CLN: improve is_unique check in Index.intersection

[phofl]

• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff

       before           after         ratio
     [e99e5ab3]       [79d6a50f]
     <cln_intersection~1>       <cln_intersection>
-        918±20ns          819±2ns     0.89  index_object.Range.time_min_trivial
-         465±7ns          414±7ns     0.89  index_object.Range.time_get_loc_dec
-         464±6ns          411±7ns     0.89  index_object.Range.time_get_loc_inc
-      16.6±0.7ms      12.6±0.03ms     0.76  index_object.SetOperations.time_operation('date_string', 'union')

Seems to have no impact.

cc @jreback

[jreback]

ok let's merge this for 1.2, thought it would have more of an impact.

[jbrockmendel]

im surprised this makes a difference. Index.unique should already be doing this check internally i guess

[phofl]

@jreback suggested to check for uniqueness before calling unique() to save a bit of time.

[jreback]

actually @jbrockmendel makes a very good point here. Let's repurpose this PR to actually do a uniquess check internal to unique (of course this may only make sense for > 1000 elements or something).

[jreback]

moving this off 1.2 for a refactor

[phofl]

Added the unique check to Index.unique. Categeorical.unique does remove values which are not within the categories. If we want to keep this, we can not add a check for uniqueness

[jreback]

can you run relevant benchmarks and see if this makes any difference?

[jreback]

this needs to return self._shallow_copy()

[phofl]

Thx.

[phofl]

asvs in index_object:

       before           after         ratio
     [e99e5ab3]       [aba0f2a3]
     <master>         <cln_intersection>
-        47.6±1ms       41.4±0.2ms     0.87  index_object.SetOperations.time_operation('strings', 'intersection')
-      17.0±0.2ms       13.7±0.5ms     0.80  index_object.SetOperations.time_operation('date_string', 'union')

[jbrockmendel]

LGTM

[jreback]

thanks @phofl