ENH: Implement interpolation for arrow and masked dtypes (#56757)

* ENH: Implement interpolation for arrow and masked dtypes

* Fixup

* Fix typing

* Update

Author: phofl

doc/source/whatsnew/v2.2.0.rst

@@ -343,6 +343,7 @@ Other enhancements
 - :meth:`ExtensionArray.duplicated` added to allow extension type implementations of the ``duplicated`` method (:issue:`55255`)
 - :meth:`Series.ffill`, :meth:`Series.bfill`, :meth:`DataFrame.ffill`, and :meth:`DataFrame.bfill` have gained the argument ``limit_area``; 3rd party :class:`.ExtensionArray` authors need to add this argument to the method ``_pad_or_backfill`` (:issue:`56492`)
 - Allow passing ``read_only``, ``data_only`` and ``keep_links`` arguments to openpyxl using ``engine_kwargs`` of :func:`read_excel` (:issue:`55027`)
+- Implement :meth:`Series.interpolate` and :meth:`DataFrame.interpolate` for :class:`ArrowDtype` and masked dtypes (:issue:`56267`)
 - Implement masked algorithms for :meth:`Series.value_counts` (:issue:`54984`)
 - Implemented :meth:`Series.dt` methods and attributes for :class:`ArrowDtype` with ``pyarrow.duration`` type (:issue:`52284`)
 - Implemented :meth:`Series.str.extract` for :class:`ArrowDtype` (:issue:`56268`)

pandas/core/arrays/arrow/array.py

@@ -184,6 +184,7 @@ def floordiv_compat(
         AxisInt,
         Dtype,
         FillnaOptions,
+        InterpolateOptions,
         Iterator,
         NpDtype,
         NumpySorter,
@@ -2068,6 +2069,45 @@ def _maybe_convert_setitem_value(self, value):
             raise TypeError(msg) from err
         return value
 
+    def interpolate(
+        self,
+        *,
+        method: InterpolateOptions,
+        axis: int,
+        index,
+        limit,
+        limit_direction,
+        limit_area,
+        copy: bool,
+        **kwargs,
+    ) -> Self:
+        """
+        See NDFrame.interpolate.__doc__.
+        """
+        # NB: we return type(self) even if copy=False
+        mask = self.isna()
+        if self.dtype.kind == "f":
+            data = self._pa_array.to_numpy()
+        elif self.dtype.kind in "iu":
+            data = self.to_numpy(dtype="f8", na_value=0.0)
+        else:
+            raise NotImplementedError(
+                f"interpolate is not implemented for dtype={self.dtype}"
+            )
+
+        missing.interpolate_2d_inplace(
+            data,
+            method=method,
+            axis=0,
+            index=index,
+            limit=limit,
+            limit_direction=limit_direction,
+            limit_area=limit_area,
+            mask=mask,
+            **kwargs,
+        )
+        return type(self)(self._box_pa_array(pa.array(data, mask=mask)))
+
     @classmethod
     def _if_else(
         cls,

pandas/core/arrays/masked.py

@@ -22,6 +22,7 @@
     AxisInt,
     DtypeObj,
     FillnaOptions,
+    InterpolateOptions,
     NpDtype,
     PositionalIndexer,
     Scalar,
@@ -99,6 +100,7 @@
         NumpyValueArrayLike,
     )
     from pandas._libs.missing import NAType
+    from pandas.core.arrays import FloatingArray
 
 from pandas.compat.numpy import function as nv
 
@@ -1521,6 +1523,58 @@ def all(
             else:
                 return self.dtype.na_value
 
+    def interpolate(
+        self,
+        *,
+        method: InterpolateOptions,
+        axis: int,
+        index,
+        limit,
+        limit_direction,
+        limit_area,
+        copy: bool,
+        **kwargs,
+    ) -> FloatingArray:
+        """
+        See NDFrame.interpolate.__doc__.
+        """
+        # NB: we return type(self) even if copy=False
+        if self.dtype.kind == "f":
+            if copy:
+                data = self._data.copy()
+                mask = self._mask.copy()
+            else:
+                data = self._data
+                mask = self._mask
+        elif self.dtype.kind in "iu":
+            copy = True
+            data = self._data.astype("f8")
+            mask = self._mask.copy()
+        else:
+            raise NotImplementedError(
+                f"interpolate is not implemented for dtype={self.dtype}"
+            )
+
+        missing.interpolate_2d_inplace(
+            data,
+            method=method,
+            axis=0,
+            index=index,
+            limit=limit,
+            limit_direction=limit_direction,
+            limit_area=limit_area,
+            mask=mask,
+            **kwargs,
+        )
+        if not copy:
+            return self  # type: ignore[return-value]
+        if self.dtype.kind == "f":
+            return type(self)._simple_new(data, mask)  # type: ignore[return-value]
+        else:
+            from pandas.core.arrays import FloatingArray
+
+            return FloatingArray._simple_new(data, mask)
+
     def _accumulate(
         self, name: str, *, skipna: bool = True, **kwargs
     ) -> BaseMaskedArray:

