slicing dataframe with date index is not working properly

[kanatti]

When slicing data time - indexed dataframe with date in string format, it works:

df.ix['2014-11-12'].head()
Out[55]: 
                     calories      gsr  heart-rate  skin-temp  steps
date                                                                
2014-11-12 00:00:00       2.1  7.04640         NaN    85.7750     30
2014-11-12 00:01:00       1.5  7.40759         NaN    85.7375      0
2014-11-12 00:02:00       1.4  7.46220          49    85.5500      0
2014-11-12 00:03:00       1.4  7.55800         NaN    85.5500      0
2014-11-12 00:04:00       1.5  7.52163         NaN    85.5125      0

But when i try it with a date object, of datetime class , it slices only for time 00:00:00.

d1=datetime.date(2014,11,12)
df.ix[d1]
Out[57]: 
calories       2.1000
gsr            7.0464
heart-rate        NaN
skin-temp     85.7750
steps         30.0000
Name: 2014-11-12 00:00:00, dtype: float64

[jreback]

conceptually similar to #10408. datetime.date objects are not well supported (and likely will not be). simply use datetime.datetime and better yet Timestamp/Period objects and all will be happy.

I suppose one could call this a bug, and if you'd like to dig-in would take a fix.

[jorisvandenbossche]

@jreback I don't know if we should call this a bug. Indexing with a datetime.date does work, but it is the question how it should be interpreted (if it should do partial string indexing or not):
As a timestamp (what you would get from pd.Timestamp(d1)), so then the above behaviour is correct. Or as a 'date' and so similar to providing a string with that date and do a partial string indexing-like thing.

I see the potential confusion in a different result of df.ix['2014-11-12'] and df.ix[date(2014,11,12)]. But personally I would keep the 'partial string indexing' feature restricted to strings.

[jreback]

@jorisvandenbossche hmm, good point. Yeh string indexing encompases internals, but Timestamp/datetime.datetime is an instant in time, its the slicing that actually introduces the interval

so

df.ix[date(2014,11,12):date(2014,11,13)] would be a day.

Introducing objects that expand into a slice then would be a fairly big API change.

[jorisvandenbossche]

Yes, indeed. Therefore I would close this as not-a-bug.
It can be a bit questionable if a date should be regarded as a instant in time like datetime, but I wouldn't start doing slices with single date objects like with strings.

[mroeschke]

Agreed that partial string indexing should be reserved for datetime strings only. Closing as this would provide too much API creep.