Index.map followups

[jreback]

xref #15081, a couple of issues

• IntervalIndex.map is xfailed
• parametrize indexex/common.py/test_map similar to how indexex/datetimelike.py/test_map is done (IOW on dict / Series) (COMPAT: map infers all-nan / empty correctly #18491)
• resolve datetimelike with empty map [2] should be all NaT for the datetimelikes. (COMPAT: map infers all-nan / empty correctly #18491)

In [2]: pd.date_range('20130101', periods=3).map({})
Out[2]: Float64Index([nan, nan, nan], dtype='float64')

In [3]: pd.Index([1,2,3]).map({})
Out[3]: Float64Index([nan, nan, nan], dtype='float64')

[jreback]

cc @jorisvandenbossche (point 3)
cc @jschendel (point 1)

[jschendel]

Regarding point 1: To be clear, the underlying issue referenced here is #16410, right? test_map xpasses locally for me, and that was the closest thing I could find based on the xfail message. The problem described in #16410 is still occurring, but I haven't immediately been able to produce any errors with map using a similar setup.

[jschendel]

Side note: should mapping an empty dictionary against an IntervalIndex return an IntervalIndex? Currently returns a Float64Index, which seems kind of strange:

In [2]: pd.interval_range(0, 4).map({})
Out[2]: Float64Index([nan, nan, nan, nan], dtype='float64')

[jreback]

@jschendel I just noticed that test_interval.py/test_map is xfailed. It might have been disabled during this re-factoring. if that is the case, then just need to re-eable.

#18491 fixes all of the Indexes to return correctly there own type if {} is passed. I think II should do the same (but again still xfailed).

[jorisvandenbossche]

@jreback Is there a specific reason you pinged me for point 3 ? (apart from general attention)

[jreback]

@jorisvandenbossche no, you had raised the point in the original issue (note this is all now resolved by #18491)

[jorisvandenbossche]

Ah, I forgot that one :-)

[louispotok]

(I don't know if this is the right place to put this, happy to open a new issue if that's better.)

According to my reading of this PR, the following should work as of 0.22.0:

s = pd.Series({0:'a',1:'b'})
s.index.map(pd.Series({0:'c',1:'d'})) 

But for me it throws a ValueError. Whereas this works fine:

s = pd.Series({0:'a',1:'b'})
pd.Series(s.index).map(pd.Series({0:'c',1:'d'}))```

[jschendel]

@louispotok : The PR you linked to has a milestone of 0.23.0, so the behavior should only work as of that release (should be relatively soon):

In [2]: pd.__version__
Out[2]: '0.23.0rc2'

In [3]: s = pd.Series({0:'a',1:'b'})

In [4]: s.index.map(pd.Series({0:'c',1:'d'}))
Out[4]: Index(['c', 'd'], dtype='object')

[louispotok]

Oh thanks @jschendel , my mistake: I saw this event putting it in the 0.22.0, and didn't see that the milestone was later moved.

[jschendel]

Ah, I was basing the 0.23.0 milestone off the right hand side summary at the top of the PR. Probably moved to 0.23.0 after the fact; the 0.22.0 release happened pretty shortly after 0.21.x and only contained one major change. Many things originally slated for 0.22.0 got moved to 0.23.0 after the being merged to accommodate the new plan for the 0.22.0 release.

[mroeschke]

@jschendel is this closable? I don't see an xfail on any test_map methods.

[jschendel]

Yes, I think this can be closed.