def test_newton_4(self):
# convergence relies heavily on x0
n, s = (40, 35)
# n, s = (5, 35)
A = mg.random_weighted_matrix_generator_dense(n,
sup_ext=100,
sym=False,
seed=s,
intweights=True)
A[0, :] = 0
bA = np.array([[1 if aa != 0 else 0 for aa in a] for a in A])
k_out = np.sum(bA, axis=1)
k_in = np.sum(bA, axis=0)
s_out = np.sum(A, axis=1)
s_in = np.sum(A, axis=0)
x0 = 0.1 * np.ones(4 * n)
# x0 = np.concatenate((-1*np.ones(2*n), np.ones(2*n)))
args = (k_out, k_in, s_out, s_in)
x0[np.concatenate(args) == 0] = 1e3
fun = lambda x: -mof.loglikelihood_prime_decm_exp(x, args)
fun_jac = lambda x: -mof.loglikelihood_hessian_decm_exp(x, args)
step_fun = lambda x: -mof.loglikelihood_decm_exp(x, args)
lin_fun = lambda x: mof.linsearch_fun_DECM_exp(x, (step_fun, ))
hes_reg = sof.matrix_regulariser_function
sol = sof.solver(
x0,
fun=fun,
step_fun=step_fun,
fun_jac=fun_jac,
linsearch_fun=lin_fun,
tol=1e-6,
eps=1e-5,
max_steps=100,
method="newton",
verbose=False,
regularise=True,
full_return=False,
linsearch=True,
hessian_regulariser=hes_reg,
)
sol = np.exp(-sol)
ek = mof.expected_decm(sol)
k = np.concatenate((k_out, k_in, s_out, s_in))
err = np.max(np.abs(ek - k))
# debug
# print(ek)
# print(k)
# print('\ntest 4: error = {}'.format(err))
# print('method: {}, matrix {}x{} with zeros'.format('newton', n,n))
# test result
self.assertTrue(err < 1e-1)
def test_quasinewton_1(self):
n, s = (4, 25)
A = mg.random_weighted_matrix_generator_dense(n,
sup_ext=10,
sym=False,
seed=s,
intweights=True)
A[0, :] = 0
bA = np.array([[1 if aa != 0 else 0 for aa in a] for a in A])
k_out = np.sum(bA, axis=1)
k_in = np.sum(bA, axis=0)
s_out = np.sum(A, axis=1)
s_in = np.sum(A, axis=0)
x0 = 0.9 * np.ones(n * 4)
args = (k_out, k_in, s_out, s_in)
fun = lambda x: -mof.loglikelihood_prime_decm_exp(x, args)
fun_jac = lambda x: -mof.loglikelihood_hessian_diag_decm_exp(x, args)
step_fun = lambda x: -mof.loglikelihood_decm_exp(x, args)
lin_fun = lambda x: mof.linsearch_fun_DECM_exp(x, (
mof.loglikelihood_decm_exp, args))
hes_reg = sof.matrix_regulariser_function
sol = sof.solver(
x0,
fun=fun,
step_fun=step_fun,
fun_jac=fun_jac,
linsearch_fun=lin_fun,
tol=1e-6,
eps=1e-10,
max_steps=300,
method="quasinewton",
verbose=False,
regularise=True,
full_return=False,
linsearch=True,
hessian_regulariser=hes_reg,
)
sol = np.exp(-sol)
ek = mof.expected_decm(sol)
k = np.concatenate((k_out, k_in, s_out, s_in))
err = np.max(np.abs(ek - k))
# debug
# print(ek)
# print(k)
# print('\ntest 0: error = {}'.format(err))
# print('method = {}, matrix {}x{}'.format('quasinewton', n, n))
# test result
self.assertTrue(err < 1e-1)