def test_disp(self):
"""Check if something is printed when `disp` is True."""
with Capturing() as output:
hfn(self.func, self.x0, self.hess_vec, disp=True)
self.assertTrue(
len(output) > 0,
'You should print the progress when `disp` is True.')
def test_default(self):
"""Check if everything works correctly with default parameters."""
with Capturing() as output:
x_min, f_min, status = hfn(self.func, self.x0, self.hess_vec)
assert_equal(status, 0)
self.assertTrue(norm(self.A.dot(x_min) - self.b, np.inf) <= 1e-4)
self.assertTrue(abs(f_min) <= 1e-8)
self.assertTrue(
len(output) == 0, 'You should not print anything by default.')
def test_trace(self):
"""Check if the history is returned correctly when `trace` is True."""
x_min, f_min, status, hist = hfn(self.func,
self.x0,
self.hess_vec,
trace=True)
self.assertTrue(isinstance(hist['f'], np.ndarray))
self.assertTrue(isinstance(hist['norm_g'], np.ndarray))
self.assertTrue(isinstance(hist['n_evals'], np.ndarray))
self.assertTrue(isinstance(hist['elaps_t'], np.ndarray))
assert_equal(len(hist['norm_g']), len(hist['f']))
assert_equal(len(hist['n_evals']), len(hist['f']))
assert_equal(len(hist['elaps_t']), len(hist['f']))
# make sure `hist['n_evals']` is a cumulative sum of integers
assert_equal(np.round(hist['n_evals']), hist['n_evals'])
self.assertTrue(np.all(hist['n_evals'] >= 0))
self.assertTrue(np.all(hist['n_evals'][1:] - hist['n_evals'][:-1] > 0))
'disp': False,
'maxiter': 100
},
tol=1e-5,
jac=True))['fun']
print("Optimal Value Found")
# Newton method
w_opt_newton, hist_newton = newton(oracle_hess,
w0,
disp=False,
maxiter=100,
tol=1e-10)
w_opt_hfn, hist_hfn = hfn(oracle,
w0,
hessvec,
disp=False,
maxiter=200,
tol=1e-7)
# optimal_value = hist_newton['f'][-1]
# Plotting
plot_performance(hist_newton['f'],
optimal_value,
'b',
"Метод Ньютона",
time_vec=hist_newton['elaps'])
plot_performance(hist_hfn['f'],
optimal_value,
'r',
"Неточный Метод\n Ньютона",
mydisp=True,
method="CG",
options={'maxiter': 10})
print("BFGS")
optimal_value_bfgs, hist_bfgs = minimize_wrapper(fun,
w0,
mydisp=True,
method="L-BFGS-B",
options={
'ftol': 0,
'maxiter': 20
})
print("Inexact Newton")
w_opt_hfn, hist_hfn = hfn(oracle, w0, hessvec, disp=True, maxiter=10, tol=1e-9)
print("Newton")
w_opt_newton, hist_newton = newton(oracle_hess,
w0,
disp=True,
maxiter=10,
tol=1e-9)
# Plotting
# print(optimal_value_cg['fun'])
# print(optimal_value_bfgs['fun'])
# print((hist_hfn['f'])[-1])
optimal_value = min(
[optimal_value_cg['fun'], optimal_value_bfgs['fun'], (hist_hfn['f'])[-1]])
def fun(w):
f, g = logreg(w.reshape((w.shape[0], 1)), Z_mat, gamma, hess=False)
return f, g[:, 0]
# def grad(w):
# return (logreg(w.reshape((w.shape[0], 1)), Z_mat, gamma, hess=True)[1])[:, 0]
optimal_value = (minimize(fun, w0[:, 0], method="CG",
options={'disp': False, 'maxiter': 100}, tol=1e-5, jac=True))['fun']
print("Optimal Value Found")
# Newton method
w_opt_newton, hist_newton = newton(oracle_hess, w0, disp=False, maxiter=100, tol=1e-10)
w_opt_hfn, hist_hfn = hfn(oracle, w0, hessvec, disp=False, maxiter=200, tol=1e-7)
# optimal_value = hist_newton['f'][-1]
# Plotting
plot_performance(hist_newton['f'], optimal_value, 'b', "Метод Ньютона", time_vec=hist_newton['elaps'])
plot_performance(hist_hfn['f'], optimal_value, 'r', "Неточный Метод\n Ньютона", time_vec=hist_hfn['elaps'])
plt.ylabel(r'$\log(f_k - f_*)$')
plt.xlabel("Время (сек)")
if real_data:
title = "Набор данных " + data_name
else:
title = "Сгенерированные данные"
title += ", D = " + str(dim + 1) + ", N = " + str(data_size) + "."
plt.title(title)
plt.legend()
def test_c2(self):
"""Check if argument `c2` is supported."""
hfn(self.func, self.x0, self.hess_vec, c2=0.1)
def test_c1(self):
"""Check if argument `c1` is supported."""
hfn(self.func, self.x0, self.hess_vec, c1=0.2)
def test_max_iter(self):
"""Check if argument `max_iter` is supported."""
hfn(self.func, self.x0, self.hess_vec, max_iter=15)
def test_tol(self):
"""Check if argument `tol` is supported."""
hfn(self.func, self.x0, self.hess_vec, tol=1e-6)