dycw-utilities 0.109.14__tar.gz → 0.109.16__tar.gz

{dycw_utilities-0.109.14 → dycw_utilities-0.109.16}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dycw-utilities
-Version: 0.109.14
+Version: 0.109.16
 Author-email: Derek Wan <d.wan@icloud.com>
 License-File: LICENSE
 Requires-Python: >=3.12

{dycw_utilities-0.109.14 → dycw_utilities-0.109.16}/pyproject.toml

@@ -92,7 +92,7 @@ dependencies = [
 name = "dycw-utilities"
 readme = "README.md"
 requires-python = ">= 3.12"
-version = "0.109.14"
+version = "0.109.16"
 
 [project.optional-dependencies]
 test = [
@@ -335,7 +335,7 @@ zzz-test-zoneinfo = [
 # bump-my-version
 [tool.bumpversion]
 allow_dirty = true
-current_version = "0.109.14"
+current_version = "0.109.16"
 
 [[tool.bumpversion.files]]
 filename = "src/utilities/__init__.py"

{dycw_utilities-0.109.14 → dycw_utilities-0.109.16}/src/tests/test_numpy.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 from re import escape
 from typing import TYPE_CHECKING, Any, Literal
 
+import numpy as np
 from hypothesis import assume, given
 from hypothesis.strategies import DataObject, data, floats, integers, none
 from numpy import (
@@ -12,13 +13,16 @@ from numpy import (
     full,
     inf,
     isclose,
+    linspace,
     median,
     nan,
     ndarray,
     ones,
+    pi,
     zeros,
     zeros_like,
 )
+from numpy.fft import fft, fftfreq
 from numpy.random import Generator
 from numpy.testing import assert_equal
 from pytest import mark, param, raises
@@ -36,7 +40,9 @@ from utilities.numpy import (
     as_int,
     discretize,
     fillna,
+    filter_frequencies,
     flatn0,
+    get_frequency_spectrum,
     has_dtype,
     is_at_least,
     is_at_least_or_nan,
@@ -268,6 +274,19 @@ class TestFillNa:
         assert_equal(result, expected)
 
 
+class TestFilterFrequencies:
+    def test_main(self) -> None:
+        n = 1000
+        x = linspace(0, 2 * pi, n)
+        noise = DEFAULT_RNG.normal(scale=0.25, size=n)
+        y = x + noise
+        result = filter_frequencies(y, lambda f: np.abs(f) <= 0.02)
+        assert result.shape == (n,)
+        fft_vals = fft(result)
+        freqs = fftfreq(n)
+        assert np.allclose(fft_vals[np.abs(freqs) > 0.02], 0.0)
+
+
 class TestFlatN0:
     @given(data=data(), n=integers(1, 10))
     def test_main(self, *, data: DataObject, n: int) -> None:
@@ -288,6 +307,18 @@ class TestFlatN0:
             _ = flatn0(ones(2, dtype=bool))
 
 
+class TestGetFrequencySpectrum:
+    def test_main(self) -> None:
+        n = 1000
+        x = linspace(0, 2 * pi, n)
+        noise = DEFAULT_RNG.normal(scale=0.25, size=n)
+        y = x + noise
+        y2 = filter_frequencies(y, lambda f: np.abs(f) <= 0.02)
+        result = get_frequency_spectrum(y2)
+        assert result.shape == (n, 2)
+        assert np.allclose(result[result[:, 0] > 0.02, 1], 0.0)
+
+
 class TestHasDtype:
     @mark.parametrize(("dtype", "expected"), [param(float, True), param(int, False)])
     @mark.parametrize("is_tuple", [param(True), param(False)])

