ENH: add BooleanArray extension array #29555

jorisvandenbossche · 2019-11-11T21:38:53Z

Closes #21778
xref closed PRs #22226 and #25415, and xref the discussions on missing values in #28095 and #28778

This PR adds a BooleanArray extension array ("boolean" dtype):

It uses the same mask-based approach as the IntegerArray: numpy array for the values + numpy array for the mask
it does not yet implement any new NA behaviour as being discussed DISCUSS: boolean dtype with missing value support #28778, but for now follows the current NaN behaviour
It is not yet "produced" from other extension arrays (eg a comparison method on IntegerArray, I would leave that for follow-up PRs)
There is certainly a lot of code to share between this BooleanArray and IntegerArray (eg in a "BaseMaskedArray" base class), but didn't do that yet here (focusing on the BooleanArray for now)

jorisvandenbossche · 2019-11-11T21:44:27Z

pandas/core/arrays/boolean.py

+            data[self._mask] = self._na_value
+            return data
+        else:
+            return self._data


I adapted this compared to IntegerArray to return bool dtype if there are no NaNs (this helps to enable boolean indexing automatically for this case). But we should probably make a consistent choice.

Yeah, I don't think we want to return object-dtype here if we can avoid it...

On the other hand, a predictable dtype (not depending on the values in the array) is also nice ...

Yeah, as I thought more about it the dtype stability was winning me over. Not sure what's best.

Switched back to always object dtype for now. There are methods (like astype(bool)) that force you to give bool dtype if possible. We should maybe think about adding a to_numpy to our arrays as well (the same for IntegerArray).

pandas/core/arrays/boolean.py

jorisvandenbossche · 2019-11-11T21:52:32Z

pandas/tests/arrays/test_boolean.py

+# def test_to_boolean_array_error(values):
+#     # error in converting existing arrays to BooleanArray
+#     with pytest.raises(TypeError):
+#         pd.array(values, dtype="boolean")


Currently those actually work, as the coercion routine is using numpy coercion: np.array(['a'], dtype=bool) actually works and use bool-ness of the values.
So the question here is if we want to allow this, or rather be more strict (eg only allow actual bools + integer 0/1)? (personally prefer to be strict)

Mmm, as long as actual bools and integer 0/1 as accepted then either works for me.

Would be nice to have a decision here, and get tests for it. I think we can justify either, since np.array([np.nan], dtype=bool) will be array([True]). So we're already deviating from NumPy in this limited case.

Do you have a plan for how to do the actual validation? Will it be somewhat expensive?

What are we allowing in __setitem__? Do we convert the value to bool, or do we require booleans?

So an option could be to allow: bool + int/float of 0's and 1's (the float is needed in case it are 0/1's with np.nan), and raise for the rest.

Do you have a plan for how to do the actual validation? Will it be somewhat expensive?

We can check for actual bool/int/float dtypes (and then for int and float check that the original is equal to the result converted back to numeric; to ensure it were only 0's and 1's). For object we can first do an infer_dtype, and then apply the same logic.

What are we allowing in setitem? Do we convert the value to bool, or do we require booleans?

Currently, we use the same coerce_to_array function in __setitem__ to convert the value(s) to be set to boolean values+mask. So if we restrict like the above, then that follows the same pattern for __setitem__ (but we could also be more strict in __setitem__ if that is preferred)

jorisvandenbossche · 2019-11-11T21:53:04Z

pandas/tests/arrays/test_boolean.py

+
+
+# @pytest.mark.parametrize("ufunc", [np.add])
+# def test_ufuncs_binary(ufunc):


still need to implement arithmetic ops

What are the expected return types? Would you use IntegerArray for things returning an integer dtype?

ndarray[bool] will sometimes return a float dtype...

In [17]: a = np.array([True, False]) In [18]: a / 1 Out[18]: array([1., 0.])

would we return an ndarray there, with NaN for the missing values?

