Merge branch 'main' into glum-v3

Merge conflicts were due to new black version.

Author: stanmart

.github/workflows/build_wheels.yml

@@ -20,9 +20,10 @@ jobs:
         uses: pypa/[email protected]
         env:
           CIBW_ARCHS_MACOS: x86_64 arm64
-      - uses: actions/upload-artifact@v3
+      - uses: actions/upload-artifact@v4
         with:
           path: ./wheelhouse/*.whl
+          name: wheels-${{ matrix.os }}
 
   build_sdist:
     name: Build source distribution
@@ -37,9 +38,10 @@ jobs:
         run: python -m pip install setuptools setuptools-scm wheel Cython numpy scikit-learn
       - name: Build sdist
         run: python setup.py sdist
-      - uses: actions/upload-artifact@v3
+      - uses: actions/upload-artifact@v4
         with:
           path: dist/*.tar.gz
+          name: sdist
 
   upload_testpypi:
     if: github.event_name == 'release' && github.event.action == 'published'
@@ -51,9 +53,9 @@ jobs:
     permissions:
       id-token: write
     steps:
-      - uses: actions/download-artifact@v3
+      - uses: actions/download-artifact@v4
         with:
-          name: artifact
+          merge-multiple: true
           path: dist
       - uses: pypa/[email protected]
         with:
@@ -69,8 +71,8 @@ jobs:
     permissions:
       id-token: write
     steps:
-      - uses: actions/download-artifact@v3
+      - uses: actions/download-artifact@v4
         with:
-          name: artifact
           path: dist
+          merge-multiple: true
       - uses: pypa/[email protected]

.pre-commit-config.yaml

@@ -1,13 +1,13 @@
 repos:
  - repo: https://github.com/Quantco/pre-commit-mirrors-black
-   rev: 23.12.1
+   rev: 24.1.1
    hooks:
      - id: black-conda
        args:
          - --safe
          - --target-version=py39
  - repo: https://github.com/Quantco/pre-commit-mirrors-flake8
-   rev: 6.1.0
+   rev: 7.0.0
    hooks:
     - id: flake8-conda
       additional_dependencies: [

setup.py

@@ -74,9 +74,9 @@
     package_dir={"": "src"},
     packages=find_packages(
         where="src",
-        include=["glum"]
-        if os.environ.get("CONDA_BUILD")
-        else ["glum", "glum_benchmarks"],
+        include=(
+            ["glum"] if os.environ.get("CONDA_BUILD") else ["glum", "glum_benchmarks"]
+        ),
     ),
     python_requires=">=3.9",
     install_requires=[
@@ -89,13 +89,15 @@
         "formulaic>=0.6",
         "tabmat>=4.0.0a3",
     ],
-    entry_points=None
-    if os.environ.get("CONDA_BUILD")
-    else """
+    entry_points=(
+        None
+        if os.environ.get("CONDA_BUILD")
+        else """
         [console_scripts]
         glm_benchmarks_run = glum_benchmarks.cli_run:cli_run
         glm_benchmarks_analyze = glum_benchmarks.cli_analyze:cli_analyze
-    """,
+    """
+    ),
     ext_modules=cythonize(
         ext_modules,
         annotate=False,

src/glum/_distribution.py

@@ -942,9 +942,7 @@ def unit_deviance(self, y: np.ndarray, mu: np.ndarray) -> np.ndarray:
         -------
         array-like
         """
-        return 2 * y * (np.arctan(y) - np.arctan(mu)) + np.log(
-            (1 + mu**2) / (1 + y**2)
-        )
+        return 2 * y * (np.arctan(y) - np.arctan(mu)) + np.log((1 + mu**2) / (1 + y**2))
 
 
 class BinomialDistribution(ExponentialDispersionModel):

src/glum/_glm.py

@@ -2759,16 +2759,18 @@ def _expand_categorical_penalties(
                             return np.array(
                                 list(
                                     chain.from_iterable(
-                                        [
-                                            elmt
-                                            for _ in range(
-                                                len(dtype.categories)
-                                                + has_missing_category[col]
-                                                - drop_first
-                                            )
-                                        ]
-                                        if pd.api.types.is_categorical_dtype(dtype)
-                                        else [elmt]
+                                        (
+                                            [
+                                                elmt
+                                                for _ in range(
+                                                    len(dtype.categories)
+                                                    + has_missing_category[col]
+                                                    - drop_first
+                                                )
+                                            ]
+                                            if pd.api.types.is_categorical_dtype(dtype)
+                                            else [elmt]
+                                        )
                                         for elmt, (col, dtype) in zip(
                                             penalty, X.dtypes.items()
                                         )

src/glum_benchmarks/bench_liblinear.py

@@ -77,9 +77,11 @@ def liblinear_bench(
     model_args = dict(
         penalty=pen,
         tol=benchmark_convergence_tolerance,
-        C=1 / (X.shape[0] * alpha)
-        if reg_multiplier is None
-        else 1 / (X.shape[0] * alpha * reg_multiplier),
+        C=(
+            1 / (X.shape[0] * alpha)
+            if reg_multiplier is None
+            else 1 / (X.shape[0] * alpha * reg_multiplier)
+        ),
         # Note that when an intercept is fitted, it is subject to regularization, unlike
         # other solvers. intercept_scaling helps combat this by inflating the intercept
         # column, though too low of a value leaves too much regularization and too high

src/glum_benchmarks/cli_run.py

@@ -81,12 +81,14 @@ def cli_run(
         for Ln in libraries.keys():
             click.echo(f"running problem={Pn} library={Ln}")
             new_params = params.update_params(problem_name=Pn, library_name=Ln)
-            result, regularization_strength_ = execute_problem_library(
+            result, _ = execute_problem_library(
                 new_params,
                 iterations,
-                defaults["diagnostics_level"]
-                if params.diagnostics_level is None
-                else params.diagnostics_level,  # type: ignore
+                (
+                    defaults["diagnostics_level"]  # type: ignore
+                    if params.diagnostics_level is None  # type: ignore
+                    else params.diagnostics_level  # type: ignore
+                ),
             )
             _save_benchmark_results(
                 output_dir,

src/glum_benchmarks/orig_sklearn_fork/_glm.py

@@ -889,9 +889,7 @@ def unit_variance_derivative(self, mu):
         return 2 * mu
 
     def unit_deviance(self, y, mu):
-        return 2 * y * (np.arctan(y) - np.arctan(mu)) + np.log(
-            (1 + mu**2) / (1 + y**2)
-        )
+        return 2 * y * (np.arctan(y) - np.arctan(mu)) + np.log((1 + mu**2) / (1 + y**2))
 
 
 class BinomialDistribution(ExponentialDispersionModel):

src/glum_benchmarks/util.py

@@ -409,9 +409,7 @@ def clear_cache(force=False):
     if cache_location is None:
         return
 
-    cache_size_limit = float(
-        os.environ.get("GLM_BENCHMARKS_CACHE_SIZE_LIMIT", 1024**3)
-    )
+    cache_size_limit = float(os.environ.get("GLM_BENCHMARKS_CACHE_SIZE_LIMIT", 1024**3))
 
     if force or _get_size_of_cache_directory() > cache_size_limit:
         shutil.rmtree(cache_location)