dycw_utilities-0.109.16/src/tests/test_polars_ols.py

@@ -0,0 +1,114 @@
+from __future__ import annotations
+
+from hypothesis import given
+from hypothesis.strategies import sampled_from
+from numpy import isclose
+from polars import DataFrame, Float64, Series, Struct, col
+from polars.testing import assert_frame_equal
+from sklearn.linear_model import LinearRegression
+
+from utilities.polars import concat_series, integers, normal
+from utilities.polars_ols import compute_rolling_ols
+
+
+class TestComputeRollingOLS:
+    def test_main_self(self) -> None:
+        df = self._df.with_columns(
+            compute_rolling_ols(
+                "y", "x1", "x2", window_size=5, min_periods=5, add_intercept=True
+            )
+        )
+        self._assert_series(df["ols"])
+
+    def test_main_series(self) -> None:
+        ols = compute_rolling_ols(
+            self._df["y"],
+            self._df["x1"],
+            self._df["x2"],
+            window_size=5,
+            min_periods=5,
+            add_intercept=True,
+        )
+        self._assert_series(ols)
+
+    @given(
+        case=sampled_from([
+            (
+                slice(-7, -2),
+                [0.3619563208480195, 0.6583229512154678],
+                -1.386023798329262,
+                7.434533329394103,
+                0.994253238813284,
+            ),
+            (
+                slice(-6, -1),
+                [0.35564162435283264, 0.6656931556738643],
+                -0.5626805730005437,
+                -51.903154626050124,
+                0.9979752966843768,
+            ),
+            (
+                slice(-5, None),
+                [0.3100421300754358, 0.6753578168818635],
+                0.48493124625502837,
+                -36.70039604095908,
+                0.9977272526713715,
+            ),
+        ])
+    )
+    def test_tail(
+        self, *, case: tuple[slice, list[float], float, float, float]
+    ) -> None:
+        slice_, coeffs, intercept, prediction, r2 = case
+        df = self._df[slice_]
+        X = df.select("x1", "x2").to_numpy()  # noqa: N806
+        y = df.select("y").to_numpy()
+        model = LinearRegression()
+        model = model.fit(X, y)
+        assert isclose(model.coef_, coeffs).all()
+        assert isclose(model.intercept_, intercept)
+        assert isclose(model.predict(X)[-1], prediction).all()
+        assert isclose(model.score(X, y), r2)
+
+    @property
+    def _df(self) -> DataFrame:
+        n = 20
+        return concat_series(
+            integers(n, -100, high=100, seed=0).alias("x1"),
+            integers(n, -100, high=100, seed=1).alias("x2"),
+        ).with_columns(
+            ((col("x1") + 2 * col("x2") + normal(n, scale=10.0, seed=2)) / 3).alias("y")
+        )
+
+    def _assert_series(self, series: Series, /) -> None:
+        df = series.struct.unnest()
+        tail = df[-10:]
+        # fmt: off
+        data = [
+            ({"x1": 0.333396198442681, "x2": 0.6845517746145712, "const": 0.2808021232120448}, 59.921571424913495, 1.67032007883995, 0.9955364659986504),
+            ({"x1": 0.322785525889542, "x2": 0.6896341527044252, "const": 0.5401793852579858}, 15.974446064929626, -0.3429678871268038, 0.9961762567103958),
+            ({"x1": 0.31042868991153927, "x2": 0.7055685710743383, "const": 1.145326562525439}, -31.310827706894123, -0.45191863996575066, 0.998022262986332),
+            ({"x1": 0.33311466967931097, "x2": 0.684137842579758, "const": -0.7961518480794516}, 50.66821598287034, -2.975371834066671, 0.9974533939791341),
+            ({"x1": 0.35299385150914864, "x2": 0.6758890569593843, "const": -0.9377907849336107}, -0.8749325340834626, 1.0581261048863142, 0.9973453833170313),
+            ({"x1": 0.351300641938209, "x2": 0.6456834722890913, "const": -1.859577387752822}, 1.6809655259738476, 0.3217076349681922, 0.9951571413022856),
+            ({"x1": 0.3583378199895871, "x2": 0.6588347796692774, "const": -1.109675446287481}, 26.65448170418155, 2.496480675700724, 0.9933751737130443),
+            ({"x1": 0.36195632084801765, "x2": 0.658322951215466, "const": -1.3860237983291754}, 7.43453332939416, -0.791818995629618, 0.9905085882663488),
+            ({"x1": 0.35564162435283225, "x2": 0.6656931556738634, "const": -0.562680573000551}, -51.90315462605006, 0.6592474497562932, 0.9973576833556038),
+            ({"x1": 0.3100421300754357, "x2": 0.675357816881863, "const": 0.48493124625501927}, -36.70039604095908, -0.6071841038068868, 0.9978541580643828),
+        ]
+        # fmt: on
+        expected = DataFrame(
+            data=data,
+            schema={
+                "coefficients": Struct({
+                    "x1": Float64,
+                    "x2": Float64,
+                    "const": Float64,
+                }),
+                "predictions": Float64,
+                "residuals": Float64,
+                "R2": Float64,
+            },
+            orient="row",
+        )
+        assert_frame_equal(tail, expected)