I guess returning a PandasArray is also an option... If we want to stay in the algebra of op(BooleanArray, other) returning an ExtensionArray

For now I went with tthe dtype that numpy ops result in, and then if result is bool -> BooleanArray, if is int -> IntegerArray, otherwise if result is float dtype (or something else) return the numpy array.

jbrockmendel · 2019-11-11T22:46:54Z

should this be marked as a WIP?

jreback

is there a reason you did not inherit from IntegeryArray (and dtype) and just copied

surely these share much more code and could easily share a bae class at the very least

jorisvandenbossche · 2019-11-12T08:11:15Z

is there a reason you did not inherit from IntegerArray (and dtype) and just copied

surely these share much more code and could easily share a bae class at the very least

As I said in the top post: yes this needs to share code with IntegerArray through common base class, but initially focussed on a working BooleanArray and deciding on its behaviour.
I am happy (and planning) to do that, here or in a follow-up.

I don't think it should directly inherit IntegerArray, but personally prefer a common base class. I think that will make it clearer what is actually shared.

jreback · 2019-11-12T08:31:38Z

is there a reason you did not inherit from IntegerArray (and dtype) and just copied
surely these share much more code and could easily share a bae class at the very least

As I said in the top post: yes this needs to share code with IntegerArray through common base class, but initially focussed on a working BooleanArray and deciding on its behaviour.
I am happy (and planning) to do that, here or in a follow-up.

I don't think it should directly inherit IntegerArray, but personally prefer a common base class. I think that will make it clearer what is actually shared.

i agree a common base class is good and as a follow up is fine

jorisvandenbossche · 2019-11-12T09:31:05Z

should this be marked as a WIP?

@jbrockmendel not in the sense of "this is ready to be reviewed", and it also passing all extension array tests (except one, see below).
There are certainly still todo's (eg the arithmetic ops and ufunc machinery), but eg some commented out tests are mainly API questions (what behaviour do we want? for which review is welcome). There are also parts that I am not planning to address in this PR (eg adding a scalar pd.NA) but rather in separate PRs.

The one failure in the base extension array tests, is for concatting mixed dtypes (result = pd.concat([df1, df2])):

(Pdb) df1
       A
0   True
1  False
2   True
(Pdb) df2
   A
0  1
1  2
2  3
(Pdb) result
   A
0  1
1  0
2  1
0  1
1  2
2  3
(Pdb) expected
       A
0   True
1  False
2   True
0      1
1      2
2      3

So it is expecting to return object dtype.
However, on master for numpy dtypes, we also coerce booleans to integer:

In [23]: df1 = pd.DataFrame({'A': [True, False]})  

In [24]: df2 = pd.DataFrame({'A': [1, 2]})  

In [25]: pd.concat([df1, df2])    
Out[25]: 
   A
0  1
1  0
0  1
1  2

What to follow here? (we should probably have some discussion about clearer rules (and document this) for which dtypes coerce with which others. Or does this exist somewhere?)

jbrockmendel · 2019-11-12T16:51:21Z

not in the sense of "this is ready to be reviewed",

I'm confused by a potential double-negative. If this is not ready to be reviewed, then I suggest it be marked as WIP so that would-be reviewers focus elsewhere for the time being.

jorisvandenbossche · 2019-11-12T16:54:39Z

Let me be clearer then: this is ready to be reviewed

jorisvandenbossche · 2019-11-13T22:11:28Z

Are people OK with first reviewing / getting merged this version of a BooleanArray without using the new NA, and only in a follow-up adapt it to use NA ?
(I prefer that to be able to work on those things in parallel, many aspects of BooleanArray do not involve NA vs np.nan, and those an already be reviewed here).

jreback · 2019-11-13T22:17:27Z

yes let’s keep BooleanArray independent and merge first

pandas/core/arrays/boolean.py

TomAugspurger · 2019-11-13T22:33:13Z

pandas/core/arrays/boolean.py

+            else:
+                mask = mask_values
+        else:
+            mask = np.asarray(mask, dtype=bool)


hmm when is this case reached? I'm a little confused about the mask | mask_values below.

