DataFrame vs Series vs Index arithmetic Roundup

[jbrockmendel]

Categorical:

• Categorical Arithmetic/Comparison Rules #19513 Comparisons between Categorical, CategoricalIndex, and Series[category] are inconsistent
• API: comparisons of categorical data and (scalar or list-like) #8995 API: comparisons of categorical data and (scalar or list-like)

Series-specific:

• Series has no __rdivmod__ #19252 Series has no __rdivmod__ (closed by Implement Series.__rdivmod__, un-xfail tests #23271)
• Index division by zero not filled #19322 Series division by zero does not get filled for some dtypes
• BUG/API Series/Index integer floordiv/mod inconsistency #13843 BUG/API Series/Index integer floordiv/mod inconsistency

DataFrame

• BUG: comparisons fail for NaT in DataFrame #15697 pd.DataFrame([pd.NaT]).eq(pd.NaT) returns NaN instead of bool. == is OK. (closed by dispatch scalar DataFrame ops to Series #22163)
• Datetime comparisons raise TypeError #13128 DataFrame.__eq__ is raising instead of returning bool (closed by dispatch scalar DataFrame ops to Series #22163)
• BUG: datetimelike subtract incorrect when broadcasting #12437 DataFrame.sub broadcasting problem (closed by (I think) dispatch scalar DataFrame ops to Series #22163)
• Inconsistent Timestamp arithmetic #8554 DataFrame[datetime64] - datetime is returning DataFrame[datetime64] instead of DataFrame[timedelta64] (closed by dispatch scalar DataFrame ops to Series #22163)
• BUG: Timestamp comparisons to dataframe should raise if there are mismatched types #8932 pd.Timestamp('2000-01-01') > pd.DataFrame({'x': range(5)}) returns 5 Trues instead of raising (closed by dispatch scalar DataFrame ops to Series #22163)
• BUG: comparing multicolumn dataframe with datetime64 values to series gives TypeError #9006 datetime64 comparisons raise incorrectly
• Unexpected exception on column with NaT #17559 DataFrame[datetime64] != pd.Series([pd.NaT]) raises (closed by (I think) dispatch scalar DataFrame ops to Series #22163)
• DataFrame[timedelta64] / timedelta64 or pydatetime has wrong dtype and wrong values #20088 DataFrame[timedelta64] / timedeltalike (closed by dispatch scalar DataFrame ops to Series #22163)
• dtype changed when DataFrame add DateOffset #21610 dtype changed when DataFrame add DateOffset (closed by dispatch scalar DataFrame ops to Series #22163)
• DataFrame(Timestamp) add Timedelta fails #22005 DataFrame(Timestamp) add Timedelta fails (closed by dispatch scalar DataFrame ops to Series #22163)
• DataFrame[datetime64].__sub__ non-nano datetime64 fails #18874 DataFrame[datetime64].sub non-nano datetime64 fails (closed by dispatch scalar DataFrame ops to Series #22163)
• Timestamp comparison inconsistency #22017 Timestamp comparison inconsistency (closed by dispatch scalar DataFrame ops to Series #22163)

Bool/Bitwise Methods:

• BUG: bool methods are wrong with DataFrame and Series (or at least wrong error) #5035 BUG: bool methods are wrong with DataFrame and Series (or at least wrong error)
• ERR: Remove bitwise operations on dtype=object? #16873 ERR: Remove bitwise operations on dtype=object?
• API: Series/DataFrame logical op NaN propagation #13896 API: Series/DataFrame logical op NaN propagation

Unsorted:

• DatetimeIndex + ndarray[int] wrong, reverse op errors #19123 DatetimeIndex + ndarray[int] wrong, reverse op errors (TODO: track down what PR fixed it)
• BUG: tz-aware datetime with column-wise comparisions failing with np.minmum/maximum #15552 BUG: tz-aware datetime with column-wise comparisions failing
• BUG: binary comparison of numpy.int/float and Series #9369 BUG: binary comparison of numpy.int/float and Series
• Inconsistent comparisons between Series and 2-D ndarray #4615 Inconsistent comparisons between Series and 2-D ndarray
• Series[Period] +/- NaT returns Series[datetime64] #19389 Series[Period] +/- NaT returns Series[datetime64]
• np.ndarray[object] - Timedelta raises #21980 np.ndarray[object] - Timedelta raises (closed by BUG: fix raise of TypeError when subtracting timedelta array #22054)
• DTI/TDI comparison ops with incompatible types #19804 DTI/TDI comparison ops with incompatible types
• BUG: Mismatched Comparisons Inconsistencies #22042 BUG: Mismatched Comparisons Inconsistencies
• Breaking change multiplication bool and int #22047 Breaking change multiplication bool and int (closed by dispatch scalar DataFrame ops to Series #22163)
• Equality comparison raises exception #7830 Equality comparison raises exception (closed by [Bug] Fix various DatetimeIndex comparison bugs #22074)
• Comparing Strings to Numbers: No ValueError #11565 Comparing Strings to Numbers: No ValueError
• Comparing DataFrame with columns with mixed types to a scalar should fail #20876 Comparing DataFrame with columns with mixed types to a scalar should fail (closed by dispatch scalar DataFrame ops to Series #22163)
• API: Index/Series/DataFrame op 1-d list-like coercion #13637 API: Index/Series/DataFrame op 1-d list-like coercion
• ENH: Period ops NaT & timedelta ops #5202 ENH: Period ops NaT & timedelta ops
• BUG: None is not equal to None #20442 BUG: None is not equal to None
• BUG: PeriodIndex comparisons break on listlike #21793 BUG: PeriodIndex comparisons break on listlike
• ERR: invalid error reporting when comparing vs. None #12689 ERR: Invalid error reporting when comparing vs None
• BUG: incorrect casting in ndarray + DataFrame #22537 BUG: incorrect casting in ndarray + DataFrame (closed by BUG: fix DataFrame+DataFrame op with timedelta64 dtype #22696)
• BUG: bool methods are wrong with DataFrame and Series (or at least wrong error) #5035 BUG: bool methods are wrong with DataFrame and Series (or at least wrong error)
• Binary operators between DataFrame and Series object doesn't seem to work #5284 Binary operators between DataFrame and Series object doesn't seem to work
• Mixed-dtype dataframe comparison with array raises incorrectly #20090 Mixed-dtype dataframe comparison with array raises incorrectly (closed by Use align_method in comp_method_FRAME #23132)

TimedeltaIndex vs Series[timedelta64]

• Series[timedelta64] // timedelta raises TypeError. See others in BUG: fix Series[timedelta64] arithmetic with Timedelta scalars #18831. (Closed by BUG: fix Series[timedelta64] arithmetic with Timedelta scalars #18831)
• Series[timedelta64] +/- TimedeltaIndex gets cast to int64 BUG:: Series of timedelta + TimedeltaIndex gets casted to int64 #17250. Update: This appears to affect 0.22.0 but no longer affect master as of 2018-01-02. Needs a test. (closed by Tests for TDI issues already fixed #19044)
• pd.Timedelta('3 hours') / pd.Series([pd.Timedelta('1 day')]) returns 0 days, not 0.125 (closed by BUG: fix Series[timedelta64] arithmetic with Timedelta scalars #18831)
• TimedeltaIndex * Series[int] returns TimedeltaIndex Index * Series returns Index #19042 (closed by Fix Index mul/div ops with Series, closes #19080, #19042 #19253)
• Series[timedelta64] + TimedeltaIndex inherits name from the latter. This does not occur in 0.22.0, occurs in master as of 2018-01-02. Series[timedelta64] +/- TimedeltaIndex inherits name from TimedeltaIndex #19043 (closed by Tests for TDI issues already fixed #19044)
• TimedeltaIndex + Series[timedelta64] raises ValueError (TODO: issue?) (Fixed by Fix Series[timedelta64]+DatetimeIndex[tz] bugs #18884 but needs test) No original issue, no longer re-producable.
• TimedeltaIndex + Series[offsets] gets transposed and returns Series of TimedeltaIndexes. Analogue to BUG: DatetimeIndex + arraylike of DateOffsets #18849 (closed by Fix Series[timedelta64]+DatetimeIndex[tz] bugs #18884 (I think))
• TimedeltaIndex +/- zero-dimensional integer arrays DatetimeIndex/TimedeltaIndex add/sub zero-dim arrays incorrect #19012 (closed by datetimelike indexes add/sub zero-dim integer arrays  #19013)