{dycw_utilities-0.109.14 → dycw_utilities-0.109.16}/src/utilities/__init__.py

@@ -1,3 +1,3 @@
 from __future__ import annotations
 
-__version__ = "0.109.14"
+__version__ = "0.109.16"

{dycw_utilities-0.109.14 → dycw_utilities-0.109.16}/src/utilities/numpy.py

@@ -1,19 +1,22 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
-from functools import reduce
+from functools import partial, reduce
 from itertools import repeat
 from typing import TYPE_CHECKING, Any, overload, override
 
 import numpy as np
 from numpy import (
+    argsort,
     array,
     bool_,
+    complex128,
     digitize,
     dtype,
     errstate,
     flatnonzero,
     float64,
+    floating,
     full_like,
     inf,
     int64,
@@ -30,6 +33,7 @@ from numpy import (
     roll,
     where,
 )
+from numpy.fft import fft, fftfreq, ifft
 from numpy.linalg import det, eig
 from numpy.random import default_rng
 from numpy.typing import NDArray
@@ -37,7 +41,7 @@ from numpy.typing import NDArray
 from utilities.iterables import is_iterable_not_str
 
 if TYPE_CHECKING:
-    from collections.abc import Iterable
+    from collections.abc import Callable, Iterable
 
 
 ##
@@ -160,6 +164,33 @@ def fillna(array: NDArrayF, /, *, value: float = 0.0) -> NDArrayF:
 ##
 
 
+def filter_frequencies(
+    array: NDArrayF,
+    /,
+    *filters: Callable[[NDArray[floating[Any]]], NDArrayB],
+    d: int = 1,
+) -> NDArrayF:
+    """Filter an array by the frequencies of its FFT."""
+    (n,) = array.shape
+    fft_vals = fft(array)
+    freqs = fftfreq(n, d=d)
+    reduced = reduce(partial(_filter_frequencies_one, freqs=freqs), filters, fft_vals)
+    return ifft(reduced).real
+
+
+def _filter_frequencies_one(
+    acc: NDArray[complex128],
+    el: Callable[[NDArray[floating[Any]]], NDArrayB],
+    /,
+    *,
+    freqs: NDArray[floating[Any]],
+) -> NDArray[complex128]:
+    return where(el(freqs), acc, 0.0)
+
+
+##
+
+
 def flatn0(array: NDArrayB, /) -> int:
     """Return the index of the unique True element."""
     if not array.any():
@@ -193,6 +224,19 @@ class FlatN0MultipleError(FlatN0Error):
 ##
 
 
+def get_frequency_spectrum(array: NDArrayF, /, *, d: int = 1) -> NDArray[floating[Any]]:
+    """Get the frequency spectrum."""
+    (n,) = array.shape
+    fft_vals = fft(array)
+    freqs = fftfreq(n, d=d)
+    amplitudes = np.abs(fft_vals)
+    data = np.hstack([freqs.reshape(-1, 1), amplitudes.reshape(-1, 1)])
+    return data[argsort(data[:, 0])]
+
+
+##
+
+
 def has_dtype(x: Any, dtype: Any, /) -> bool:
     """Check if an object has the required dtype."""
     if is_iterable_not_str(dtype):
@@ -851,7 +895,9 @@ __all__ = [
     "datetime64us",
     "discretize",
     "fillna",
+    "filter_frequencies",
     "flatn0",
+    "get_frequency_spectrum",
     "has_dtype",
     "is_at_least",
     "is_at_least_or_nan",