In principle, you can pass a mask manually to this function (and in that case it gets combined with the potential mask of the values). However, this functionality is actually not used right now by BooleanArray.

I added tests for this (as I think we want to expose this publicly at some point)

pandas/core/arrays/boolean.py

jorisvandenbossche · 2019-11-14T08:05:58Z

Can you check: does this handle indexing a BooleanArray with another BooleanArray well? Does it have to go through object dtype?

@TomAugspurger boolean indexing does not yet work. Well, it did work when BooleanArray coerced to a bool ndarray, and then it worked automatically (for the case that there are no NAs).
But now I changed back to always convert to object ndarray, we will need to handle this case specifically. Since that will probably involve adding a check and proper conversion to numpy array in several places in the indexing code, I was planning on doing that later / separate PR.

TomAugspurger · 2019-11-14T19:27:57Z

Since that will probably involve adding a check and proper conversion to numpy array in several places in the indexing code, I was planning on doing that later / separate PR.

SGTM. Have we discussed the API on that anywhere? Do missing values propagate, or are they treated as False?

pandas/core/dtypes/missing.py

pandas/core/arrays/boolean.py

jorisvandenbossche · 2019-11-14T19:33:54Z

Have we discussed the API on that anywhere? Do missing values propagate, or are they treated as False?

That's where I want feedback, see the discussion in #28778

jbrockmendel · 2019-11-14T19:34:30Z

pandas/core/arrays/boolean.py

+        # may need to fill infs
+        # and mask wraparound
+        if is_float_dtype(result):
+            mask |= (result == np.inf) | (result == -np.inf)


This looks like it is copied from IntegerArray, and should not be an inplace operation. It risks alterting self._mask inplace. xref #27829

This looks like it is copied from IntegerArray, and should not be an inplace operation. It risks alterting self._mask inplace. xref #27829

OK, I removed this here altogether, as I don't think we should fill any infs in the BooleanArray division (we should separately also look at this for IntegerArray to see what behaviour we want)

This means that for BooleanArray I follow numpy's behaviour in returning float with inf:

In [19]: True / np.array([True, False, True]) /home/joris/miniconda3/envs/dev/bin/ipython:1: RuntimeWarning: divide by zero encountered in true_divide #!/home/joris/miniconda3/envs/dev/bin/python Out[19]: array([ 1., inf, 1.])

jorisvandenbossche · 2019-11-20T14:51:10Z

@jreback @TomAugspurger OK, made some substantial changes, recommend to look at the diff of the last commits: https://github.com/pandas-dev/pandas/pull/29555/files/a3e1e931392edd2631c3731a081e4eb16128c80a..90558d696c5407523e9fd0e8204c0a2ac8950653

pandas/core/arrays/boolean.py

jreback · 2019-11-20T15:47:14Z

have u tested this is indexing? i assume that would be a follow up

jorisvandenbossche · 2019-11-20T15:50:46Z

have u tested this is indexing? i assume that would be a follow up

Yep, see #29555 (comment)

jorisvandenbossche · 2019-11-20T15:56:15Z

So to gather the follow-ups:

Use pd.NA instead of np.nan for missing value indicator + update the logical ops to use the pd.NA behaviour
Reduce duplication with IntegerArray (implement a BaseMaskedArray base class or something like that). This might depend on the decision on whether we already want to use pd.NA in IntegerArray as well or not.
Update the behaviour of any/all reductions with skipna=False (API: any/all in context of boolean dtype with missing values #29686)
Enable boolean indexing with BooleanArray (we can start with handling the case of no NAs; the behaviour for with NAs still need to be decided: DISCUSS: boolean dtype with missing value support #28778)

(will open an issue for this)

jorisvandenbossche · 2019-11-21T10:54:37Z

Are you all fine with almost merging this? (or maybe a final round of review?)
Merging it would make it easier to start working on the follow-ups

jreback · 2019-11-21T11:26:59Z

let me look one more round

jorisvandenbossche · 2019-11-21T11:49:01Z

Other question: what do we want pd.array([..]) to return by default (when not specifying a dtype)?

