetime64[ns] dtype

Recently the ability to convert a datetimelike series to strings with astype(str) was introduced (#9758 by @evanpw):

In [27]: pd.Series(pd.date_range('2012-01-01', periods=3))
Out[27]:
0   2012-01-01
1   2012-01-02
2   2012-01-03
dtype: datetime64[ns]

In [28]: pd.Series(pd.date_range('2012-01-01', periods=3)).astype(str)
Out[28]:
0    2012-01-01T01:00:00.000000000+0100
1    2012-01-02T01:00:00.000000000+0100
2    2012-01-03T01:00:00.000000000+0100
dtype: object

it would be nice if this also possible on a index (now you still get the same error as previously for a series):

In [29]: pd.date_range('2012-01-01', periods=3).astype(str)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-29-68a28f79c8e9> in <module>()
----> 1 pd.date_range('2012-01-01', periods=3).astype(str)

C:\Users\vdbosscj\Scipy\pandas-joris\pandas\tseries\index.py in astype(self, dty
pe)
    692             return self.asi8.copy()
    693         else:  # pragma: no cover
--> 694             raise ValueError('Cannot cast DatetimeIndex to dtype %s' % d
type)
    695
    696     def _get_time_micros(self):

ValueError: Cannot cast DatetimeIndex to dtype |S0

[jreback]

sure, this is pretty easy.

this is competes a bit with .format() (which takes a formatting argument and is 'internal'; and the .strptime, which I think is still outstanding). But ok by me

[jorisvandenbossche]

yes, for more control over the conversion the .format is certainly the preferred way to do, but for consistency with series this would be good.

Something else I was thinking: we are now using the numpy machinery I suppose to do the conversion, with the result that this also uses the odd numpy representation in local time:

`In [46]: s.astype(str)[0]
Out[46]: '2012-01-01T01:00:00.000000000+0100'

In [47]: str(s[0])
Out[47]: '2012-01-01 00:00:00'

Would it be possible to also have the second output for astype?

[jreback]

hmm, that's a problem, these should all go thru format (with no format specified) for consistency. Further I really though that we added .strptime, but if we didn't ok. should also expose this as Series.dt.format then. I'll add a list to the top. All these pretty straightforward.