ERR: invalid MultiIndex construction

[flipdazed]

Code Sample, a copy-pastable example if possible

import pandas as pd
import numpy as np
idx0 = range(2)
idx1 = np.repeat(range(2), 2)

midx = pd.MultiIndex(
    levels=[idx0, idx1],
    labels=[
        np.repeat(range(len(idx0)), len(idx1)),
        np.tile(range(len(idx1)), len(idx0))
    ],
    names=['idx0', 'idx1']
)

df = pd.DataFrame(
    [
        [i**2/float(j), 'example{}'.format(i), i**3/float(j)]
        for j in range(1, len(idx0) + 1)
        for i in range(1, len(idx1) + 1)
    ],
    columns=['col0', 'col1', 'col2'],
    index=midx
)

example = df.loc[[(0, 1)]]

For display

In [13]: example
Out[13]:
           col0      col1  col2
idx0 idx1
0    1     16.0  example4  64.0

Problem description

Taken from this StackOverflow post:

• https://stackoverflow.com/q/46098097/4013571

Given the following dataframe

           col0      col1  col2
idx0 idx1
0    0      1.0  example1   1.0
     0      4.0  example2   8.0
     1      9.0  example3  27.0
     1     16.0  example4  64.0
1    0      0.5  example1   0.5
     0      2.0  example2   4.0
     1      4.5  example3  13.5
     1      8.0  example4  32.0

the .xs operation will select

In [121]: df.xs((0,1), level=[0,1])
Out[121]:
           col0      col1  col2
idx0 idx1
0    1      9.0  example3  27.0
     1     16.0  example4  64.0

whilst the .loc operation will select

In [125]: df.loc[[(0,1)]]
Out[125]:
           col0      col1  col2
idx0 idx1
0    1     16.0  example4  64.0

This is highlighted even further by the following

In [149]: df.loc[pd.IndexSlice[:, 1], :]
Out[149]:
           col0      col1  col2
idx0 idx1
0    1      9.0  example3  27.0
     1     16.0  example4  64.0

In [150]: df.loc[pd.IndexSlice[0, 1], :]
Out[150]:
col0          16
col1    example4
col2          64
Name: (0, 1), dtype: object

Expected Output

Note that this only works for this minimal example because there is only one label in level 0 index axis

In [8]: df.loc[pd.IndexSlice[:, 1], :]
Out[8]:
           col0      col1  col2
idx0 idx1
0    1      9.0  example3  27.0
     1     16.0  example4  64.0

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 2.7.13.final.0
python-bits: 64
OS: Windows
OS-release: 7
machine: AMD64
processor: Intel64 Family 6 Model 79 Stepping 1, GenuineIntel
byteorder: little
LC_ALL: None
LANG: None
LOCALE: None.None

pandas: 0.20.3
pytest: 3.0.7
pip: 9.0.1
setuptools: 36.2.7
Cython: 0.25.2
numpy: 1.13.1
scipy: 0.19.0
xarray: None
IPython: 5.3.0
sphinx: 1.5.6
patsy: 0.4.1
dateutil: 2.6.1
pytz: 2017.2
blosc: None
bottleneck: 1.2.1
tables: 3.2.2
numexpr: 2.6.2
feather: None
matplotlib: 2.0.2
openpyxl: 2.4.7
xlrd: 1.0.0
xlwt: 1.2.0
xlsxwriter: 0.9.6
lxml: 3.7.3
bs4: 4.6.0
html5lib: 0.999
sqlalchemy: 1.1.9
pymysql: None
psycopg2: None
jinja2: 2.9.6
s3fs: None
pandas_gbq: None
pandas_datareader: 0.5.0

[colinalexander]

The midx returned as constructed above is as follows:

MultiIndex(levels=[[0, 1], 
                   [0, 0, 1, 1]],
           labels=[[0, 0, 0, 0, 1, 1, 1, 1], 
                   [0, 0, 1, 1, 0, 0, 1, 1]],
           names=[u'idx0', u'idx1'])

Whereas typical construction would be:

midx = pd.MultiIndex.from_product([[0, 1], [0, 0, 1, 1]], names=['idx0', 'idx1'])
>>> midx
MultiIndex(levels=[[0, 1], 
                   [0, 1]],
           labels=[[0, 0, 0, 0, 1, 1, 1, 1], 
                   [0, 0, 1, 1, 0, 0, 1, 1]],
           names=[u'idx0', u'idx1'])

Using the typical construction, indexing via df.loc[[(0, 1)]] appears to behave as expected.

I believe the issue here is how this indexing handles duplicates in one of the levels (i.e. the 1).

[jreback]

your directly constructing the MultiIndex is violating guarantees, namely that the levels are each unique. We don't explicitly check this as the public constructors guarantee this. I suppose we could check this when verify_integrity=True.

want to do a PR for this?

to clarify, you certainly can have a non-unique MultiIndex (though generally discouraged as they are not that performant), but you would have duplicate labels, never level values.