Currently, you get this behavour:

In [9]: pd.array([True, False]) 
Out[9]: 
<PandasArray>
[True, False]
Length: 2, dtype: bool

In [10]: pd.array([True, False, np.nan]) 
Out[10]: 
<PandasArray>
[1.0, 0.0, nan]
Length: 3, dtype: float64

In [12]: pd.array([True, False, pd.NA])
Out[12]: 
<PandasArray>
[True, False, NA]
Length: 3, dtype: object

Do we want to return BooleanArray also when there are no missing values (which currently returns boolean PandasArray)?
Eg for integer, we also do not return the IntegerArray (I think the "rule" now is to basically have the same default behaviour as pandas, so only infer the extensions dtypes that are used by default (datetimetz, period, interval, etc). So if we want to be consistent with that, we should not return BooleanArray.

When a pd.NA is present, that's of course something new, and we could in principle decide to already infer the new dtype in that case (although then the inference of the dtype depends on the presence of a missing value or not ..) Although I am not sure this is technically easy to do, as the infer_dtypes does not distinguish different types of NA/NaN at the moment (and this out of scope of this PR, could be a future enhancement).

jreback · 2019-11-21T12:18:04Z

I would be +1 on using all of the extension arrays (SA, IA, BA) in pa.array by default. I dont' see any downside here. I would create an issue to update infer_dtype to do this, shouldn't be very hard as that can now return IA (with an opt in flag).

jorisvandenbossche · 2019-11-21T13:03:59Z

I think our reasoning before was to keep this consistent with the main constructors (eg in Series), so that we could in principle move towards Series(..) being Series(array(..)) at some point.

TomAugspurger · 2019-11-22T14:17:55Z

Do we want to return BooleanArray also when there are no missing values (which currently returns boolean PandasArray)?

I don't think that we should have value-dependent behavior here.

On the broader issue, I'd be fine with pd.array(..., dtype=None) starting to infer these new EAs. Will make a new issue for that.

jorisvandenbossche · 2019-11-25T13:06:45Z

@jreback ping

jreback

looks fine. some questions, can be done on followup, mostly asking about test coverage. if you want to do a minor pass ok, or can merge.

doc/source/whatsnew/v1.0.0.rst

jreback · 2019-11-25T13:34:06Z

pandas/core/arrays/boolean.py

+        return True
+
+
+def coerce_to_array(values, mask=None, copy=False):


can you type these as much as possible

I am not really sure how to type those (I don't see anything in pandas._typing for "list like" ?)

pandas/core/arrays/boolean.py

jreback · 2019-11-25T13:46:27Z

pandas/core/arrays/boolean.py

+        inferred_dtype = lib.infer_dtype(values_object, skipna=True)
+        integer_like = ("floating", "integer", "mixed-integer-float")
+        if inferred_dtype not in ("boolean", "empty") + integer_like:
+            raise TypeError("Need to pass bool-like values")


tests hit here?

tests hit here?

Yes, there is a test that passes all kinds of non-boolean-like values

In general, I ran locally pytest with coverage, and there is 97% coverage for this file. The main non-covered things are some parts of the ufunc related code, and some length mismatch errors in the ops code.

pandas/core/arrays/boolean.py

jreback · 2019-11-25T13:53:33Z

pandas/core/arrays/boolean.py

+    @classmethod
+    def _from_sequence(cls, scalars, dtype=None, copy=False):
+        if dtype:
+            assert dtype == "boolean"


do we have tests for this?

do we have tests for this?

Do we need that? This is an internal assertion, that should never be raised to the user but is here to help the developer (I don't think we should add tests for those asserts). So the "test" is that this actually never occurs in the tests.

