Label slices on Series with MultiIndex

[LeartS]

I have a Series with a MultiIndex like this:

In [174]: a
Out[174]: 
hour  year  weekday  workingday       
0     2011  0        0           count      4.000000
                                 mean      76.750000
                                 std       53.368998
                                 min       17.000000
                                 25%       44.000000
                                 50%       75.000000
                                 75%      107.750000
                                 max      140.000000
                     1           count     28.000000
                                 mean      29.500000
                                 std       31.792732
                                 min        4.000000
                                 25%       15.000000
                                 50%       24.500000
                                 75%       35.500000
...
23    2012  5        0           mean     152.687500
                                 std       55.711490
                                 min       20.000000
                                 25%      119.000000
                                 50%      148.000000
                                 75%      187.750000
                                 max      239.000000
            6        0           count     34.000000
                                 mean      71.617647
                                 std       32.471913
                                 min       23.000000
                                 25%       46.000000
                                 50%       68.500000
                                 75%      104.250000
                                 max      123.000000
Length: 3456, dtype: float64

And I'd like to get, for example, the rows where hour=1, year=2011, weekday=[1,2,3,4]

Intuitively, I would do this: a.loc[1,2011,1:4] which would work in a 4-dimensional numpy array. Unfortunately it raises the error:

/usr/lib/python2.7/dist-packages/pandas/core/index.pyc in get_loc_level(self, key, level, drop_level)
   3214                             continue
   3215                         else:
-> 3216                             raise TypeError(key)
   3217 
   3218                     if indexer is None:

TypeError: (0, 2011, slice(0, 4, None))

Even If I try suppling the index list instead of a slice: a.loc[0,2011,[1,2,3,4]] I get an error: unhashable type: 'list'
Is there a reason why this syntax is not supported? How can I select labels slices in a MultiIndex?

[jreback]

docs are here: http://pandas.pydata.org/pandas-docs/dev/indexing.html#multiindexing-using-slicers

this is a 2-d structure so what you are writing is ambiguous at best

what if you had a MultiIndex in 2 dimensions?

using MultiIndex slicers this is easily accomplishes though

requires pandas >= 0.14

[LeartS]

Ah! Thanks. Unfortunately I have 0.13, at least for a while.

Given that as long as I use single labels it works (e.g. a.loc[0,2011,1,0] works) it seemed really intuitive that a.loc[0,2001,0:3,0], for example, would work too. I don't see any ambiguity on Series with column-only (or row-only) MultiIndex.

[jreback]

you can use a tuple to specify the various axes snd a list of tuples
doing this in 0.13 is a bit non trivial with slices

their is no ambiguity in a series but the syntax works for multi dimensions and is much more clear

the syntax you are using really shouldn't be allowed IMHO but remains for back compat

best bet is to upgrade

[jorisvandenbossche]

@jreback this is a Series, not a DataFrame (so 1d structure), so I think what the OP did should work given pandas >= 0.14 (and it does also work) without an index slicer. Why shouldn't this be allowed?

[jreback]

This is fixed in 0.15.0 by #8132

Yes this is valid syntax for a Series (only)

In [1]: index = pd.MultiIndex.from_product([['one','two'],
   ...:                                     range(3),
   ...:                                     range(3),
   ...:                                     range(3)],names=['first','seconds','third','fourth'],
   ...:                                    )

In [2]: df = pd.DataFrame(np.arange(len(index)).reshape(-1,1),columns=['value'],index=index)

In [3]: 

In [3]: 

In [3]: df.loc['one',1,1,1]
Out[3]: 
value    13
Name: (one, 1, 1, 1), dtype: int64

In [4]: 

In [4]: df.loc['one',1,1,1:3]
Out[4]: 
                            value
first seconds third fourth       
one   1       1     1          13
                    2          14

In [5]: 

In [5]: df.loc['one',1,1:3]
Out[5]: 
                            value
first seconds third fourth       
one   1       1     0          12
                    1          13
                    2          14
              2     0          15
                    1          16
                    2          17