Pandas timeseries indexing fails when the index is hierarchical - pandas

Question

I tried the following code snippet.
In [84]:
from datetime import datetime
from dateutil.parser import parse
​
rng = [datetime(2017,1,13), datetime(2017,1,14), datetime(2017,2,15), datetime(2017,2,16)]
​
s = Series([1,2,3,4], index=rng)
​s['2017/1']
Out[84]:
2017-01-13 1
2017-01-14 2
dtype: int64
As I expected, I could successfully retrieve only those items belonging to JAN by only specifying up to JAN like s['2017/1'].
Next time, I tried a bit extended version of the above code, where a hierarchical index was used instead:
from datetime import datetime
from dateutil.parser import parse
rng1 = [datetime(2017,1,1), datetime(2017,1,1), datetime(2017,2,1), datetime(2017,2,1)]
rng2 = [datetime(2017,1,13), datetime(2017,1,14), datetime(2017,2,15), datetime(2017,2,16)]
midx = pd.MultiIndex.from_arrays([rng1, rng2])
s = Series([1,2,3,4], index=midx)
s['2017/1']
The above code snippet, however, generates an error:
TypeError: unorderable types: int() > slice()
Would you give me some help?

Answer 1

It seems it is more complicated.
Partial string indexing on datetimeindex when part of a multiindex is implemented in DataFrame in pandas 0.18.
So if use:
rng1 = [pd.Timestamp(2017,5,1), pd.Timestamp(2017,5,1),
pd.Timestamp(2017,6,1), pd.Timestamp(2017,6,1)]
rng2 = pd.date_range('2017-01-13', periods=2).tolist() +
pd.date_range('2017-02-15', periods=2).tolist()
s = pd.Series([1,2,3,4], index=[rng1, rng2])
print (s)
2017-05-01 2017-01-13 1
2017-01-14 2
2017-06-01 2017-02-15 3
2017-02-16 4
Then for me works:
print (s.to_frame().loc[pd.IndexSlice[:, '2017/1'],:].squeeze())
2017-05-01 2017-01-13 1
2017-01-14 2
Name: 0, dtype: int64
print (s.loc['2017/6'])
2017-06-01 2017-02-15 3
2017-02-16 4
dtype: int64
But this return empty Series:
print (s.loc[pd.IndexSlice[:, '2017/2']])
Series([], dtype: int64