BUG: Empty PeriodIndex issues

[max-sixty]

• closes BUG: Period resample with length=0 doesn't set freq #13067 (existing PR)
• tests added / passed
• passes git diff upstream/master | flake8 --diff
• whatsnew entry

closes #13067
closes #13212

[sinhrks]

can u also test _simple_new adding name attribute?

[max-sixty]

The test above does that, no?

[sinhrks]

above doesn't check empty case. Though _shallow_copy internally uses _simple_new, I meant to guarantee _simple_new API.

[max-sixty]

@jreback Is this the best way to do this? Previously it was creating float64 arrays.

[jreback]

that would probably work. IIUC, this is a non ndarray, that could be ints, or objects (meaning Periods, or coercable to).

have a look at com.is_period_arraylike as well. Though that actually traverses.

[max-sixty]

com.is_period_arraylike didn't work, so we're back to the try except

[jreback]

explain what changing is.

[jreback]

also say: .resample(..) with a PeriodIndex

[max-sixty]

@jreback @sinhrks:
Any idea what this might be? It only happens on 2.7 and so I would need to rebuild to step through. No prob if nothing comes to mind though:

======================================================================
ERROR: test_resample_empty (pandas.tseries.tests.test_resample.TestPeriodIndex)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/travis/build/pydata/pandas/pandas/tseries/tests/test_resample.py", line 2054, in test_resample_empty
    result = getattr(s.resample(freq), method)()
  File "/home/travis/build/pydata/pandas/pandas/tseries/resample.py", line 511, in f
    return self._downsample(_method)
  File "/home/travis/build/pydata/pandas/pandas/tseries/resample.py", line 805, in _downsample
    return self._groupby_and_aggregate(how, grouper=grouper)
  File "/home/travis/build/pydata/pandas/pandas/tseries/resample.py", line 379, in _groupby_and_aggregate
    result = grouped.apply(how, *args, **kwargs)
  File "/home/travis/build/pydata/pandas/pandas/core/groupby.py", line 645, in apply
    @wraps(func)
  File "/home/travis/miniconda/envs/pandas/lib/python2.7/functools.py", line 33, in update_wrapper
    setattr(wrapper, attr, getattr(wrapped, attr))
AttributeError: 'str' object has no attribute '__module__'
----------------------------------------------------------------------
Ran 10105 tests in 537.776s
FAILED (SKIP=177, errors=1)

[jreback]

I remember seeing those. I think an attribute is being passed which isn't defined for that object
e.g. nunique for a DataFrame will cause this.

[jreback]

doesn't make sense; the bug is returning an incorrect freq when empty, right?

[sinhrks]

This looks to change the current behavior below.

pd.PeriodIndex._simple_new(np.array([pd.Period('2011-01', freq='M')]))
# PeriodIndex(['2011-01'], dtype='int64', freq='M')

[jreback]

I think that is correct though, any reason we don't want that?

@MaximilianR as an aside pls run asv on --bench period to verify no changes (under py3) in perf

[max-sixty]

This is ugly and I would love a better way of doing it.
I do think the outcomes are correct though - @sinhrks do you disagree?
Open to nice implementation though

[sinhrks]

I just wondering whether the following behavior (on current master) are the same. Also is no inconsistencies between list and ndarray.

pd.PeriodIndex._simple_new([pd.Period('2011-01', freq='M')])
# PeriodIndex(['2011-01'], dtype='int64', freq='M')

pd.PeriodIndex._simple_new(np.array([pd.Period('2011-01', freq='M')]))
# PeriodIndex(['2011-01'], dtype='int64', freq='M')

pd.PeriodIndex._simple_new([1.1], freq='M')
# ValueError: Ordinal must be an integer

pd.PeriodIndex._simple_new(np.array([1.1]), freq='M')
# ValueError: Ordinal must be an integer

[jreback]

@sinhrks not sure I understand, those seem right.

[max-sixty]

I think @sinhrks is asking whether the new implementation is consistent with those examples, from master. I'll check

[max-sixty]

It wasn't working well. Changed & tested.

[jreback]

want to rebase/update

[max-sixty]

is that #13212 ?

Yup!

[sinhrks]

can u explicitly compare .freq also, as assert_index_equal doesn't (i'll submit another PR once this is fixed).

[max-sixty]

Should I remove this test now @jreback? I think it's covered by the Base tests now https://github.com/pydata/pandas/blob/master/pandas/tseries/tests/test_resample.py#L1411

[jreback]

yes (just inspect it to make sure that we are covering the bases)

[max-sixty]

OK, I moved a couple around to ensure we weren't losing anything

[jreback]

can you put this in Base? (e.g. its a generic test)

[max-sixty]

done. I had thought generic was to run tests across a number of types, while this doesn't depend on the index type.

[max-sixty]

@jreback @sinhrks Green!

[jreback]

thanks @MaximilianR @sinhrks this is great8!

small issue. I am not convinced that we are testing IncompatibleFrequency fully, IOW we may be just testing for ValueError. But that's a minor point & can be done as a followup.

[sinhrks]

thanks @MaximilianR :)