Skip to content

Fix error value_counts result with pyarrow categorical columns #60949

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Feb 19, 2025
Merged

Fix error value_counts result with pyarrow categorical columns #60949

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
1e4e2f4
BUG: Fix PyArrow array access in Categorical constructor for Index ob…
chilin0525 Feb 17, 2025
c9282de
TST: Add test for value_counts with Arrow dictionary dtype (#60563)
chilin0525 Feb 17, 2025
a4ad466
DOC: Add changelog entry for PyArrow array access fix in Categorical …
chilin0525 Feb 17, 2025
4f79b07
Update doc/source/whatsnew/v3.0.0.rst
chilin0525 Feb 19, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions doc/source/whatsnew/v3.0.0.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -782,6 +782,7 @@ Sparse

ExtensionArray
^^^^^^^^^^^^^^
- Bug in :class:`Categorical` when constructing with an :class:`Index` with :class:`ArrowDtype` (:issue:`60563`)
- Bug in :meth:`.arrays.ArrowExtensionArray.__setitem__` which caused wrong behavior when using an integer array with repeated values as a key (:issue:`58530`)
- Bug in :meth:`api.types.is_datetime64_any_dtype` where a custom :class:`ExtensionDtype` would return ``False`` for array-likes (:issue:`57055`)
- Bug in comparison between object with :class:`ArrowDtype` and incompatible-dtyped (e.g. string vs bool) incorrectly raising instead of returning all-``False`` (for ``==``) or all-``True`` (for ``!=``) (:issue:`59505`)
Expand Down
7 changes: 6 additions & 1 deletion pandas/core/arrays/categorical.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -447,7 +447,12 @@ def __init__(
if isinstance(values.dtype, ArrowDtype) and issubclass(
values.dtype.type, CategoricalDtypeType
):
arr = values._pa_array.combine_chunks()
from pandas import Index

if isinstance(values, Index):
arr = values._data._pa_array.combine_chunks()
else:
arr = values._pa_array.combine_chunks()
Comment on lines +450 to +455
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
from pandas import Index
if isinstance(values, Index):
arr = values._data._pa_array.combine_chunks()
else:
arr = values._pa_array.combine_chunks()
arr = values.array._pa_array.combine_chunks()

Copy link
Contributor Author

@chilin0525 chilin0525 Feb 19, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@mroeschke Thanks for review. After applied this change got failed of 2 testcase in pandas/tests/extension/test_arrow.py 🤔 (testing in my local):

Following is error message from 2 fail testcase, values.array raise error when value is ArrowExtensionArray Object:

  • test_sort_values_dictionary:

        [1/1] Generating write_version_file with a custom command
+ /usr/local/bin/ninja
============================= test session starts ==============================
platform linux -- Python 3.10.8, pytest-8.3.4, pluggy-1.5.0
PyQt5 5.15.11 -- Qt runtime 5.15.16 -- Qt compiled 5.15.14
rootdir: /home/pandas
configfile: pyproject.toml
plugins: xdist-3.6.1, hypothesis-6.125.3, qt-4.4.0, anyio-4.8.0, localserver-0.9.0.post0, cython-0.3.1, cov-6.0.0
collected 26414 items / 26413 deselected / 1 selected

pandas/tests/extension/test_arrow.py F

=================================== FAILURES ===================================
_________________________ test_sort_values_dictionary __________________________

    def test_sort_values_dictionary():
        df = pd.DataFrame(
            {
                "a": pd.Series(
                    ["x", "y"], dtype=ArrowDtype(pa.dictionary(pa.int32(), pa.string()))
                ),
                "b": [1, 2],
            },
        )
        expected = df.copy()
>       result = df.sort_values(by=["a", "b"])

pandas/tests/extension/test_arrow.py:1699: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
pandas/core/frame.py:7077: in sort_values
    indexer = lexsort_indexer(
pandas/core/sorting.py:350: in lexsort_indexer
    cat = Categorical(k, ordered=True)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <[AttributeError("'NoneType' object has no attribute 'categories'") raised in repr()] Categorical object at 0x7f5aa42d8a90>
values = <ArrowExtensionArray>
['x', 'y']
Length: 2, dtype: dictionary<values=string, indices=int32, ordered=0>[pyarrow]
categories = None, ordered = True
dtype = CategoricalDtype(categories=None, ordered=True, categories_dtype=None)
copy = True

    def __init__(
        self,
        values,
        categories=None,
        ordered=None,
        dtype: Dtype | None = None,
        copy: bool = True,
    ) -> None:
        dtype = CategoricalDtype._from_values_or_dtype(
            values, categories, ordered, dtype
        )
        # At this point, dtype is always a CategoricalDtype, but
        # we may have dtype.categories be None, and we need to
        # infer categories in a factorization step further below
    
        if not is_list_like(values):
            # GH#38433
            raise TypeError("Categorical input must be list-like")
    
