ENH change Ridge tol to 1e-4 (scikit-learn#24465)

lorentzenchr · glemaitre · commit 6be5770095b1 · 2022-10-31T11:25:37.000+01:00
* ENH set defaul tol of Ridge to 1e-4 instead of 1e-3

* MNT rename internal least square method in discrimant_analysis

* DOC add versionchanged

* DOC add whats_new entry

* DOC add pr to whatsnew

* DOC better versionchanged message
diff --git a/doc/whats_new/v1.2.rst b/doc/whats_new/v1.2.rst
@@ -31,6 +31,11 @@ random sampling procedures.
   to a tiny value. Moreover, `verbose` is now properly propagated to L-BFGS-B.
   :pr:`23619` by :user:`Christian Lorentzen <lorentzenchr>`.
 
+- |API| The default value of `tol` was changed from `1e-3` to `1e-4` for
+  :func:`linear_model.ridge_regression`, :class:`linear_model.Ridge` and
+  :class:`linear_model.`RidgeClassifier`.
+  :pr:`24465` by :user:`Christian Lorentzen <lorentzenchr>`.
+
 - |Fix| Make sign of `components_` deterministic in :class:`decomposition.SparsePCA`.
   :pr:`23935` by :user:`Guillaume Lemaitre <glemaitre>`.
 
@@ -333,6 +338,11 @@ Changelog
   in :class:`linear_model.LogisticRegression`, and will be removed in version 1.4.
   Use `None` instead. :pr:`23877` by :user:`Zhehao Liu <MaxwellLZH>`.
 
+- |API| The default value of `tol` was changed from `1e-3` to `1e-4` for
+  :func:`linear_model.ridge_regression`, :class:`linear_model.Ridge` and
+  :class:`linear_model.`RidgeClassifier`.
+  :pr:`24465` by :user:`Christian Lorentzen <lorentzenchr>`.
+
 - |Fix| :class:`linear_model.SGDOneClassSVM` no longer performs parameter
   validation in the constructor. All validation is now handled in `fit()` and
   `partial_fit()`.
diff --git a/sklearn/discriminant_analysis.py b/sklearn/discriminant_analysis.py
@@ -350,7 +350,7 @@ def __init__(
         self.tol = tol  # used only in svd solver
         self.covariance_estimator = covariance_estimator
 
-    def _solve_lsqr(self, X, y, shrinkage, covariance_estimator):
+    def _solve_lstsq(self, X, y, shrinkage, covariance_estimator):
         """Least squares solver.
 
         The least squares solver computes a straightforward solution of the
@@ -621,7 +621,7 @@ def fit(self, X, y):
                 )
             self._solve_svd(X, y)
         elif self.solver == "lsqr":
-            self._solve_lsqr(
+            self._solve_lstsq(
                 X,
                 y,
                 shrinkage=self.shrinkage,
diff --git a/sklearn/linear_model/_ridge.py b/sklearn/linear_model/_ridge.py
@@ -66,7 +66,7 @@ def _solve_sparse_cg(
     y,
     alpha,
     max_iter=None,
-    tol=1e-3,
+    tol=1e-4,
     verbose=0,
     X_offset=None,
     X_scale=None,
@@ -153,7 +153,7 @@ def _solve_lsqr(
     alpha,
     fit_intercept=True,
     max_iter=None,
-    tol=1e-3,
+    tol=1e-4,
     X_offset=None,
     X_scale=None,
     sample_weight_sqrt=None,
@@ -303,7 +303,7 @@ def _solve_lbfgs(
     alpha,
     positive=True,
     max_iter=None,
-    tol=1e-3,
+    tol=1e-4,
     X_offset=None,
     X_scale=None,
     sample_weight_sqrt=None,
@@ -381,7 +381,7 @@ def ridge_regression(
     sample_weight=None,
     solver="auto",
     max_iter=None,
-    tol=1e-3,
+    tol=1e-4,
     verbose=0,
     positive=False,
     random_state=None,
@@ -471,8 +471,13 @@ def ridge_regression(
         by scipy.sparse.linalg. For 'sag' and saga solver, the default value is
         1000. For 'lbfgs' solver, the default value is 15000.
 