dycw_utilities-0.109.16/src/utilities/polars_ols.py

@@ -0,0 +1,171 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, overload
+
+from polars import Expr, Series, struct
+from polars_ols import RollingKwargs, compute_rolling_least_squares
+
+from utilities.errors import ImpossibleCaseError
+from utilities.functions import is_sequence_of
+from utilities.polars import concat_series, ensure_expr_or_series
+
+if TYPE_CHECKING:
+    from polars._typing import IntoExprColumn
+    from polars_ols import NullPolicy
+
+    from utilities.polars import ExprLike
+
+
+@overload
+def compute_rolling_ols(
+    target: ExprLike,
+    *features: ExprLike,
+    sample_weights: ExprLike | None = None,
+    add_intercept: bool = False,
+    null_policy: NullPolicy = "drop_window",
+    window_size: int = 1000000,
+    min_periods: int | None = None,
+    use_woodbury: bool | None = None,
+    alpha: float | None = None,
+) -> Expr: ...
+@overload
+def compute_rolling_ols(
+    target: Series,
+    *features: Series,
+    sample_weights: Series | None = None,
+    add_intercept: bool = False,
+    null_policy: NullPolicy = "drop_window",
+    window_size: int = 1000000,
+    min_periods: int | None = None,
+    use_woodbury: bool | None = None,
+    alpha: float | None = None,
+) -> Series: ...
+@overload
+def compute_rolling_ols(
+    target: IntoExprColumn,
+    *features: IntoExprColumn,
+    sample_weights: IntoExprColumn | None = None,
+    add_intercept: bool = False,
+    null_policy: NullPolicy = "drop_window",
+    window_size: int = 1000000,
+    min_periods: int | None = None,
+    use_woodbury: bool | None = None,
+    alpha: float | None = None,
+) -> Expr | Series: ...
+def compute_rolling_ols(
+    target: IntoExprColumn,
+    *features: IntoExprColumn,
+    sample_weights: IntoExprColumn | None = None,
+    add_intercept: bool = False,
+    null_policy: NullPolicy = "drop_window",
+    window_size: int = 1000000,
+    min_periods: int | None = None,
+    use_woodbury: bool | None = None,
+    alpha: float | None = None,
+) -> Expr | Series:
+    """Compute a rolling OLS."""
+    target = ensure_expr_or_series(target)
+    features2 = tuple(map(ensure_expr_or_series, features))
+    sample_weights = (
+        None if sample_weights is None else ensure_expr_or_series(sample_weights)
+    )
+    if (
+        isinstance(target, Expr)
+        and is_sequence_of(features2, Expr)
+        and ((sample_weights is None) or isinstance(sample_weights, Expr))
+    ):
+        return _compute_rolling_ols_expr(
+            target,
+            *features2,
+            sample_weights=sample_weights,
+            add_intercept=add_intercept,
+            null_policy=null_policy,
+            window_size=window_size,
+            min_periods=min_periods,
+            use_woodbury=use_woodbury,
+            alpha=alpha,
+        )
+    if (
+        isinstance(target, Series)
+        and is_sequence_of(features2, Series)
+        and ((sample_weights is None) or isinstance(sample_weights, Series))
+    ):
+        return concat_series(
+            target, *features2, *([] if sample_weights is None else [sample_weights])
+        ).with_columns(
+            _compute_rolling_ols_expr(
+                target.name,
+                *(f.name for f in features2),
+                sample_weights=None if sample_weights is None else sample_weights.name,
+                add_intercept=add_intercept,
+                null_policy=null_policy,
+                window_size=window_size,
+                min_periods=min_periods,
+                use_woodbury=use_woodbury,
+                alpha=alpha,
+            )
+        )["ols"]
+    raise ImpossibleCaseError(  # pragma: no cover
+        case=[f"{target=}", f"{features2=}", f"{sample_weights=}"]
+    )
+
+
+def _compute_rolling_ols_expr(
+    target: ExprLike,
+    *features: ExprLike,
+    sample_weights: ExprLike | None = None,
+    add_intercept: bool = False,
+    null_policy: NullPolicy = "drop_window",
+    window_size: int = 1000000,
+    min_periods: int | None = None,
+    use_woodbury: bool | None = None,
+    alpha: float | None = None,
+) -> Expr:
+    """Compute a rolling OLS."""
+    target = ensure_expr_or_series(target)
+    features2 = tuple(map(ensure_expr_or_series, features))
+    sample_weights = (
+        None if sample_weights is None else ensure_expr_or_series(sample_weights)
+    )
+    rolling_kwargs = RollingKwargs(
+        null_policy=null_policy,
+        window_size=window_size,
+        min_periods=min_periods,
+        use_woodbury=use_woodbury,
+        alpha=alpha,
+    )
+    coefficients = compute_rolling_least_squares(
+        target,
+        *features2,
+        sample_weights=sample_weights,
+        add_intercept=add_intercept,
+        mode="coefficients",
+        rolling_kwargs=rolling_kwargs,
+    ).alias("coefficients")
+    predictions = compute_rolling_least_squares(
+        target,
+        *features2,
+        sample_weights=sample_weights,
+        add_intercept=add_intercept,
+        mode="predictions",
+        rolling_kwargs=rolling_kwargs,
+    ).alias("predictions")
+    residuals = compute_rolling_least_squares(
+        target,
+        *features2,
+        sample_weights=sample_weights,
+        add_intercept=add_intercept,
+        mode="residuals",
+        rolling_kwargs=rolling_kwargs,
+    ).alias("residuals")
+    ssr = (residuals**2).rolling_sum(window_size, min_samples=min_periods).alias("SSR")
+    sst = (
+        ((target - target.rolling_mean(window_size, min_samples=min_periods)) ** 2)
+        .rolling_sum(window_size, min_samples=min_periods)
+        .alias("SST")
+    )
+    r2 = (1 - ssr / sst).alias("R2")
+    return struct(coefficients, predictions, residuals, r2).alias("ols")
+
+
+__all__ = ["compute_rolling_ols"]

