FIX/MNT install R with conda and use python 3.10 on test workflow (#282)

Author: mathurinm

.github/workflows/main.yml

@@ -13,20 +13,21 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@v3
-    - name: Set up Python 3.8
+    - name: Set up Python 3.10
       uses: actions/setup-python@v3
       with:
-        python-version: 3.8
+        python-version: "3.10"
     - name: Install package and testing tools
       run: |
         python -m pip install --upgrade pip
         pip install .
         pip install .[test]
     - name: Install other dependencies
       run: |
+        conda install rpy2 -c conda-forge
         pip install celer
         pip install statsmodels cvxopt
-        pip install git+https://github.com/jolars/pyslope.git
+        pip install git+https://github.com/jolars/sortedl1.git
         # for testing Cox estimator
         pip install lifelines
         pip install pandas

skglm/estimators.py

@@ -10,7 +10,7 @@
 from sklearn.utils.validation import (check_is_fitted, check_array,
                                       check_consistent_length)
 from sklearn.linear_model._base import (
-    LinearModel, RegressorMixin,
+    RegressorMixin, LinearModel,
     LinearClassifierMixin, SparseCoefMixin, BaseEstimator
 )
 from sklearn.utils.extmath import softmax
@@ -302,7 +302,7 @@ def get_params(self, deep=False):
         return params
 
 
-class Lasso(LinearModel, RegressorMixin):
+class Lasso(RegressorMixin, LinearModel):
     r"""Lasso estimator based on Celer solver and primal extrapolation.
 
     The optimization objective for Lasso is:
@@ -449,7 +449,7 @@ def path(self, X, y, alphas, coef_init=None, return_n_iter=True, **params):
         return solver.path(X, y, datafit, penalty, alphas, coef_init, return_n_iter)
 
 
-class WeightedLasso(LinearModel, RegressorMixin):
+class WeightedLasso(RegressorMixin, LinearModel):
     r"""WeightedLasso estimator based on Celer solver and primal extrapolation.
 
     The optimization objective for WeightedLasso is:
@@ -612,7 +612,7 @@ def fit(self, X, y):
         return _glm_fit(X, y, self, Quadratic(), penalty, solver)
 
 
-class ElasticNet(LinearModel, RegressorMixin):
+class ElasticNet(RegressorMixin, LinearModel):
     r"""Elastic net estimator.
 
     The optimization objective for Elastic net is:
@@ -766,7 +766,7 @@ def fit(self, X, y):
                         L1_plus_L2(self.alpha, self.l1_ratio, self.positive), solver)
 
 
-class MCPRegression(LinearModel, RegressorMixin):
+class MCPRegression(RegressorMixin, LinearModel):
     r"""Linear regression with MCP penalty estimator.
 
     The optimization objective for MCPRegression is, with :math:`x >= 0`:
@@ -1381,7 +1381,7 @@ def fit(self, X, y):
         return self
 
 
-class MultiTaskLasso(LinearModel, RegressorMixin):
+class MultiTaskLasso(RegressorMixin, LinearModel):
     r"""MultiTaskLasso estimator.
 
     The optimization objective for MultiTaskLasso is:
@@ -1557,7 +1557,7 @@ def path(self, X, Y, alphas, coef_init=None, return_n_iter=False, **params):
         return solver.path(X, Y, datafit, penalty, alphas, coef_init, return_n_iter)
 
 
-class GroupLasso(LinearModel, RegressorMixin):
+class GroupLasso(RegressorMixin, LinearModel):
     r"""GroupLasso estimator based on Celer solver and primal extrapolation.
 
     The optimization objective for GroupLasso is:

skglm/tests/test_penalties.py

@@ -86,25 +86,24 @@ def test_slope_lasso():
 
 
 def test_slope():
-    # compare solutions with `pyslope`: https://github.com/jolars/pyslope
+    # compare solutions with `sortedl1`: https://github.com/jolars/sortedl1
     try:
-        from slope.solvers import pgd_slope  # noqa
-        from slope.utils import lambda_sequence  # noqa
+        from sortedl1 import Slope as SlopeEst  # noqa
     except ImportError:
         pytest.xfail(
             "This test requires slope to run.\n"
-            "https://github.com/jolars/pyslope")
+            "https://github.com/jolars/sortedl1")
 
-    q = 0.1
-    alphas = lambda_sequence(
-        X, y, fit_intercept=False, reg=alpha / alpha_max, q=q)
+    # q = 0.1
+    # alphas = lambda_sequence(
+    #     X, y, fit_intercept=False, reg=alpha / alpha_max, q=q)
+    clf = SlopeEst(alpha=0.01, fit_intercept=False).fit(X, y)
+    alphas = clf.lambda_
     ours = GeneralizedLinearEstimator(
-        penalty=SLOPE(alphas),
+        penalty=SLOPE(clf.alpha * alphas),
         solver=FISTA(max_iter=1000, tol=tol, opt_strategy="fixpoint"),
     ).fit(X, y)
-    pyslope_out = pgd_slope(
-        X, y, alphas, fit_intercept=False, max_it=1000, gap_tol=tol)
-    np.testing.assert_allclose(ours.coef_, pyslope_out["beta"], rtol=1e-5)
+    np.testing.assert_allclose(ours.coef_, clf.coef_, rtol=1e-5)
 
 
 @pytest.mark.parametrize("fit_intercept", [True, False])