BTW, I can also leave it out (the assert). For BooleanArray, the dtype is not very useful. I think this parameter is mainly used for cases where multiple dtypes are possible per array (eg int64, int32 etc for IntegerArray)

right i think it actually should always be None here, as this is internally passed. however a user might try to pass something that is not None (so this should maybe be a ValueError)

pandas/core/arrays/boolean.py

jreback · 2019-11-25T13:56:38Z

pandas/tests/extension/test_boolean.py

+                "__rmod__",
+            ):
+                # combine keeps boolean type
+                expected = expected.astype("Int8")


is this right?

It's what numpy does, eg:

In [17]: np.array([True, False]) ** 2 Out[17]: array([1, 0], dtype=int8)

So for those ops, I just followed numpy's behaviour with boolean arrays.

jreback · 2019-11-25T13:57:50Z

pandas/tests/extension/test_boolean.py

+
+
+class TestComparisonOps(base.BaseComparisonOpsTests):
+    def check_opname(self, s, op_name, other, exc=None):


do you need these overrides here? (you are already doing it above)

Yes, it's to override that there is no exception being raised (will add a comment about that).

(above is for the arithmetic ones)

jorisvandenbossche

Thanks for the review!

doc/source/whatsnew/v1.0.0.rst

jorisvandenbossche · 2019-11-25T14:16:12Z

pandas/core/arrays/boolean.py

+        return True
+
+
+def coerce_to_array(values, mask=None, copy=False):


I am not really sure how to type those (I don't see anything in pandas._typing for "list like" ?)

pandas/core/arrays/boolean.py

jorisvandenbossche · 2019-11-25T14:19:00Z

pandas/core/arrays/boolean.py

+        inferred_dtype = lib.infer_dtype(values_object, skipna=True)
+        integer_like = ("floating", "integer", "mixed-integer-float")
+        if inferred_dtype not in ("boolean", "empty") + integer_like:
+            raise TypeError("Need to pass bool-like values")


tests hit here?

Yes, there is a test that passes all kinds of non-boolean-like values

In general, I ran locally pytest with coverage, and there is 97% coverage for this file. The main non-covered things are some parts of the ufunc related code, and some length mismatch errors in the ops code.

pandas/core/arrays/boolean.py

jorisvandenbossche · 2019-11-25T14:23:04Z

pandas/core/arrays/boolean.py

+            raise TypeError(
+                "mask should be boolean numpy array. Use "
+                "the 'array' function instead"
+            )


only when you actually coerce, not here though

Not fully sure I understand. We don't do any coercing in this __init__ constructor, and a few lines above there are isinstance checks that the input can only be boolean ndarrays.

this is something we should check (here and in IntegerArray) as if you accidently pass a non ndim==1 then it would be an error (can be a followup)

As following your comment, I already added a ndim check; I now raise an error if ndim is not 1 (on the lines below)

jorisvandenbossche · 2019-11-25T14:25:40Z

pandas/core/arrays/boolean.py

+    @classmethod
+    def _from_sequence(cls, scalars, dtype=None, copy=False):
+        if dtype:
+            assert dtype == "boolean"


do we have tests for this?

Do we need that? This is an internal assertion, that should never be raised to the user but is here to help the developer (I don't think we should add tests for those asserts). So the "test" is that this actually never occurs in the tests.

jorisvandenbossche · 2019-11-25T14:27:11Z

pandas/core/arrays/boolean.py

+    @classmethod
+    def _from_sequence(cls, scalars, dtype=None, copy=False):
+        if dtype:
+            assert dtype == "boolean"


BTW, I can also leave it out (the assert). For BooleanArray, the dtype is not very useful. I think this parameter is mainly used for cases where multiple dtypes are possible per array (eg int64, int32 etc for IntegerArray)

jorisvandenbossche · 2019-11-25T14:29:49Z

pandas/tests/extension/test_boolean.py

+                "__rmod__",
+            ):
+                # combine keeps boolean type
+                expected = expected.astype("Int8")


It's what numpy does, eg:

In [17]: np.array([True, False]) ** 2 Out[17]: array([1, 0], dtype=int8)

So for those ops, I just followed numpy's behaviour with boolean arrays.

jorisvandenbossche · 2019-11-25T14:30:43Z

pandas/tests/extension/test_boolean.py

+
+
+class TestComparisonOps(base.BaseComparisonOpsTests):
+    def check_opname(self, s, op_name, other, exc=None):


Yes, it's to override that there is no exception being raised (will add a comment about that).