        # null_mask indicates missing values we want to exclude from inference.
        # This means: only missing values in list-likes (not arrays/ndframes).
        null_mask = np.array(False)
    
        # sanitize input
        vdtype = getattr(values, "dtype", None)
        if isinstance(vdtype, CategoricalDtype):
            if dtype.categories is None:
                dtype = CategoricalDtype(values.categories, dtype.ordered)
        elif isinstance(values, range):
            from pandas.core.indexes.range import RangeIndex
    
            values = RangeIndex(values)
        elif not isinstance(values, (ABCIndex, ABCSeries, ExtensionArray)):
            values = com.convert_to_list_like(values)
            if isinstance(values, list) and len(values) == 0:
                # By convention, empty lists result in object dtype:
                values = np.array([], dtype=object)
            elif isinstance(values, np.ndarray):
                if values.ndim > 1:
                    # preempt sanitize_array from raising ValueError
                    raise NotImplementedError(
                        "> 1 ndim Categorical are not supported at this time"
                    )
                values = sanitize_array(values, None)
            else:
                # i.e. must be a list
                arr = sanitize_array(values, None)
                null_mask = isna(arr)
                if null_mask.any():
                    # We remove null values here, then below will re-insert
                    #  them, grep "full_codes"
                    arr_list = [values[idx] for idx in np.where(~null_mask)[0]]
    
                    # GH#44900 Do not cast to float if we have only missing values
                    if arr_list or arr.dtype == "object":
                        sanitize_dtype = None
                    else:
                        sanitize_dtype = arr.dtype
    
                    arr = sanitize_array(arr_list, None, dtype=sanitize_dtype)
                values = arr
    
        if dtype.categories is None:
            if isinstance(values.dtype, ArrowDtype) and issubclass(
                values.dtype.type, CategoricalDtypeType
            ):
                # from pandas import Index
    
                # if isinstance(values, Index):
                #     arr = values._data._pa_array.combine_chunks()
                # else:
                #     arr = values._pa_array.combine_chunks()
>               arr = values.array._pa_array.combine_chunks()
E               AttributeError: 'ArrowExtensionArray' object has no attribute 'array'

pandas/core/arrays/categorical.py:456: AttributeError
---------------- generated xml file: /home/pandas/test-data.xml ----------------
============================= slowest 30 durations =============================

(3 durations < 0.005s hidden.  Use -vv to show these durations.)
=========================== short test summary info ============================
FAILED pandas/tests/extension/test_arrow.py::test_sort_values_dictionary - At...
===================== 1 failed, 26413 deselected in 1.61s ======================

  • test_dictionary_astype_categorical:

        [1/1] Generating write_version_file with a custom command
+ /usr/local/bin/ninja
============================= test session starts ==============================
platform linux -- Python 3.10.8, pytest-8.3.4, pluggy-1.5.0
PyQt5 5.15.11 -- Qt runtime 5.15.16 -- Qt compiled 5.15.14
rootdir: /home/pandas
configfile: pyproject.toml
plugins: xdist-3.6.1, hypothesis-6.125.3, qt-4.4.0, anyio-4.8.0, localserver-0.9.0.post0, cython-0.3.1, cov-6.0.0
collected 26414 items / 26413 deselected / 1 selected

pandas/tests/extension/test_arrow.py F

=================================== FAILURES ===================================
______________________ test_dictionary_astype_categorical ______________________

    def test_dictionary_astype_categorical():
        # GH#56672
        arrs = [
            pa.array(np.array(["a", "x", "c", "a"])).dictionary_encode(),
            pa.array(np.array(["a", "d", "c"])).dictionary_encode(),
        ]
        ser = pd.Series(ArrowExtensionArray(pa.chunked_array(arrs)))
>       result = ser.astype("category")

pandas/tests/extension/test_arrow.py:3339: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
pandas/core/generic.py:6435: in astype
    new_data = self._mgr.astype(dtype=dtype, errors=errors)
pandas/core/internals/managers.py:588: in astype
    return self.apply("astype", dtype=dtype, errors=errors)
pandas/core/internals/managers.py:438: in apply
    applied = getattr(b, f)(**kwargs)
pandas/core/internals/blocks.py:610: in astype
    new_values = astype_array_safe(values, dtype, errors=errors)
pandas/core/dtypes/astype.py:234: in astype_array_safe
    new_values = astype_array(values, dtype, copy=copy)
pandas/core/dtypes/astype.py:176: in astype_array
    values = values.astype(dtype, copy=copy)
pandas/core/arrays/base.py:769: in astype
    return cls._from_sequence(self, dtype=dtype, copy=copy)
pandas/core/arrays/categorical.py:531: in _from_sequence
    return cls(scalars, dtype=dtype, copy=copy)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <[AttributeError("'NoneType' object has no attribute 'categories'") raised in repr()] Categorical object at 0x7feb7f621940>
values = <ArrowExtensionArray>
['a', 'x', 'c', 'a', 'a', 'd', 'c']
Length: 7, dtype: dictionary<values=string, indices=int32, ordered=0>[pyarrow]
categories = None, ordered = None
dtype = CategoricalDtype(categories=None, ordered=None, categories_dtype=None)
copy = False

