Issues Creating Period with DateOffset as freq

[cancan101]

Something seems wrong here:

print repr(pd.Period("2013-12", freq=pd.offsets.MonthEnd()))

correctly prints:

Period('2013-12', 'M')

but:

print pd.Period("2013-12", freq=pd.offsets.BusinessMonthEnd())

leads to:

Traceback (most recent call last):
  File "/home/alex/git/pandas/pandas/tseries/tests/test_52.py", line 29, in testName
    a = pd.Period("2013-12", freq=BusinessMonthEnd())
  File "/home/alex/git/pandas/pandas/tseries/period.py", line 121, in __init__
    base, mult = _gfc(freq)
  File "/home/alex/git/pandas/pandas/tseries/frequencies.py", line 92, in get_freq_code
    code = _period_str_to_code(freqstr[0])
  File "/home/alex/git/pandas/pandas/tseries/frequencies.py", line 757, in _period_str_to_code
    alias = _period_alias_dict[freqstr]
KeyError: 'BM'

Strangely, the following both work:

In [10]: pd.date_range("2013-01","2013-10",freq=pd.offsets.MonthEnd())
Out[10]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-01-31 00:00:00, ..., 2013-09-30 00:00:00]
Length: 9, Freq: M, Timezone: None

In [11]: pd.date_range("2013-01","2013-10",freq=pd.offsets.BusinessMonthEnd())
Out[11]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2013-01-31 00:00:00, ..., 2013-09-30 00:00:00]
Length: 9, Freq: BM, Timezone: None

[cancan101]

The exception actually looks like:

Traceback (most recent call last)
<ipython-input-19-fe5191046ba2> in <module>()
----> 1 print pd.Period("2013-12", freq=pd.offsets.BusinessMonthEnd())

/usr/local/lib/python2.7/dist-packages/pandas/tseries/period.pyc in __init__(self, value, freq, ordinal, year, month, quarter, day, hour, minute, second)
    118             raise ValueError(msg)
    119 
--> 120         base, mult = _gfc(freq)
    121         if mult != 1:
    122             raise ValueError('Only mult == 1 supported')

/usr/local/lib/python2.7/dist-packages/pandas/tseries/frequencies.pyc in get_freq_code(freqstr)
     91                 stride = freqstr[1]
     92             except:
---> 93                 code = _period_str_to_code(freqstr[1])
     94                 stride = freqstr[0]
     95             return code, stride

/usr/local/lib/python2.7/dist-packages/pandas/tseries/frequencies.pyc in _period_str_to_code(freqstr)
    738     # hack
    739     freqstr = _rule_aliases.get(freqstr, freqstr)
--> 740     freqstr = _rule_aliases.get(freqstr.lower(), freqstr)
    741 
    742     try:

AttributeError: 'int' object has no attribute 'lower'

The more informative stack trace in the OP comes from making the following change: cancan101@6789ef8#L1R95

[cancan101]

Some more examples of inconsistencies:

pd.period_range('1/1/2012', periods=4, freq=pd.offsets.BusinessMonthEnd())

yields:

/usr/local/lib/python2.7/dist-packages/pandas/tseries/frequencies.pyc in get_freq_code(freqstr)
     99 
    100     base, stride = _base_and_stride(freqstr)
--> 101     code = _period_str_to_code(base)
    102 
    103     return code, stride

/usr/local/lib/python2.7/dist-packages/pandas/tseries/frequencies.pyc in _period_str_to_code(freqstr)
    744         return _period_code_map[freqstr]
    745     except:
--> 746         alias = _period_alias_dict[freqstr]
    747         return _period_code_map[alias]
    748 

KeyError: 'BM'

whereas:

pd.date_range('1/1/2012', periods=4, freq=pd.offsets.BusinessMonthEnd()).to_period()

works correctly.

I have tracked part of this issue down:
period_range uses _period_str_to_code which in turn uses _period_code_map (or _period_alias_dict). Currently 'BM' is not in these dicts.
to_period on the other hand uses get_period_alias which in turn uses _offset_to_period_map.

