33 lines
771 B
Python
33 lines
771 B
Python
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)
|
|
)
|