Offsets Roundup (original) (raw)

Confusion about offsets with Period. AFAICT the underlying issue is that DateOffset are for the most part intended to be used with date_range and its just a coincidence that some of them work with Period.

• DOC: Unclear documentation for Period Construction #4591 Unclear documentation for Period Construction
• Missing Periods for some DateOffsets #5091 Missing Periods for some DateOffsets
• No str for FY5253Quarter #12379 No str for FY5253Quarter
• Issues Creating Period with DateOffset as freq #4878 Issues Creating Period with DateOffset as freq, e.g. pd.Period('2013-12', freq='pd.offsets.BusinessMonthEnd()) raises

Unanchored Offsets

• The docstrings/comments don't actually say what "isAnchored()` is supposed to mean. There is some explanation https://pandas.pydata.org/pandas-docs/stable/timeseries.html#anchored-offsets
• Feature request: Add annual offset that is not anchored #15685 Request for unanchored YearOffset
• YearOffset() raises AttributeError #7706 YearOffset() raises AttributeError
• BUG: seems that subtracting a YearBegin offset from an index loses the months/days #4804 BUG: seems that subtracting a YearBegin offset from an index loses the months/days

Quarter start/end offset confusion

• DOC: Anchored Offset Starting Quarter and Year defaults to JAN not DEC #17787 Anchored offset Starting Quarter and Year defaults to JAN not DEC
• Default for "Q" is not the same default for QuarterEnd #2885 Default for "Q" is not the same as default for "QuarterEnd"
• Parameter Name of startingMonth for QuarterEnd is Misleading #5307 Parameter Name of startingMonth for QuarterEnd is Misleading
• Quarter.onOffset looks fishy #18235 Quarter.onOffset looks fishy
• DEPR: QuarterBegin and BQuarterBegin return days that are not quarter beginnings #8435 DEPR: QuarterBegin and BQuarterBegin return days that are not quarter beginnings

Timezone issues

• Adding DateOffset(days=1) produces NonExistentTimeError #12156 Adding DateOffset(days=1) produces NonExistentTimeError, given example is pd.date_range(pd.Timestamp('2015-1-1', tz='US/Eastern'), pd.Timestamp('2016-1-1', tz='US/Eastern'), freq='H') + pd.DateOffset(days=1)
• API: Wrong DateOffset behaviour with DST changes #16980 API: Wrong DateOffset behaviour with DST changes. I think the OP wants Day to behave like relativedelta(days=1) rather than timedelta(hours=24)
• Lots of errors surfaced when running tests with hypothesis in [WIP] implement tests using hypothesis #18761

RelativeDelta issues. Some of these may be solved or affected by #18329, #18226

• DOC: hour/hours (and other plurals) based on dateutil.relativedelta #7418 BUG: hour/hours (and other plurals) should be the same as singulars. DateOffset behavior here just passes through to dateutil.relativedelta, so the issue seems to be confusingly similar kwarg options.
• MonthOffset() gives increase of 1 day instead of 1 month #7707 MonthOffset() gives increase of 1 day instead of 1 month (This raises NotImplementedError)
• BUG: seems that subtracting a YearBegin offset from an index loses the months/days #4804 goes here too

Perf

• DateOffset.__eq__ (called from Period.__eq__) is very slow because DateOffset._params is very slow. This could be improved a lot if DateOffset were immutable. Attempts to do this so far have run into pickle problems, see Tslibs offsets immutable #18224. Assistance requested.

Possible Bugs, needs confirmation:

• Week.onOffset looks fishy #18510 Week(weekday=None).onOffset will always return False, while I think it should always be True. (Closed by Fix bugs in WeekOfMonth.apply, Week.onOffset #18875)
• A bunch of offsets fail to satisfy offset.onOffset(ts) \Leftrigharrow (ts + offset) - offset == ts. I think the latter is effectively the definition of onOffset in the general case, need confirmation that this is a bug.

Misc

• BUG: Cannot add non-vectorized DateOffset to empty DatetimeIndex #12724 BUG: Cannot add non-vectorized DateOffset to empty DatetimeIndex
• normalize should be accounted for in __eq__ (BUG: offsets normalize and __eq__ #17689) (closed by fix DateOffset eq to depend on normalize attr #21404)
• Do Tick classes with normalize=True make sense? (Tick with normalize=True should be disallowed #21434)
• CacheableOffset mixin is never used
• liboffsets.WeekDay is not used outside of tests
• DateOffset._should_cache is never true
• DateOffset.isAnchored needs a docstring
• I screwed up a while back confusing BusinessMixin.offset with DateOffset._offset. No real harm done, but should be reverted. Might be worth finding a different name than offset
• BusinessHourMixin.apply raises an ApplyTypeError with an incomplete error message
• remove camelCase
• WeekOfMonth._from_name has a comment # TODO: handle n here..., same with LastWeekOfMonth
• BQuarterBegin has a comment # I suspect this is wrong for *all* of them, could really use some clarification.
• YearOffset._get_offset_day has a comment suggesting a more performant implementation may be available.
• FY5253.apply has a couple of assert False statements for cases which presumably we don't expect to get hit. These should probably raise with a useful error message (or better yet, confirm that they will never get reached.
• Does Tick need its own apply_index?
• Easter.apply handles n==0 differently from all others.
• Not all DateOffsets are comparable #8386 missing comparison methods (Implement missing offset comparison methods #18738 should close this)