Allow level wildcard via slice(None) in df.ix[] with MultiIndex

[timcera]

arrays = [['bar', 'bar', 'baz', 'baz', 'foo', 'foo', 'qux', 'qux'],['one', 'two', 'one', 'two', 'one', 'two', 'one', 'two']]
tuples = zip(*arrays)
index = pd.MultiIndex.from_tuples(tuples, names=['first', 'second'])
df = DataFrame(randn(3, 8), index=['A', 'B', 'C'], columns=index)
df = df.T

The following works:

df.A.ix[slice(None), 'two']

first
bar      0.233442
baz     -1.551486
foo      0.175177
qux      1.524440
Name: A

It is very possible that this is something about a MultiIndex DataFrame that I don't understand, but I think the following should work:

df.ix[slice(None), 'two']

Instead of the expected:

first
bar      0.233442      0.359638  2.061794
baz     -1.551486    -1.888951  0.172784
foo      0.175177     -1.085777 -0.988670
qux      1.524440     0.070535  1.006851

I get a KeyError:

...
/sjr/beodata/local/python_linux/lib/python2.6/site-packages/pandas/core/internals.pyc in _check_have(self, item)
949 def _check_have(self, item):
950 if item not in self.items:
--> 951 raise KeyError('no item named %s' % com.pprint_thing(item))
952
953 def reindex_axis(self, new_axis, method=None, axis=0, copy=True):

KeyError: u'no item named two'

[ghost]

AFICT from the code, the args of ix correspond to the axes, so you're asking for all rows
and column "two". which fails.
case in point:

In [4]: df.ix[:]
Out[4]: 
                     A         B         C
first second                              
bar   one    -0.720545 -0.382630  0.573031
      two    -0.263034  0.462324 -0.126281
baz   one     1.676899 -0.660316  1.216486
      two    -0.343970  0.234571  0.347938
foo   one    -0.563490 -1.136923 -0.450143
      two    -1.209016  0.044605  0.879672
qux   one    -0.276785  0.563070 -0.133299
      two    -0.449211  0.545187 -0.869852

In [6]: df.ix[:,('A','B')]
Out[6]: 
                     A         B
first second                    
bar   one    -0.720545 -0.382630
      two    -0.263034  0.462324
baz   one     1.676899 -0.660316
      two    -0.343970  0.234571
foo   one    -0.563490 -1.136923
      two    -1.209016  0.044605
qux   one    -0.276785  0.563070
      two    -0.449211  0.545187

This also works

In [10]: df.ix[('foo','two'),:]
Out[10]: 
A   -1.209016
B    0.044605
C    0.879672
Name: (foo, two)

This however doesn't, which is disappointing:

In [7]: df.ix[(slice(None),'two')]
/home/user1/src/pandas/pandas/core/internals.pyc in _check_have(self, item)
   1002     def _check_have(self, item):
   1003         if item not in self.items:
-> 1004             raise KeyError('no item named %s' % com.pprint_thing(item))
   1005 
   1006     def reindex_axis(self, new_axis, method=None, axis=0, copy=True):

KeyError: u'no item named two'

and maybe that is a bug.

If your question comes from a real need,

In [7]: df.xs('two',level=1)
Out[7]: 
              A         B         C
first                              
bar   -0.677739 -0.740875  0.072675
baz    0.061356 -1.522032  1.084492
foo   -0.124634 -2.342294 -0.625460
qux   -0.647809 -0.051477  0.724003

will get you there, until this is otherwise addressed.

p.s. I think you may have mixed up columns and index in your example.

[timcera]

Fixed my example about the column/row confusion - forgot a command during copy/paste...

[timcera]

I have a 5 level MultiIndex DataFrame and want to select rows by wildcard one or more of the levels. I figured out a solution yesterday - outside of the pandas framework by using 'get_level_values' for each level to build a pseudo database. Works, and actually is fast enough.

When trying to use native pandas I was hoping that if 'df.ix' supported 'slice(None)' I could use it for my DataFrame as something like:

df.ix[('PERLND', 112, 'PWATER', slice(None), 2)]

Which would wildcard the fourth level - right? If that make sense, that is what I think I want.

[ghost]

related #1766

[jreback]

closing in favor of #4036