feat(numeric_solver): add a new numeric algo 'BFGS'

use scipy.optimize.minimize with method 'BFGS'

Author: bbayukari

pytest/test_args.py

@@ -20,15 +20,26 @@
 solvers_ids = ("scope", "Base")  # , "GraHTP", "GraSP", "IHT")
 
 
+@pytest.mark.parametrize("model", models, ids=models_ids)
+@pytest.mark.parametrize("solver_creator", solvers, ids=solvers_ids)
+def test_numeric_solver(model, solver_creator):
+    from skscope.numeric_solver import convex_solver_BFGS
+
+    solver = solver_creator(
+        model["n_features"], model["n_informative"], numeric_solver=convex_solver_BFGS
+    )
+    solver.solve(model["loss"], jit=True)
+
+    assert set(model["support_set"]) == set(solver.get_support())
+
+
 @pytest.mark.parametrize("model", models, ids=models_ids)
 @pytest.mark.parametrize("solver_creator", solvers, ids=solvers_ids)
 def test_init_support_set(model, solver_creator):
     solver = solver_creator(model["n_features"], model["n_informative"])
     solver.solve(model["loss"], init_support_set=[0, 1, 2], jit=True)
-    solver.get_result()
-    solver.get_estimated_params()
-    solver.get_support()
-    assert set(model["support_set"]) == set(solver.support_set)
+
+    assert set(model["support_set"]) == set(solver.get_support())
 
 
 @pytest.mark.parametrize("model", models, ids=models_ids)

setup.py

@@ -18,6 +18,7 @@
     "win-arm64": "ARM64",
 }
 
+
 def get_info():
     # get information from `__init__.py`
     labels = ["__version__", "__author__"]
@@ -39,6 +40,7 @@ def __init__(self, name, sourcedir=""):
         self.sourcedir = os.path.abspath(sourcedir)
         self.parallel = 4
 
+
 class CMakeBuild(build_ext):
     def build_extension(self, ext):
         extdir = os.path.abspath(os.path.dirname(self.get_ext_fullpath(ext.name)))
@@ -59,7 +61,7 @@ def build_extension(self, ext):
         cmake_args = [
             f"-DCMAKE_LIBRARY_OUTPUT_DIRECTORY={extdir}",
             f"-DPYTHON_EXECUTABLE={sys.executable}",
-            f"-DCMAKE_BUILD_TYPE={cfg}"
+            f"-DCMAKE_BUILD_TYPE={cfg}",
         ]
         build_args = []
         # Adding CMake arguments set as environment variable
@@ -128,27 +130,29 @@ def build_extension(self, ext):
         subprocess.check_call(["cmake", ext.sourcedir] + cmake_args, cwd=build_temp)
         subprocess.check_call(["cmake", "--build", "."] + build_args, cwd=build_temp)
 
-with open(os.path.join(CURRENT_DIR, 'README.md'), encoding='utf-8') as f:
+
+with open(os.path.join(CURRENT_DIR, "README.md"), encoding="utf-8") as f:
     long_description = f.read()
 
 package_info = get_info()
 
 setup(
-    name='skscope',
-    version=package_info['__version__'],
-    author=package_info['__author__'],
+    name="skscope",
+    version=package_info["__version__"],
+    author=package_info["__author__"],
     author_email="[email protected]",
     maintainer="Zezhi Wang",
     maintainer_email="[email protected]",
     packages=find_packages(),
-    description="Sparsity-Constraint OPtimization via itErative-algorithm", 
+    description="Sparsity-Constraint OPtimization via itErative-algorithm",
     long_description=long_description,
-    long_description_content_type='text/markdown',
+    long_description_content_type="text/markdown",
     install_requires=[
         "numpy",
         "scikit-learn>=1.2.2",
         "jax[cpu]",
         "nlopt",
+        "scipy",
     ],
     license="MIT",
     url="https://skscope.readthedocs.io",
@@ -176,7 +180,7 @@ def build_extension(self, ext):
         "Programming Language :: Python :: 3.10",
         "Programming Language :: Python :: 3.11",
     ],
-    python_requires='>=3.8', 
+    python_requires=">=3.8",
     ext_modules=[CMakeExtension("skscope._scope")],
-    cmdclass={"build_ext": CMakeBuild}
+    cmdclass={"build_ext": CMakeBuild},
 )

skscope/numeric_solver.py

@@ -7,6 +7,7 @@
 import numpy as np
 import math
 import nlopt
+from scipy.optimize import minimize
 
 
 def convex_solver_nlopt(
@@ -69,3 +70,24 @@ def cache_opt_fn(x, grad):
     except RuntimeError:
         init_params[optim_variable_set] = best_params
         return best_loss, init_params
+
+
+def convex_solver_BFGS(
+    objective_func,
+    value_and_grad,
+    init_params,
+    optim_variable_set,
+    data,
+):
+    def fun(x):
+        init_params[optim_variable_set] = x
+        return objective_func(init_params, data)
+
+    def jac(x):
+        init_params[optim_variable_set] = x
+        _, grad = value_and_grad(init_params, data)
+        return grad[optim_variable_set]
+
+    res = minimize(fun, init_params[optim_variable_set], method="BFGS", jac=jac)
+    init_params[optim_variable_set] = res.x
+    return res.fun, init_params