def test_model_specification_error(self):
n_l, prior_l = 30, .3
n_u, prior_u = 200, .5
n_t = 100
x, y, x_tp, x_tn = gendata(n_l, prior_l, n_u, prior_u, n_t)
eta_list = [
pnu_mr.calc_etab(np.sum(y == +1), np.sum(y == -1), prior_u)
]
with self.assertRaises(ValueError):
pnu_mr.PNU_SL(x, y, prior_u, eta_list, model='DNN')
def test_linear_model(self):
n_l, prior_l = 30, .3
n_u, prior_u = 200, .5
n_t = 100
eta_list = np.arange(-.9, 1, .1)
x, y, x_tp, x_tn = gendata(n_l, prior_l, n_u, prior_u, n_t)
f_dec, outs, funcs = pnu_mr.PNU_SL(x,
y,
prior_u,
eta_list,
model='lm',
nargout=3)
err = 100 * calc_err(f_dec, x_tp, x_tn, prior_u)
def test_linear_model_one_eta(self):
n_l, prior_l = 30, .5
n_u, prior_u = 200, .3
n_t = 100
x, y, x_tp, x_tn = gendata(n_l, prior_l, n_u, prior_u, n_t)
eta_list = [
pnu_mr.calc_etab(np.sum(y == +1), np.sum(y == -1), prior_u)
]
f_dec, outs, funcs = pnu_mr.PNU_SL(x,
y,
prior_u,
eta_list,
model='lm',
nargout=3)
err = 100 * calc_err(f_dec, x_tp, x_tn, prior_u)
def test_model_nargout2(self):
n_l, prior_l = 30, .5
n_u, prior_u = 200, .3
n_t = 100
x, y, x_tp, x_tn = gendata(n_l, prior_l, n_u, prior_u, n_t)
eta_list = [
pnu_mr.calc_etab(np.sum(y == +1), np.sum(y == -1), prior_u)
]
lambda_list = [.1]
f_dec, outs = pnu_mr.PNU_SL(x,
y,
prior_u,
eta_list,
lambda_list=lambda_list,
model='lm',
nargout=2)
err = 100 * calc_err(f_dec, x_tp, x_tn, prior_u)
def test_gauss_model(self):
n_l, prior_l = 30, .3
n_u, prior_u = 200, .5
n_t = 100
lambda_list = [.1]
x, y, x_tp, x_tn = gendata(n_l, prior_l, n_u, prior_u, n_t)
eta_list = [
pnu_mr.calc_etab(np.sum(y == +1), np.sum(y == -1), prior_u)
]
f_dec, outs, funcs = pnu_mr.PNU_SL(x,
y,
prior_u,
eta_list,
model='gauss',
lambda_list=lambda_list,
nargout=3)
err = 100 * calc_err(f_dec, x_tp, x_tn, prior_u)
prior_u = .3
eta_list = np.arange(-.9, 1, .1)
n_trial = 20
best_err = np.inf
errs1 = np.empty(n_trial)
errs2 = np.empty((n_trial, len(eta_list)))
priors = np.empty(n_trial)
for ite in range(n_trial):
x, y, x_tp, x_tn = gendata(n_l, prior_l, n_u, prior_u, n_t)
priorh = cpe.cpe(x[y != 0, :], y[y != 0], x[y == 0, :])
f_dec, outs, funcs = pnu.PNU_SL(x,
y,
priorh,
eta_list,
model='lm',
nargout=3)
errs1[ite] = 100 * calc_err(f_dec, x_tp, x_tn, prior_u)
if errs1[ite] < best_err:
best_err = errs1[ite]
best_w = outs['w']
best_x, best_y = x, y
for ite_eta in range(len(eta_list)):
errs2[ite, ite_eta] = 100 * calc_err(funcs[ite_eta], x_tp, x_tn,
prior_u)
priors[ite] = priorh
print("Average of misclassification rates: {:.1f} ({:.2f})".format(
np.mean(errs1),