[cancan101]

So this could be a simple fix, but I want to get some feedback:

I plan to add this test:

    def test_period_range_alias(self):
        self.assertTrue(
            pd.date_range('1/1/2012', periods=4, freq=pd.offsets.MonthEnd()).to_period().identical(
            pd.period_range('1/1/2012', periods=4, freq=pd.offsets.MonthEnd())))

        #GH 4878
        self.assertTrue(
           pd.date_range('1/1/2012', periods=4, freq=pd.offsets.BusinessMonthEnd()).to_period().identical(
           pd.period_range('1/1/2012', periods=4, freq=pd.offsets.BusinessMonthEnd())))

Currently the first line passes, but the second fails.

To summarize the problem: Pandas uses DateOffset to represent fixed point in time but a string to represent time periods. Currently not all DateOffset have corresponding period strings. BusinessMonthEnd is an example of a DateOffset that does not have a corresponding period string. Currently when calling to_period on a DatetimeIndex, pandas will use the following:

def get_period_alias(offset_str):
    """ alias to closest period strings BQ->Q etc"""
    return _offset_to_period_map.get(offset_str, None)

for which there is in fact a mapping 'BM': 'M'. This same aliasing is NOT used when directly calling period_range for which the correct period string must be specified. If a DateOffset is passed into to period_range, Pandas uses the rulecode for the DateOffset which for BusinessMonthEnd is BM which, as stated above, is not a valid period string.

One idea would be for DateOffset to have a periodstr property. Another would be if the user passes in an DateOffset to period_range to pass its rulecode through get_period_alias.

Thoughts?

[jtratner]

@cancan101 weren't you thinking of reworking the backend for all of this? Would it make more sense to address this after you've made the changes there? I don't think there's any reason why we would not want your example to run without an error.

[cancan101]

@jtratner #5148 is my WIP for this. So far the above unit test passes.

[jtratner]

@cancan101 Now it has a new error - hurray:

Traceback (most recent call last):
  File "testcase2.py", line 5, in <module>
    pd.Period("2013-12", freq=BusinessMonthEnd())
  File "../pandas/tseries/period.py", line 122, in __init__
    base, mult = _gfc(freq)
  File "../pandas/tseries/frequencies.py", line 98, in get_freq_code
    code = _period_str_to_code(freqstr[1])
  File "../pandas/tseries/frequencies.py", line 619, in _period_str_to_code
    freqstr = _rule_aliases.get(freqstr.lower(), freqstr)
AttributeError: 'int' object has no attribute 'lower'

I guess I messed something up here.

[cancan101]

@jtratner That issue (at least the poor exception raised) is fixed on my retail branch.

[jtratner]

great - is that ready to go or soon to be ready? (not sure if you rebased it)

[cancan101]

See
https://github.com/pydata/pandas/pull/5004/files#diff-18135e4852fd97f51aabf9d5be9f528dR99
for
reference. I am rebasing and dealing with conflicts as we speak.

On Mon, Oct 14, 2013 at 11:04 PM, Jeff Tratner notifications@github.comwrote:

great - is that ready to go or soon to be ready? (not sure if you rebased
it)

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/4878#issuecomment-26306283
.

[jreback]

This is with your PR; is this right?

In [10]: pd.period_range('1/1/2012', periods=4, freq=pd.offsets.BusinessMonthEnd())
Out[10]: 
<class 'pandas.tseries.period.PeriodIndex'>
freq: M
[2012-01, ..., 2012-04]
length: 4

In [11]: pd.period_range('1/1/2012', periods=4, freq=pd.offsets.MonthEnd())
Out[11]: 
<class 'pandas.tseries.period.PeriodIndex'>
freq: M
[2012-01, ..., 2012-04]
length: 4

[cancan101]

Yes. Why?

[jreback]

should the first not be freq 'BM'?

[cancan101]

I believe this is the "issue" I address here: https://github.com/pydata/pandas/pull/5148/files#diff-18135e4852fd97f51aabf9d5be9f528dR414

LMK if that does not make sense.

[jreback]

ok...so why are they not different?