better error message for to_datetime

[cpcloud]

cc @danbirken

In [2]: to_datetime([1,'1'])
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-2-1d4cd9e078aa> in <module>()
----> 1 to_datetime([1,'1'])

/home/phillip/Documents/code/py/pandas/pandas/tseries/tools.pyc in to_datetime(arg, errors, dayfirst, utc, box, format, coerce, unit)
    137         return Series(values, index=arg.index, name=arg.name)
    138     elif com.is_list_like(arg):
--> 139         return _convert_listlike(arg, box=box)
    140
    141     return _convert_listlike(np.array([ arg ]), box=box)[0]

/home/phillip/Documents/code/py/pandas/pandas/tseries/tools.pyc in _convert_listlike(arg, box)
    117                 result = tslib.array_to_datetime(arg, raise_=errors == 'raise',
    118                                                  utc=utc, dayfirst=dayfirst,
--> 119                                                  coerce=coerce, unit=unit)
    120             if com.is_datetime64_dtype(result) and box:
    121                 result = DatetimeIndex(result, tz='utc' if utc else None)

/home/phillip/Documents/code/py/pandas/pandas/tslib.so in pandas.tslib.array_to_datetime (pandas/tslib.c:16487)()

TypeError: object of type 'int' has no len()

[danbirken]

Yeah this is a little strange. I don't know enough about the history of these functions, but in the documentation of to_datetime() it says:

arg : string, datetime, array of strings (with possible NAs)

However, array_to_datetime() and by proxy to_datetime() supports certain cases of being passed arrays of ints or floats:

In [2]: pd.to_datetime([1, 2])
Out[2]:
<class 'pandas.tseries.index.DatetimeIndex'>
[1970-01-01 00:00:00.000000001, 1970-01-01 00:00:00.000000002]
Length: 2, Freq: None, Timezone: None

In [3]: pd.to_datetime([1.5, 2.5])
Out[3]:
<class 'pandas.tseries.index.DatetimeIndex'>
[1970-01-01 00:00:00.000000001, 1970-01-01 00:00:00.000000002]
Length: 2, Freq: None, Timezone: None

So I can:

a) Make it so you must pass an array of strings to to_datetime(), otherwise throw a nice exception that says you passed in a list that wasn't strings

b) Make it so array_to_datetime() (and by proxy to_datetime()) returns the same array back in case it cannot be processed. So like:

# Current behavior for two un-processable strings
In [4]: pd.to_datetime(['1', '2'])
Out[4]: array(['1', '2'], dtype=object)

# Potential behavior for your input.  Un-processable, just return it back
In [4]: pd.to_datetime([1, '1'])
Out[4]: array([1, '1'], dtype=object)

c) some other option I didn't think of

[jreback]

their are a bunch test case I think in tseries/test/test_timeseries IIRC.

on errors it will just return the input, unless errors='strict', or if coerce=True then will make errors into NaT

[jreback]

@danbirken can you address this one? I think that b) is the correct response here (it will just return it to you), unless errors='raise' is passed in which case it will raise

although, maybe in this case as the input is completely bogus, you could raise....

[danbirken]

Sorry for the delay in responding -- I'm still on a very long vacation :)

I think this change should do it.

[jreback]

gr8...just roll that into a PR and put a release notes entry...thankxs

[danbirken]

Put into PR, release notes should be in there.