dycw_utilities-0.109.14/src/tests/test_polars_ols.py

@@ -1,103 +0,0 @@
-from __future__ import annotations
-
-from hypothesis import given
-from hypothesis.strategies import sampled_from
-from numpy import isclose
-from polars import DataFrame, Float64, Struct, UInt32, col
-from polars.testing import assert_frame_equal
-from sklearn.linear_model import LinearRegression
-
-from utilities.polars import concat_series, integers, normal
-from utilities.polars_ols import compute_rolling_ols
-
-
-class TestComputeRollingOLS:
-    def test_ols_results(self) -> None:
-        result = self._df["ols"].struct.unnest().with_row_index()
-        # fmt: off
-        exp_values = [
-            (10, {"x1": 0.333396198442681, "x2": 0.6845517746145712, "const": 0.2808021232120448}, 59.921571424913495, 1.67032007883995, 0.9955364659986504),
-            (11, {"x1": 0.322785525889542, "x2": 0.6896341527044252, "const": 0.5401793852579858}, 15.974446064929626, -0.3429678871268038, 0.9961762567103958),
-            (12, {"x1": 0.31042868991153927, "x2": 0.7055685710743383, "const": 1.145326562525439}, -31.310827706894123, -0.45191863996575066, 0.998022262986332),
-            (13, {"x1": 0.33311466967931097, "x2": 0.684137842579758, "const": -0.7961518480794516}, 50.66821598287034, -2.975371834066671, 0.9974533939791341),
-            (14, {"x1": 0.35299385150914864, "x2": 0.6758890569593843, "const": -0.9377907849336107}, -0.8749325340834626, 1.0581261048863142, 0.9973453833170313),
-            (15, {"x1": 0.351300641938209, "x2": 0.6456834722890913, "const": -1.859577387752822}, 1.6809655259738476, 0.3217076349681922, 0.9951571413022856),
-            (16, {"x1": 0.3583378199895871, "x2": 0.6588347796692774, "const": -1.109675446287481}, 26.65448170418155, 2.496480675700724, 0.9933751737130443),
-            (17, {"x1": 0.36195632084801765, "x2": 0.658322951215466, "const": -1.3860237983291754}, 7.43453332939416, -0.791818995629618, 0.9905085882663488),
-            (18, {"x1": 0.35564162435283225, "x2": 0.6656931556738634, "const": -0.562680573000551}, -51.90315462605006, 0.6592474497562932, 0.9973576833556038),
-            (19, {"x1": 0.3100421300754357, "x2": 0.675357816881863, "const": 0.48493124625501927}, -36.70039604095908, -0.6071841038068868, 0.9978541580643828),
-        ]
-        # fmt: on
-        expected = DataFrame(
-            data=exp_values,
-            schema={
-                "index": UInt32,
-                "coefficients": Struct({
-                    "x1": Float64,
-                    "x2": Float64,
-                    "const": Float64,
-                }),
-                "predictions": Float64,
-                "residuals": Float64,
-                "R2": Float64,
-            },
-            orient="row",
-        )
-        assert_frame_equal(result[-10:], expected)
-
-    @given(
-        case=sampled_from([
-            (
-                slice(-7, -2),
-                [0.3619563208480195, 0.6583229512154678],
-                -1.386023798329262,
-                7.434533329394103,
-                0.994253238813284,
-            ),
-            (
-                slice(-6, -1),
-                [0.35564162435283264, 0.6656931556738643],
-                -0.5626805730005437,
-                -51.903154626050124,
-                0.9979752966843768,
-            ),
-            (
-                slice(-5, None),
-                [0.3100421300754358, 0.6753578168818635],
-                0.48493124625502837,
-                -36.70039604095908,
-                0.9977272526713715,
-            ),
-        ])
-    )
-    def test_tail(
-        self, *, case: tuple[slice, list[float], float, float, float]
-    ) -> None:
-        slice_, coeffs, intercept, prediction, r2 = case
-        df = self._df[slice_]
-        X = df.select("x1", "x2").to_numpy()  # noqa: N806
-        y = df.select("y").to_numpy()
-        model = LinearRegression()
-        model = model.fit(X, y)
-        assert isclose(model.coef_, coeffs).all()
-        assert isclose(model.intercept_, intercept)
-        assert isclose(model.predict(X)[-1], prediction).all()
-        assert isclose(model.score(X, y), r2)
-
-    @property
-    def _df(self) -> DataFrame:
-        n = 20
-        return (
-            concat_series(
-                integers(n, -100, high=100, seed=0).alias("x1"),
-                integers(n, -100, high=100, seed=1).alias("x2"),
-            )
-            .with_columns(
-                y=(col("x1") + 2 * col("x2") + normal(n, scale=10.0, seed=2)) / 3
-            )
-            .with_columns(
-                compute_rolling_ols(
-                    "y", "x1", "x2", window_size=5, min_periods=5, add_intercept=True
-                )
-            )
-        )