-    tol : float, default=1e-3
-        Precision of the solution.
+    tol : float, default=1e-4
+        Precision of the solution. Note that `tol` has no effect for solvers 'svd' and
+        'cholesky'.
+
+        .. versionchanged:: 1.2
+           Default value changed from 1e-3 to 1e-4 for consistency with other linear
+           models.
 
     verbose : int, default=0
         Verbosity level. Setting verbose > 0 will display additional
@@ -556,7 +561,7 @@ def _ridge_regression(
     sample_weight=None,
     solver="auto",
     max_iter=None,
-    tol=1e-3,
+    tol=1e-4,
     verbose=0,
     positive=False,
     random_state=None,
@@ -803,7 +808,7 @@ def __init__(
         normalize="deprecated",
         copy_X=True,
         max_iter=None,
-        tol=1e-3,
+        tol=1e-4,
         solver="auto",
         positive=False,
         random_state=None,
@@ -977,8 +982,13 @@ class Ridge(MultiOutputMixin, RegressorMixin, _BaseRidge):
         by scipy.sparse.linalg. For 'sag' solver, the default value is 1000.
         For 'lbfgs' solver, the default value is 15000.
 
-    tol : float, default=1e-3
-        Precision of the solution.
+    tol : float, default=1e-4
+        Precision of the solution. Note that `tol` has no effect for solvers 'svd' and
+        'cholesky'.
+
+        .. versionchanged:: 1.2
+           Default value changed from 1e-3 to 1e-4 for consistency with other linear
+           models.
 
     solver : {'auto', 'svd', 'cholesky', 'lsqr', 'sparse_cg', \
             'sag', 'saga', 'lbfgs'}, default='auto'
@@ -1096,7 +1106,7 @@ def __init__(
         normalize="deprecated",
         copy_X=True,
         max_iter=None,
-        tol=1e-3,
+        tol=1e-4,
         solver="auto",
         positive=False,
         random_state=None,
@@ -1276,8 +1286,13 @@ class RidgeClassifier(_RidgeClassifierMixin, _BaseRidge):
         Maximum number of iterations for conjugate gradient solver.
         The default value is determined by scipy.sparse.linalg.
 
-    tol : float, default=1e-3
-        Precision of the solution.
+    tol : float, default=1e-4
+        Precision of the solution. Note that `tol` has no effect for solvers 'svd' and
+        'cholesky'.
+
+        .. versionchanged:: 1.2
+           Default value changed from 1e-3 to 1e-4 for consistency with other linear
+           models.
 
     class_weight : dict or 'balanced', default=None
         Weights associated with classes in the form ``{class_label: weight}``.
@@ -1397,7 +1412,7 @@ def __init__(
         normalize="deprecated",
         copy_X=True,
         max_iter=None,
-        tol=1e-3,
+        tol=1e-4,
         class_weight=None,
         solver="auto",
         positive=False,
diff --git a/sklearn/linear_model/tests/test_ridge.py b/sklearn/linear_model/tests/test_ridge.py
@@ -1660,7 +1660,7 @@ def test_ridge_fit_intercept_sparse_sag(with_sample_weight, global_random_seed):
     assert_allclose(dense_ridge.intercept_, sparse_ridge.intercept_, rtol=1e-4)
     assert_allclose(dense_ridge.coef_, sparse_ridge.coef_, rtol=1e-4)
     with pytest.warns(UserWarning, match='"sag" solver requires.*'):
-        Ridge(solver="sag").fit(X_csr, y)
+        Ridge(solver="sag", fit_intercept=True, tol=1e-3, max_iter=None).fit(X_csr, y)
 
 
 @pytest.mark.parametrize("return_intercept", [False, True])