pandas/core/missing.py

@@ -349,6 +349,7 @@ def interpolate_2d_inplace(
     limit_direction: str = "forward",
     limit_area: str | None = None,
     fill_value: Any | None = None,
+    mask=None,
     **kwargs,
 ) -> None:
     """
@@ -396,6 +397,7 @@ def func(yvalues: np.ndarray) -> None:
             limit_area=limit_area_validated,
             fill_value=fill_value,
             bounds_error=False,
+            mask=mask,
             **kwargs,
         )
 
@@ -440,6 +442,7 @@ def _interpolate_1d(
     fill_value: Any | None = None,
     bounds_error: bool = False,
     order: int | None = None,
+    mask=None,
     **kwargs,
 ) -> None:
     """
@@ -453,8 +456,10 @@ def _interpolate_1d(
     -----
     Fills 'yvalues' in-place.
     """
-
-    invalid = isna(yvalues)
+    if mask is not None:
+        invalid = mask
+    else:
+        invalid = isna(yvalues)
     valid = ~invalid
 
     if not valid.any():
@@ -531,7 +536,10 @@ def _interpolate_1d(
             **kwargs,
         )
 
-    if is_datetimelike:
+    if mask is not None:
+        mask[:] = False
+        mask[preserve_nans] = True
+    elif is_datetimelike:
         yvalues[preserve_nans] = NaT.value
     else:
         yvalues[preserve_nans] = np.nan

pandas/tests/frame/methods/test_interpolate.py

@@ -498,8 +498,41 @@ def test_interpolate_empty_df(self):
         assert result is None
         tm.assert_frame_equal(df, expected)
 
-    def test_interpolate_ea_raise(self):
+    def test_interpolate_ea(self, any_int_ea_dtype):
         # GH#55347
-        df = DataFrame({"a": [1, None, 2]}, dtype="Int64")
-        with pytest.raises(NotImplementedError, match="does not implement"):
-            df.interpolate()
+        df = DataFrame({"a": [1, None, None, None, 3]}, dtype=any_int_ea_dtype)
+        orig = df.copy()
+        result = df.interpolate(limit=2)
+        expected = DataFrame({"a": [1, 1.5, 2.0, None, 3]}, dtype="Float64")
+        tm.assert_frame_equal(result, expected)
+        tm.assert_frame_equal(df, orig)
+
+    @pytest.mark.parametrize(
+        "dtype",
+        [
+            "Float64",
+            "Float32",
+            pytest.param("float32[pyarrow]", marks=td.skip_if_no("pyarrow")),
+            pytest.param("float64[pyarrow]", marks=td.skip_if_no("pyarrow")),
+        ],
+    )
+    def test_interpolate_ea_float(self, dtype):
+        # GH#55347
+        df = DataFrame({"a": [1, None, None, None, 3]}, dtype=dtype)
+        orig = df.copy()
+        result = df.interpolate(limit=2)
+        expected = DataFrame({"a": [1, 1.5, 2.0, None, 3]}, dtype=dtype)
+        tm.assert_frame_equal(result, expected)
+        tm.assert_frame_equal(df, orig)
+
+    @pytest.mark.parametrize(
+        "dtype",
+        ["int64", "uint64", "int32", "int16", "int8", "uint32", "uint16", "uint8"],
+    )
+    def test_interpolate_arrow(self, dtype):
+        # GH#55347
+        pytest.importorskip("pyarrow")
+        df = DataFrame({"a": [1, None, None, None, 3]}, dtype=dtype + "[pyarrow]")
+        result = df.interpolate(limit=2)
+        expected = DataFrame({"a": [1, 1.5, 2.0, None, 3]}, dtype="float64[pyarrow]")
+        tm.assert_frame_equal(result, expected)