PERF: PeriodIndex speed up

[max-sixty]

• closes #PERF: PeriodIndex.size #14822
• tests added / passed
• passes git diff upstream/master | flake8 --diff
• whatsnew entry

[jreback]

great! pls add a whatsnew note. and if you can post the asv results would be great.

[max-sixty]

    before     after       ratio
  [3ccb5013] [13aa1309]
+    4.48ms     6.10ms      1.36  period.period_standard_indexing.time_shape
-  659.85μs   572.45μs      0.87  period.period_standard_indexing.time_intersection
-  258.36μs   166.30μs      0.64  period.period_standard_indexing.time_series_loc
-  135.68μs    48.90μs      0.36  period.period_standard_indexing.time_get_loc
-    3.95ms    31.21μs      0.01  period.period_standard_indexing.time_shallow_copy

This is before the Shape change.

Current branch is v fast:

In [5]: index = pd.PeriodIndex(start=2000, periods=20000, freq='B')

In [7]: %timeit -n 1 -r 1 index.shape
1 loop, best of 1: 12.2 µs per loop

Re-running ASV now (doesn't seem easy to do a quick dev run...)

[jreback]

@MaximilianR that's fine. maybe needs np.random.seed to be set (to make it consitent). but no big deal.

[max-sixty]

Ideally this would be parameterized with Index type. For another day though.

[jreback]

lgtm. ping on green.

[jreback]

======================================================================
FAIL: test_getitem_ndim2 (pandas.tseries.tests.test_period.TestPeriodIndex)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "C:\Python27_64\envs\pandas\lib\site-packages\pandas\tseries\tests\test_period.py", line 2110, in test_getitem_ndim2
    self.assertEqual(result.shape, (len(idx), ))
AssertionError: Tuples differ: (3L, 1L) != (3,)
First tuple contains 1 additional elements.
First extra element 1:
1L
- (3L, 1L)
+ (3,)
------------------

[max-sixty]

I'll look

[max-sixty]

Any idea why this is here? Creating another axis on an index?

   2104     def test_getitem_ndim2(self):
   2105         idx = period_range('2007-01', periods=3, freq='M')
   2106
   2107         result = idx[:, None]
   2108         # MPL kludge, internally has incorrect shape
   2109         tm.assertIsInstance(result, PeriodIndex)
-> 2110         self.assertEqual(result.shape, (len(idx), ))

[max-sixty]

Weird but different things happen with other indexes:

In [7]: pd.DatetimeIndex(start='2000', periods=10, freq='B')
Out[7]:
DatetimeIndex(['2000-01-03', '2000-01-04', '2000-01-05', '2000-01-06',
               '2000-01-07', '2000-01-10', '2000-01-11', '2000-01-12',
               '2000-01-13', '2000-01-14'],
              dtype='datetime64[ns]', freq='B')

In [8]: pd.DatetimeIndex(start='2000', periods=10, freq='B')[:, None].shape
Out[8]: (10, 1)

In [9]: pd.DatetimeIndex(start='2000', periods=10, freq='B')[:, None]
Out[9]:
array([['2000-01-03T00:00:00.000000000'],
       ['2000-01-04T00:00:00.000000000'],
       ['2000-01-05T00:00:00.000000000'],
       ['2000-01-06T00:00:00.000000000'],
       ['2000-01-07T00:00:00.000000000'],
       ['2000-01-10T00:00:00.000000000'],
       ['2000-01-11T00:00:00.000000000'],
       ['2000-01-12T00:00:00.000000000'],
       ['2000-01-13T00:00:00.000000000'],
       ['2000-01-14T00:00:00.000000000']], dtype='datetime64[ns]')

In [10]: pd.DatetimeIndex(start='2000', periods=10, freq='B').shape
Out[10]: (10,)

[jreback]

I would leave this here, maybe causing the ndim test sto fail

[max-sixty]

It's not what's causing the test failure (it could only be different if cls != PeriodIndex

[jreback]

this is a bit bogus, but I suspect this is in there for some matplotib compat (we pass periods to it I think for it to display some things).

[max-sixty]

It's only changing the output of .shape - nothing else. This is from master:

In [3]: pd.PeriodIndex(start='2000', periods=10, freq='B')[:, None].shape
Out[3]: (10,)

In [4]: pd.PeriodIndex(start='2000', periods=10, freq='B')[:, None].values.shape
Out[4]: (10,)

In [5]: pd.PeriodIndex(start='2000', periods=10, freq='B')[:, None]._values.shape
Out[5]: (10, 1)

I can hack change .shape to only look at the first dimension, or if nothing else fails are you OK taking that test out?

[jreback]

ok taking the test out, if all plotting things pass (iow everything else is ok).

[codecov-io]

Current coverage is 84.64% (diff: 100%)

Merging #14931 into master will decrease coverage by <.01%

@@             master     #14931   diff @@
==========================================
  Files           144        144          
  Lines         51023      51025     +2   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits          43190      43191     +1   
- Misses         7833       7834     +1   
  Partials          0          0          

Powered by Codecov. Last update 3ab369c...72b14aa

[jreback]

thanks!