DatetimeIndex vs Series[datetime64]

• Series[offsets] + Series[datetime64] raises Series[offsets] + Series[datetime64] fails #19211 (closed sometime before 4ebdc50)
• Series.__sub__ incorrect with non-nano datetime64 BUG: coercion of non-M8[ns] in datetime ops #7996 (closed by Fix Series.__sub__ non-nano datetime64 #18783)
• Series[datetime64] - pd.NaT returns datetime64-dtype, DatetimeIndex - pd.NaT returns timedelta64-dtype. Series.__sub__(NaT) vs DatetimeIndex.__sub__(NaT) #18808, diff of datetime column with NaT does not result into TimeDelta  #17837. (closed by dispatch Series[datetime64] ops to DatetimeIndex #19024)
• Series +/- Series[offsets] applies element-wise. DatetimeIndex +/- box([offsets]) raises for box in [np.array, pd.Index] and for box == Series returns a Series of DatetimeIndex objects. (Closed by BUG: DatetimeIndex + arraylike of DateOffsets #18849)
• Series[datetime64tz] - DatetimeIndex[tz] (with same tz) raises, same with reverse op Series[datetime64tz] - DatetimeIndex[tz] raises ValueError #19071. (Fixed by Fix Series[timedelta64]+DatetimeIndex[tz] bugs #18884 but needs test) (closed by Tests for fixed issues in Series[datetime64] #19072)
• DatetimeIndex[tz] + Series[timedelta64] returns DatetimeIndex (?). Example from DatetimeIndex + TimeDelta gives wrong results when timezone is set #13905 but the behavior seems to have changed since then. As of late-Dec 2017 behavior is as in DatetimeIndex + Series[timedelta64] returns DatetimeIndex #18963. (Closed by Fix Series[timedelta64]+DatetimeIndex[tz] bugs #18884)
• Series is missing overflow checks Negative timedeltas for deltas > 292 years #12534 (Note: this is not specific to datetime/timedelta. Even int64 is missing overflow checks) (closed by Tests for fixed issues in Series[datetime64] #19072 (but only for datetime dtypes)
• Series[datetime64] - PeriodIndex is allowed, DatetimeIndex - PeriodIndex is not. Series[datetime64] - PeriodIndex vs DatetimeIndex - PeriodIndex #18850. (closed by dispatch Series[datetime64] ops to DatetimeIndex #19024)
• DatetimeIndex +/- zero-dimensional integer arrays DatetimeIndex/TimedeltaIndex add/sub zero-dim arrays incorrect #19012 (closed by datetimelike indexes add/sub zero-dim integer arrays  #19013)

Timedelta (strictly speaking out of scope)

• pd.Timedelta(minutes=5, seconds=4) // np.timedelta64('NaT') --> -1 instead of NaN, also messes up for timedelta / Series([Timedelta, NaT]) (closed by Fix Timedelta.__floordiv__, __rfloordiv__ #18961)

[nmusolino]

I was asked to comment here about issue #20088.

Issue #20088 is closest in spirit to #8554, in the sense that both bugs involve arithmetic operations (subtraction and division, respectively) for dataframes containing timedelta values.

That said, the bugs are somewhat distinct... paraphrasing, #8554 is roughly this: DataFrame[date] - date should give timedeltas, whereas #20088 is roughly this: DataFrame[timedelta64] / timedelta64 should give float.

I suggest you re-open #20088 and add it to the list of "DataFrame" issues above.

[jreback]

none of the issues here need to be reopened this is a master tracking issue and they are referenced

[jbrockmendel]

Down to 3 checkboxes, none of which are actionable. Closing.