(above is for the arithmetic ones)

jreback · 2019-11-25T15:50:04Z

thanks @jorisvandenbossche other things can be done as followups.

jbrockmendel · 2019-11-25T16:26:14Z

@jorisvandenbossche for those of us who didn't follow the thread in real-time, can you summarize the design decisions that got made in the end? and any decisions that remain un-decided?

…ndexing-1row-df * upstream/master: (185 commits) ENH: add BooleanArray extension array (pandas-dev#29555) DOC: Add link to dev calendar and meeting notes (pandas-dev#29737) ENH: Add built-in function for Styler to format the text displayed for missing values (pandas-dev#29118) DEPR: remove statsmodels/seaborn compat shims (pandas-dev#29822) DEPR: remove Index.summary (pandas-dev#29807) DEPR: passing an int to read_excel use_cols (pandas-dev#29795) STY: fstrings in io.pytables (pandas-dev#29758) BUG: Fix melt with mixed int/str columns (pandas-dev#29792) TST: add test for ffill/bfill for non unique multilevel (pandas-dev#29763) Changed description of parse_dates in read_excel(). (pandas-dev#29796) BUG: pivot_table not returning correct type when margin=True and aggfunc='mean' (pandas-dev#28248) REF: Create _lib/window directory (pandas-dev#29817) Fixed small mistake (pandas-dev#29815) minor cleanups (pandas-dev#29798) DEPR: enforce deprecations in core.internals (pandas-dev#29723) add test for unused level raises KeyError (pandas-dev#29760) Add documentation linking to sqlalchemy (pandas-dev#29373) io/parsers: ensure decimal is str on PythonParser (pandas-dev#29743) Reenabled no-unused-function (pandas-dev#29767) CLN:F-string in pandas/_libs/tslibs/*.pyx (pandas-dev#29775) ... # Conflicts: # pandas/tests/frame/indexing/test_indexing.py

TomAugspurger · 2019-11-25T17:01:15Z

@jbrockmendel started at #29556 (comment)

jorisvandenbossche · 2019-11-25T17:23:14Z

@jbrockmendel this PR has no NA-related design decisions (it's just adding a BooleanArray that has not yet any specific NA behaviour). All discussions are still in the respective issues (#28095, #28778) and in the open PR to add NA (#29597), and will come in a future PR to add NA to BooleanArray.

jorisvandenbossche · 2019-11-25T17:46:30Z

@jreback opened two follow-up issues related to improving the conversion / astype (which currently goes through object dtype where it is not always needed): #29838 and #29839

ENH: add BooleanArray extension array

640dac9

jorisvandenbossche added the ExtensionArray Extending pandas with custom dtypes or arrays. label Nov 11, 2019

