import numpy as np from sklearn.utils.optimize import _newton_cg from scipy.optimize import fmin_ncg from sklearn.utils._testing import assert_array_almost_equal def test_newton_cg(): # Test that newton_cg gives same result as scipy's fmin_ncg rng = np.random.RandomState(0) A = rng.normal(size=(10, 10)) x0 = np.ones(10) def func(x): Ax = A.dot(x) return .5 * (Ax).dot(Ax) def grad(x): return A.T.dot(A.dot(x)) def hess(x, p): return p.dot(A.T.dot(A.dot(x.all()))) def grad_hess(x): return grad(x), lambda x: A.T.dot(A.dot(x)) assert_array_almost_equal( _newton_cg(grad_hess, func, grad, x0, tol=1e-10)[0], fmin_ncg(f=func, x0=x0, fprime=grad, fhess_p=hess) )