SOLVER - add skglm L-BFGS

solvers/skglm.py

@@ -0,0 +1,54 @@
+import warnings
+from benchopt import BaseSolver, safe_import_context
+
+with safe_import_context() as import_ctx:
+    from sklearn.exceptions import ConvergenceWarning
+
+    from skglm.penalties import L2
+    from skglm.solvers import LBFGS
+    from skglm.datafits import Logistic
+
+    from skglm.utils.jit_compilation import compiled_clone
+
+
+class Solver(BaseSolver):
+    name = 'skglm'
+    stopping_strategy = "iteration"
+
+    install_cmd = 'conda'
+    requirements = [
+        'pip:git+https://github.com/scikit-learn-contrib/skglm.git@main'
+    ]
+
+    references = [
+        'Q. Bertrand and Q. Klopfenstein and P.-A. Bannier and G. Gidel'
+        'and M. Massias'
+        '"Beyond L1: Faster and Better Sparse Models with skglm", '
+        'https://arxiv.org/abs/2204.07826'
+    ]
+
+    def set_objective(self, X, y, lmbd):
+        self.X, self.y, self.lmbd = X, y, lmbd
+        n_samples = X.shape[0]
+
+        self.datafit = compiled_clone(Logistic())
+        self.penalty = compiled_clone(L2(lmbd / n_samples))
+
+        warnings.filterwarnings('ignore', category=ConvergenceWarning)
+        self.solver = LBFGS(tol=1e-20)
+
+        # cache Numba compilation
+        self.run(3)
+
+    def run(self, n_iter):
+        self.solver.max_iter = n_iter
+
+        self.coef_ = self.solver.solve(
+            self.X, self.y, self.datafit, self.penalty)[0]
+
+    @staticmethod
+    def get_next(stop_val):
+        return stop_val + 1
+
+    def get_result(self):
+        return self.coef_.flatten()