jorisvandenbossche commented Nov 11, 2019

View reviewed changes

TomAugspurger mentioned this pull request Nov 11, 2019

Missing values proposal: concrete steps for 1.0 #29556

Closed

13 tasks

jreback requested changes Nov 12, 2019

View reviewed changes

jorisvandenbossche added this to the 1.0 milestone Nov 12, 2019

jorisvandenbossche added 2 commits November 12, 2019 13:53

enable arithmetic ops + ufuncs

b9597bb

switch back to object dtype for __array__ + astype tests

fa77b7a

jorisvandenbossche mentioned this pull request Nov 12, 2019

ROADMAP: Consistent missing value handling with new NA scalar #28095

Open

temp

29415a9

TomAugspurger reviewed Nov 13, 2019

View reviewed changes

jbrockmendel reviewed Nov 14, 2019

View reviewed changes

pandas/core/dtypes/missing.py Show resolved Hide resolved

jbrockmendel reviewed Nov 14, 2019

View reviewed changes

pandas/core/arrays/boolean.py Outdated Show resolved Hide resolved

jbrockmendel reviewed Nov 14, 2019

View reviewed changes

jorisvandenbossche added 5 commits November 14, 2019 20:45

Merge remote-tracking branch 'upstream/master' into boolean-EA

c4a53f2

updates for feedback + add BooleanArray docstring

b1182bc

Merge remote-tracking branch 'upstream/master' into boolean-EA

94c5a90

try fix test for old numpy

1861602

fix in place modification of mask / follow numpy for division

ad6c477

TomAugspurger reviewed Nov 20, 2019

View reviewed changes

pandas/core/arrays/boolean.py Show resolved Hide resolved

jreback approved these changes Nov 25, 2019

View reviewed changes

Merge remote-tracking branch 'upstream/master' into boolean-EA

af82754

jorisvandenbossche commented Nov 25, 2019

View reviewed changes

small edits

0eb3ca2

jreback merged commit 7d7f885 into pandas-dev:master Nov 25, 2019

jorisvandenbossche deleted the boolean-EA branch November 25, 2019 17:15

TomAugspurger pushed a commit to TomAugspurger/pandas that referenced this pull request Nov 25, 2019

ENH: add BooleanArray extension array (pandas-dev#29555)

1a32d19

TomAugspurger pushed a commit to TomAugspurger/pandas that referenced this pull request Nov 25, 2019

ENH: add BooleanArray extension array (pandas-dev#29555)

bb904cb

jorisvandenbossche mentioned this pull request Dec 2, 2019

Use new NA scalar in BooleanArray #29961

Merged

proost pushed a commit to proost/pandas that referenced this pull request Dec 19, 2019

ENH: add BooleanArray extension array (pandas-dev#29555)

2713cc6

proost pushed a commit to proost/pandas that referenced this pull request Dec 19, 2019

ENH: add BooleanArray extension array (pandas-dev#29555)

abaef60

jorisvandenbossche mentioned this pull request Dec 3, 2021

CLN: TODOs #44733

Merged

4 tasks



		# @pytest.mark.parametrize("ufunc", [np.add])
		# def test_ufuncs_binary(ufunc):

		return True


		def coerce_to_array(values, mask=None, copy=False):



		class TestComparisonOps(base.BaseComparisonOpsTests):
		def check_opname(self, s, op_name, other, exc=None):

ENH: add BooleanArray extension array #29555

ENH: add BooleanArray extension array #29555

Conversation

jorisvandenbossche commented Nov 11, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jbrockmendel commented Nov 11, 2019

jreback left a comment

Choose a reason for hiding this comment

jorisvandenbossche commented Nov 12, 2019

jreback commented Nov 12, 2019

jorisvandenbossche commented Nov 12, 2019

jbrockmendel commented Nov 12, 2019

jorisvandenbossche commented Nov 12, 2019

jorisvandenbossche commented Nov 13, 2019

jreback commented Nov 13, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jorisvandenbossche commented Nov 14, 2019

TomAugspurger commented Nov 14, 2019

jorisvandenbossche commented Nov 14, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jorisvandenbossche commented Nov 20, 2019

jreback commented Nov 20, 2019

jorisvandenbossche commented Nov 20, 2019

jorisvandenbossche commented Nov 20, 2019

jorisvandenbossche commented Nov 21, 2019

jreback commented Nov 21, 2019

jorisvandenbossche commented Nov 21, 2019

jreback commented Nov 21, 2019

jorisvandenbossche commented Nov 21, 2019

TomAugspurger commented Nov 22, 2019

jorisvandenbossche commented Nov 25, 2019

jreback left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jorisvandenbossche left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jreback commented Nov 25, 2019

jbrockmendel commented Nov 25, 2019

TomAugspurger commented Nov 25, 2019

jorisvandenbossche commented Nov 25, 2019

jorisvandenbossche commented Nov 25, 2019