fetch upstream main, edit whats new, and add intercept to unit test

Author: floriankozikowski

doc/changes/0.5.rst

@@ -1,7 +1,7 @@
 .. _changes_0_5:
 
-Version 0.5 (in progress)
--------------------------
+Version 0.5 (29/07/2025)
+-----------------------
 - Add support for fitting an intercept in :ref:`SqrtLasso <skglm.experimental.SqrtLasso>` (PR: :gh:`298`)
 - Add experimental :ref:`QuantileHuber <skglm.experimental.QuantileHuber>` and :ref:`SmoothQuantileRegressor <skglm.experimental.SmoothQuantileRegressor>` for quantile regression (PR: :gh:`312`).
 - Add :ref:`GeneralizedLinearEstimatorCV <skglm.GeneralizedLinearEstimatorCV>` for cross-validation with automatic parameter selection for L1 and elastic-net penalties (PR: :gh:`299`)

doc/changes/0.6.rst

@@ -0,0 +1,6 @@
+.. _changes_0_6:
+
+Version 0.6 (in progress)
+-------------------------
+
+- :class:`skglm.solvers.LBFGS` now supports fitting an intercept with the `fit_intercept` parameter.

doc/changes/whats_new.rst

@@ -5,6 +5,8 @@ What's new
 
 .. currentmodule:: skglm
 
+.. include:: 0.6.rst
+
 .. include:: 0.5.rst
 
 .. include:: 0.4.rst

issue320.py

@@ -1,4 +1,4 @@
-__version__ = '0.5dev'
+__version__ = '0.6dev'
 
 from skglm.estimators import (  # noqa F401
     Lasso, WeightedLasso, ElasticNet, MCPRegression, MultiTaskLasso, LinearSVC,

skglm/__init__.py

@@ -12,7 +12,8 @@
 
 
 @pytest.mark.parametrize("X_sparse", [True, False])
-def test_lbfgs_L2_logreg(X_sparse):
+@pytest.mark.parametrize("fit_intercept", [True, False])
+def test_lbfgs_L2_logreg(X_sparse, fit_intercept):
     reg = 1.0
     X_density = 1.0 if not X_sparse else 0.5
     n_samples, n_features = 100, 50
@@ -28,17 +29,21 @@ def test_lbfgs_L2_logreg(X_sparse):
     # fit L-BFGS
     datafit = Logistic()
     penalty = L2(reg)
-    w, *_ = LBFGS(tol=1e-12).solve(X, y, datafit, penalty)
+    w, *_ = LBFGS(tol=1e-12, fit_intercept=fit_intercept).solve(X, y, datafit, penalty)
 
     # fit scikit learn
     estimator = LogisticRegression(
         penalty="l2",
         C=1 / (n_samples * reg),
-        fit_intercept=False,
+        fit_intercept=fit_intercept,
         tol=1e-12,
     ).fit(X, y)
 
-    np.testing.assert_allclose(w, estimator.coef_.flatten(), atol=1e-5)
+    if fit_intercept:
+        np.testing.assert_allclose(w[:-1], estimator.coef_.flatten(), atol=1e-5)
+        np.testing.assert_allclose(w[-1], estimator.intercept_[0], atol=1e-5)
+    else:
+        np.testing.assert_allclose(w, estimator.coef_.flatten(), atol=1e-5)
 
 
 @pytest.mark.parametrize("use_efron", [True, False])