dycw_utilities-0.109.14/src/utilities/polars_ols.py

@@ -1,71 +0,0 @@
-from __future__ import annotations
-
-from typing import TYPE_CHECKING
-
-from polars import struct
-from polars_ols import RollingKwargs, compute_rolling_least_squares
-
-from utilities.polars import ensure_expr_or_series
-
-if TYPE_CHECKING:
-    from polars import Expr
-    from polars_ols import NullPolicy
-
-    from utilities.polars import ExprLike
-
-
-def compute_rolling_ols(
-    target: ExprLike,
-    *features: ExprLike,
-    sample_weights: ExprLike | None = None,
-    add_intercept: bool = False,
-    null_policy: NullPolicy = "drop_window",
-    window_size: int = 1000000,
-    min_periods: int | None = None,
-    use_woodbury: bool | None = None,
-    alpha: float | None = None,
-) -> Expr:
-    """Compute a rolling OLS."""
-    target = ensure_expr_or_series(target)
-    rolling_kwargs = RollingKwargs(
-        null_policy=null_policy,
-        window_size=window_size,
-        min_periods=min_periods,
-        use_woodbury=use_woodbury,
-        alpha=alpha,
-    )
-    coefficients = compute_rolling_least_squares(
-        target,
-        *features,
-        sample_weights=sample_weights,
-        add_intercept=add_intercept,
-        mode="coefficients",
-        rolling_kwargs=rolling_kwargs,
-    ).alias("coefficients")
-    predictions = compute_rolling_least_squares(
-        target,
-        *features,
-        sample_weights=sample_weights,
-        add_intercept=add_intercept,
-        mode="predictions",
-        rolling_kwargs=rolling_kwargs,
-    ).alias("predictions")
-    residuals = compute_rolling_least_squares(
-        target,
-        *features,
-        sample_weights=sample_weights,
-        add_intercept=add_intercept,
-        mode="residuals",
-        rolling_kwargs=rolling_kwargs,
-    ).alias("residuals")
-    ssr = (residuals**2).rolling_sum(window_size, min_samples=min_periods).alias("SSR")
-    sst = (
-        ((target - target.rolling_mean(window_size, min_samples=min_periods)) ** 2)
-        .rolling_sum(window_size, min_samples=min_periods)
-        .alias("SST")
-    )
-    r2 = (1 - ssr / sst).alias("R2")
-    return struct(coefficients, predictions, residuals, r2).alias("ols")
-
-
-__all__ = ["compute_rolling_ols"]