API: support "unique=True" in MultiIndex.get_level_values()

[toobaz]

Code Sample, a copy-pastable example if possible

I often find my self doing

In [2]: df = pd.Series(index=pd.MultiIndex.from_product([['A', 'B'], ['a', 'b']]))

In [3]: df.index.get_level_values(0).unique()
Out[3]: Index(['A', 'B'], dtype='object')

Problem description

The above is very inefficient, because first a Series is built which includes a copy of the entire level (possibly using way more memory than the index itself), and only then duplicates are stripped. Other people on SO have faced the same problem, and this is also blocking a fix I wrote for #17845.

I'm pushing a simple PR in seconds.

Expected Output

Same as above, but in an efficient way.

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.5.3.final.0
python-bits: 64
OS: Linux
OS-release: 4.9.0-3-amd64
machine: x86_64
processor:
byteorder: little
LC_ALL: None
LANG: it_IT.UTF-8
LOCALE: it_IT.UTF-8

pandas: 0.21.0rc1+19.gb15d92d14
pytest: 3.0.6
pip: 9.0.1
setuptools: None
Cython: 0.25.2
numpy: 1.12.1
scipy: 0.19.0
pyarrow: None
xarray: None
IPython: 5.1.0.dev
sphinx: 1.5.6
patsy: 0.4.1
dateutil: 2.6.0
pytz: 2017.2
blosc: None
bottleneck: 1.2.1
tables: 3.3.0
numexpr: 2.6.1
feather: 0.3.1
matplotlib: 2.0.0
openpyxl: None
xlrd: 1.0.0
xlwt: 1.1.2
xlsxwriter: 0.9.6
lxml: None
bs4: 4.5.3
html5lib: 0.999999999
sqlalchemy: 1.0.15
pymysql: None
psycopg2: None
jinja2: 2.9.6
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: 0.2.1

[toobaz]

By the way: this is in principle related to #2770, which however is being tackled in a different and complementary way.

[jreback]

how is this not just .get_levels(..) ?

[jreback]

#2770 is handled by remove_unused_levels()

[toobaz]

how is this not just .levels ?

.levels includes unused labels (which is why users are often confused by it)

[jreback]

ok, you are adding it there, ok!.

I am not sure unique is the right word here.

[jreback]

.get_level_values(level, used=False), though I am not sure I like this either.

[jorisvandenbossche]

I agree it would be nice to have a clean way to get those unique values, but IMO it does not belong in get_level_values. That method returns the actual values of the Index level, with a length equal to the length of the Index, and IMO we should stick to that contract. Having such a keyword would completely alter the return type of this method.

(not directly a good idea for alternative though)