BUG: Spurious IncompatibleFrequency error prevented plotting of non-standard intervals (Fixes #14763)

[jeffcarey]

• closes Period Index plotting method, non standard intervals #14763
• tests added: tests/plotting/test_series.py: test_non_standard_intervals
• passes git diff upstream/master | flake8 --diff
• whatsnew entry - Confirm version and I'll add

Fix Summary: A check on frequency compatibility was unnecessarily strict. Now IncompatibleFrequency exceptions will only get raised if the unit types differ.

[codecov-io]

Current coverage is 85.26% (diff: 100%)

Merging #14850 into master will increase coverage by <.01%

@@             master     #14850   diff @@
==========================================
  Files           144        144          
  Lines         50979      50979          
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits          43468      43469     +1   
+ Misses         7511       7510     -1   
  Partials          0          0          

Powered by Codecov. Last update 856476b...fb5f046

[jorisvandenbossche]

Looks good to me!

Can you add

• a whatsnew notice in v0.20.0.txt in the bug fixes section?
• a test for PeriodIndex(array of Periods) with different but compatible freqs

cc @sinhrks

[jorisvandenbossche]

Can you add the github issue number as a comment?

[jreback]

pls also add a whatsnew entry (0.20.0) is fine.

[jeffcarey]

@jorisvandenbossche @jreback @sinhrks

After doing some more testing I don't think the fix I've made here is the correct one, so I am going to close this pull request for now. What I submitted does result in the desired plots but I believe it also breaks some behavior in PeriodIndex. The right way to do this is a good deal more complicated than I originally thought, and I don't think I am the right one to do it given my relative inexperience working with this codebase. Happy to discuss my findings if anyone else wants to try to solve this.

Example of an undesired side effect of the fix:

idx1 = PeriodIndex(period_range('2000-01-01', '2000-01-02', freq='6H'))
idx2 = idx1.append([ period_range('2000-01-03', '2000-01-04', freq='5H')])

# idx2
PeriodIndex(['2000-01-01 00:00', '2000-01-01 06:00', '2000-01-01 12:00',
'2000-01-01 18:00', '2000-01-02 00:00', '2000-01-03 00:00',
'2000-01-03 05:00', '2000-01-03 10:00', '2000-01-03 15:00',
'2000-01-03 20:00'],
dtype='period[6H]', freq='6H')

As you can see, you get a PeriodIndex which says its freq is 6H, but some of the points have a 5H difference. In the current master this code would just result in a plain ndarray.