to_datetime conversion inconsistency in 0.15.0

[pmart123]

to_datetime is not converting input consistently between a pandas series or a python list of either strings or datetime objects. for example:

import pandas as pd
s = pd.Series(['102214','102314'])
df = pd.Series({'dates':['102214','102314']})
lst = ['102214','102314']

pd.to_datetime(s,format='%m%d%y') and pd.to_datetime(df['dates'],format='%m%d%y') return a numpy.datetime64 ndarray, while pd.to_datetime(lst,format='%m%d%y') returns a pandas.DateTimeIndex object. Is this behavior an expected result in 0.15.0?

[jorisvandenbossche]

The actual values will be the same, but the object type of the output that holds those values depends on the input type: by default it created a DatetimeIndex (eg for lists, arrays, etc as input), only when passed a Series, the output will also be a Series.

See http://pandas.pydata.org/pandas-docs/stable/timeseries.html#converting-to-timestamps as explained in the docs, but this should actually also be stated in the docstring here http://pandas.pydata.org/pandas-docs/stable/generated/pandas.to_datetime.html

[jreback]

this is correct and expected

In [5]: pd.to_datetime(s,format='%m%d%y')
Out[5]: 
0   2014-10-22
1   2014-10-23
dtype: datetime64[ns]

In [6]: pd.to_datetime(df['dates'],format='%m%d%y')
Out[6]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2014-10-22, 2014-10-23]
Length: 2, Freq: None, Timezone: None

In [7]: pd.to_datetime(lst,format='%m%d%y')
Out[7]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2014-10-22, 2014-10-23]
Length: 2, Freq: None, Timezone: None

[6] and [7] are lists, while [5] is a series (so it returns the same).

[jorisvandenbossche]

Are you interested in making a PR to add this to the docstring? (otherwise I will do it quickly)

[jorisvandenbossche]

This was closed by #8921