BUG: xs with drop_level=True with a fully specified tuples buggy

[jreback]

related #6501

In [1]: n=23

In [2]: df = pd.DataFrame({'a':np.append(np.random.randint(0,10,n), -1),
   ...:                    'b':np.append(np.random.randint(0,10,n), -1),
   ...:                    'c':np.append(np.random.randint(0,10,n), -1),
   ...:                    'value':np.random.randint(0,100,n+1)})
In [3]: df.set_index(['a','b','c'], inplace=True)

In [4]: df.sortlevel(inplace = True)

In [7]: df.xs((-1,-1,-1),drop_level=True)
Out[7]: 
value    40
Name: (-1, -1, -1), dtype: int64

In [8]: df.xs((-1,-1),drop_level=True)
Out[8]: 
    value
c        
-1     40

[1 rows x 1 columns]

[hayd]

Pretty racy example...

[hayd]

So it becomes a Series when it should be a frame?

Also, incorrect for drop_level=False, presumably same reason.

In [21]: df.xs((-1,-1, -1),drop_level=False)
Out[21]:
value    59
Name: (-1, -1, -1), dtype: int64

expected = df.xs((-1,-1),drop_level=False)

Annoying no way to get from Series (name) to DataFrame with correct mi... maybe we could use namedtuple (ag, but this would fail on dupe names :S)

[jreback]

I was thinking that this should be a scalar
but looking again this example only has 1 column
so it should be a series always I think

this is equiv to
df.loc[indexer]

drop_level=False with a fully specified tuple is just
like passing [indexer]

hmm one could argue that it should be a scalar
but what if their are multiple return values then it had to be a series

[hayd]

it's weird cos everything is inconsistent (the index goes from having 3 levels to 2 levels to 1 level to ...er). IMO should be DataFrame as could have more columns.

Confusing.