Surprising results of pandas.to_datetime()

[jankatins]

I just used pd.to_datetime on a string column and inspected the head of the dataframe and it looked good. Unfortunately, I had german dates (day.month.year) with high days in the first rows. These were converted correctly, but then I got dates (of the next month) with low days and the conversion switched to american style (month.day.year). I would expect that I at least get a warning that the date formats are different or that the first found date format is used for all dates:

>>> pd.to_datetime(["29.01.1945","1.3.1945", "02.03.1945"])
DatetimeIndex(['1945-01-29', '1945-01-03', '1945-02-03'], dtype='datetime64[ns]', freq=None)

[jreback]

you need to specify dayfirst (which defaults to False)

In [4]: pd.to_datetime(["29.01.1945","1.3.1945", "02.03.1945"],dayfirst=True)
Out[4]: DatetimeIndex(['1945-01-29', '1945-03-01', '1945-03-02'], dtype='datetime64[ns]', freq=None)

[jankatins]

I know how to work around it, I'm just surprised that pd.to_datetime() can change the interpretation in between values of the same Series.

E.g. if dayfirst == False, then why does it sometimes ("29.01.1945") use a month value as day and not raise/warn?

[jorisvandenbossche]

@JanSchulz You are fully correct this is really confusing/problematic behaviour, but that is a limitation by the dependency on dateutil, as explained (briefly) in the docstring: http://pandas.pydata.org/pandas-docs/stable/generated/pandas.to_datetime.html

[jorisvandenbossche]

See also #3341