s.loc[[]] raises error (only) if index is not unique

[toobaz]

Code Sample, a copy-pastable example if possible

In [2]: s = pd.Series(0, index=pd.MultiIndex.from_product([[0], [1,2]]))

In [3]: s.loc[[]]
Out[3]: Series([], dtype: int64)

In [4]: s = pd.Series(0, index=pd.MultiIndex.from_product([[0], [1,1]]))

In [5]: s.loc[[]]
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-5-f9b6211189ca> in <module>()
----> 1 s.loc[[]]

/home/pietro/nobackup/repo/pandas/pandas/core/indexing.py in __getitem__(self, key)
   1304             return self._getitem_tuple(key)
   1305         else:
-> 1306             return self._getitem_axis(key, axis=0)
   1307 
   1308     def _getitem_axis(self, key, axis=0):

/home/pietro/nobackup/repo/pandas/pandas/core/indexing.py in _getitem_axis(self, key, axis)
   1464                     raise ValueError('Cannot index with multidimensional key')
   1465 
-> 1466                 return self._getitem_iterable(key, axis=axis)
   1467 
   1468             # nested tuple slicing

/home/pietro/nobackup/repo/pandas/pandas/core/indexing.py in _getitem_iterable(self, key, axis)
   1095 
   1096                 new_target, indexer, new_indexer = labels._reindex_non_unique(
-> 1097                     keyarr)
   1098 
   1099                 if new_indexer is not None:

/home/pietro/nobackup/repo/pandas/pandas/indexes/base.py in _reindex_non_unique(self, target)
   2497                 new_indexer[~check] = -1
   2498 
-> 2499         new_index = self._shallow_copy_with_infer(new_labels, freq=None)
   2500         return new_index, indexer, new_indexer
   2501 

/home/pietro/nobackup/repo/pandas/pandas/indexes/multi.py in _shallow_copy_with_infer(self, values, **kwargs)
    393 
    394     def _shallow_copy_with_infer(self, values=None, **kwargs):
--> 395         return self._shallow_copy(values, **kwargs)
    396 
    397     @Appender(_index_shared_docs['_shallow_copy'])

/home/pietro/nobackup/repo/pandas/pandas/indexes/multi.py in _shallow_copy(self, values, **kwargs)
    402             # discards freq
    403             kwargs.pop('freq', None)
--> 404             return MultiIndex.from_tuples(values, **kwargs)
    405         return self.view()
    406 

/home/pietro/nobackup/repo/pandas/pandas/indexes/multi.py in from_tuples(cls, tuples, sortorder, names)
    889         if len(tuples) == 0:
    890             # I think this is right? Not quite sure...
--> 891             raise TypeError('Cannot infer number of levels from empty list')
    892 
    893         if isinstance(tuples, (np.ndarray, Index)):

TypeError: Cannot infer number of levels from empty list

Expected Output

Out[5]: Series([], dtype: int64)

By the way, on top of the inconsistency above,

In [6]: s.loc[s.iloc[0:0]]

results in the same error, which is even more unexpected because s.iloc[0:0].index is a perfectly valid MultiIndex, and so it is certainly possible to infer the number of levels from it.

Might be related to #13490 (the error is the same, but I ignore the level at which the open PR applies).

output of pd.show_versions()

In [7]: pd.show_versions()

INSTALLED VERSIONS
------------------
commit: None
python: 3.5.1.final.0
python-bits: 64
OS: Linux
OS-release: 4.5.0-2-amd64
machine: x86_64
processor: 
byteorder: little
LC_ALL: None
LANG: it_IT.utf8
LOCALE: it_IT.UTF-8

pandas: 0.18.1+206.gc42455b
nose: 1.3.7
pip: 1.5.6
setuptools: 18.4
Cython: 0.23.4
numpy: 1.10.4
scipy: 0.16.0
statsmodels: 0.8.0.dev0+111ddc0
xarray: None
IPython: 5.0.0.dev
sphinx: 1.3.1
patsy: 0.3.0-dev
dateutil: 2.2
pytz: 2012c
blosc: None
bottleneck: 1.1.0dev
tables: 3.2.2
numexpr: 2.5
matplotlib: 1.5.1
openpyxl: None
xlrd: 0.9.4
xlwt: 1.1.2
xlsxwriter: 0.7.3
lxml: None
bs4: 4.4.0
html5lib: 0.999
httplib2: 0.9.1
apiclient: 1.5.0
sqlalchemy: 1.0.11
pymysql: None
psycopg2: None
jinja2: 2.8
boto: 2.38.0
pandas_datareader: 0.2.1

[jreback]

ok, will mark it, though non-unique multi-indexes are not very well tested, nor should they really be allowed (but they are) :<

[mborysow]

I have just recently run into this issue as well. I do this very often in fact, and now a lot of my code is broken. I would probably do a lot of differently now, as I've learned more about pandas, but it is what it is. (FYI, merging with df1_good with [left_index=True, right_index=True, how='inner'] produces an equivalent result)

df1 = pd.DataFrame(data=dict(A=[1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2],
B=[1, 1, 2, 2, 2, 3, 1, 1, 1, 2, 3, 4],
C=[1, 2, 1, 2, 3, 2, 1, 2, 3, 2, 1, 4],
V=[1, 5, 2, 3, 8, 3, 3, 3, 1, 2, 1, 4]))
df1 = df1.set_index(['A', 'B', 'C']).sortlevel()

df2 = df1.groupby(level=[0, 2])[['V']].max()

df2_good = df2[df2.V > 9]

...this apparently works and returns an empty dataframe with the index labels and columns intact.
df1_good = df1.loc[df2_good.index]

...this fails with TypeError: Cannot infer number of levels from empty list
df1_good = df1.reset_index().set_index(['A', 'C']).sortlevel().loc[df2_good.index]

Anyhow, it's perfectly acceptable in my analysis that initial query returns an empty DataFrame. This used to work just fine. Now, any time I want to do anything remotely like this I need to always check and make sure that the result of the initial query isn't empty.

This works fine in pandas 0.16, and does not in 0.17 or 0.18. I don't know exactly where it failed, but I get the same error above.