    def __init__(
        self,
        values,
        categories=None,
        ordered=None,
        dtype: Dtype | None = None,
        copy: bool = True,
    ) -> None:
        dtype = CategoricalDtype._from_values_or_dtype(
            values, categories, ordered, dtype
        )
        # At this point, dtype is always a CategoricalDtype, but
        # we may have dtype.categories be None, and we need to
        # infer categories in a factorization step further below
    
        if not is_list_like(values):
            # GH#38433
            raise TypeError("Categorical input must be list-like")
    
        # null_mask indicates missing values we want to exclude from inference.
        # This means: only missing values in list-likes (not arrays/ndframes).
        null_mask = np.array(False)
    
        # sanitize input
        vdtype = getattr(values, "dtype", None)
        if isinstance(vdtype, CategoricalDtype):
            if dtype.categories is None:
                dtype = CategoricalDtype(values.categories, dtype.ordered)
        elif isinstance(values, range):
            from pandas.core.indexes.range import RangeIndex
    
            values = RangeIndex(values)
        elif not isinstance(values, (ABCIndex, ABCSeries, ExtensionArray)):
            values = com.convert_to_list_like(values)
            if isinstance(values, list) and len(values) == 0:
                # By convention, empty lists result in object dtype:
                values = np.array([], dtype=object)
            elif isinstance(values, np.ndarray):
                if values.ndim > 1:
                    # preempt sanitize_array from raising ValueError
                    raise NotImplementedError(
                        "> 1 ndim Categorical are not supported at this time"
                    )
                values = sanitize_array(values, None)
            else:
                # i.e. must be a list
                arr = sanitize_array(values, None)
                null_mask = isna(arr)
                if null_mask.any():
                    # We remove null values here, then below will re-insert
                    #  them, grep "full_codes"
                    arr_list = [values[idx] for idx in np.where(~null_mask)[0]]
    
                    # GH#44900 Do not cast to float if we have only missing values
                    if arr_list or arr.dtype == "object":
                        sanitize_dtype = None
                    else:
                        sanitize_dtype = arr.dtype
    
                    arr = sanitize_array(arr_list, None, dtype=sanitize_dtype)
                values = arr
    
        if dtype.categories is None:
            if isinstance(values.dtype, ArrowDtype) and issubclass(
                values.dtype.type, CategoricalDtypeType
            ):
                # from pandas import Index
    
                # if isinstance(values, Index):
                #     arr = values._data._pa_array.combine_chunks()
                # else:
                #     arr = values._pa_array.combine_chunks()
>               arr = values.array._pa_array.combine_chunks()
E               AttributeError: 'ArrowExtensionArray' object has no attribute 'array'

pandas/core/arrays/categorical.py:456: AttributeError
---------------- generated xml file: /home/pandas/test-data.xml ----------------
============================= slowest 30 durations =============================

(3 durations < 0.005s hidden.  Use -vv to show these durations.)
=========================== short test summary info ============================
FAILED pandas/tests/extension/test_arrow.py::test_dictionary_astype_categorical
===================== 1 failed, 26413 deselected in 1.67s ======================

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah ok. It appears something is passing in an ArrowExtensionArray here already. OK what you have here is fine then

categories = arr.dictionary.to_pandas(types_mapper=ArrowDtype)
codes = arr.indices.to_numpy()
dtype = CategoricalDtype(categories, values.dtype.pyarrow_dtype.ordered)
Expand Down
17 changes: 17 additions & 0 deletions pandas/tests/extension/test_arrow.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3511,3 +3511,20 @@ def test_map_numeric_na_action():
result = ser.map(lambda x: 42, na_action="ignore")
expected = pd.Series([42.0, 42.0, np.nan], dtype="float64")
tm.assert_series_equal(result, expected)


def test_categorical_from_arrow_dictionary():
# GH 60563
df = pd.DataFrame(
{"A": ["a1", "a2"]}, dtype=ArrowDtype(pa.dictionary(pa.int32(), pa.utf8()))
)
result = df.value_counts(dropna=False)
expected = pd.Series(
[1, 1],
index=pd.MultiIndex.from_arrays(
[pd.Index(["a1", "a2"], dtype=ArrowDtype(pa.string()), name="A")]
),
name="count",
dtype="int64",
)
tm.assert_series_equal(result, expected)
Loading