Function test_line_search_armijo

test/test_optim.py:132–181

(nx)

def test_line_search_armijo(nx):
    xk = np.array([[0.25, 0.25], [0.25, 0.25]])
    pk = np.array([[-0.25, 0.25], [0.25, -0.25]])
    gfk = np.array([[23.04273441, 23.0449082], [23.04273441, 23.0449082]])
    old_fval = -123.0

    xkb, pkb, gfkb = nx.from_numpy(xk, pk, gfk)

    def f(x):
        return 1.0

    # Should not throw an exception and return 0. for alpha
    alpha, a, b = ot.optim.line_search_armijo(f, xkb, pkb, gfkb, old_fval)
    alpha_np, anp, bnp = ot.optim.line_search_armijo(f, xk, pk, gfk, old_fval)
    assert a == anp
    assert b == bnp
    assert alpha == 0.0

    # check line search armijo
    def f(x):
        return nx.sum((x - 5.0) ** 2)

    def grad(x):
        return 2 * (x - 5.0)

    xk = nx.from_numpy(np.array([[[-5.0, -5.0]]]))
    pk = nx.from_numpy(np.array([[[100.0, 100.0]]]))
    gfk = grad(xk)
    old_fval = f(xk)

    # chech the case where the optimum is on the direction
    alpha, _, _ = ot.optim.line_search_armijo(f, xk, pk, gfk, old_fval)
    np.testing.assert_allclose(alpha, 0.1)

    # check the case where the direction is not far enough
    pk = nx.from_numpy(np.array([[[3.0, 3.0]]]))
    alpha, _, _ = ot.optim.line_search_armijo(f, xk, pk, gfk, old_fval, alpha0=1.0)
    np.testing.assert_allclose(alpha, 1.0)

    # check the case where checking the wrong direction
    alpha, _, _ = ot.optim.line_search_armijo(f, xk, -pk, gfk, old_fval)
    assert alpha <= 0

    # check the case where the point is not a vector
    xk = nx.from_numpy(np.array(-5.0))
    pk = nx.from_numpy(np.array(100.0))
    gfk = grad(xk)
    old_fval = f(xk)
    alpha, _, _ = ot.optim.line_search_armijo(f, xk, pk, gfk, old_fval)
    np.testing.assert_allclose(alpha, 0.1)


def test_line_search_armijo_dtype_device(nx):