def test_LearnerSCCS_confidence_intervals(self):
lrn = ConvSCCS(n_lags=self.n_lags, penalized_features=[])
coeffs, _ = lrn.fit(self.features, self.labels, self.censoring)
p_features, p_labels, p_censoring = lrn._preprocess_data(
self.features, self.labels, self.censoring)
confidence_intervals = lrn._bootstrap(p_features,
p_labels, p_censoring,
np.hstack(coeffs), 5, .90)
for i, c in enumerate(coeffs):
self.assertTrue(
np.all(confidence_intervals.lower_bound[i] <= c),
"lower bound of the confidence interval\
should be <= coeffs at index %i" % i)
self.assertTrue(
np.all(c <= confidence_intervals.upper_bound[i]),
"upper bound of the confidence interval\
should be >= coeffs at index %i" % i)
# Same with 0 lags
n_lags = np.zeros_like(self.n_lags, dtype='uint64')
lrn = ConvSCCS(n_lags=n_lags, penalized_features=[])
coeffs, _ = lrn.fit(self.features, self.labels, self.censoring)
p_features, p_labels, p_censoring = lrn._preprocess_data(
self.features, self.labels, self.censoring)
confidence_intervals = lrn._bootstrap(p_features,
p_labels, p_censoring,
np.hstack(coeffs), 5, .90)
for i, c in enumerate(coeffs):
self.assertTrue(
np.all(confidence_intervals.lower_bound[i] <= c),
"lower bound of the confidence interval\
should be <= coeffs at index %i" % i)
self.assertTrue(
np.all(c <= confidence_intervals.upper_bound[i]),
"upper bound of the confidence interval\
should be >= coeffs at index %i" % i)
def test_LearnerSCCS_score(self):
lrn = ConvSCCS(n_lags=self.n_lags,
penalized_features=[],
random_state=self.seed)
lrn.fit(self.features, self.labels, self.censoring)
self.assertEqual(lrn.score(),
lrn.score(self.features, self.labels, self.censoring))
def test_LearnerSCCS_fit_KFold_CV(self):
lrn = ConvSCCS(n_lags=self.n_lags,
penalized_features=np.arange(self.n_features),
random_state=self.seed,
C_tv=1e-1,
C_group_l1=1e-1)
lrn.fit(self.features, self.labels, self.censoring)
score = lrn.score()
tv_range = (-5, -1)
groupl1_range = (-5, -1)
lrn.fit_kfold_cv(self.features,
self.labels,
self.censoring,
C_tv_range=tv_range,
C_group_l1_range=groupl1_range,
n_cv_iter=4)
self.assertTrue(lrn.score() <= score)
def test_LearnerSCCS_fit(self):
seed = 42
n_lags = np.repeat(2, 2).astype('uint64')
sim = SimuSCCS(n_cases=800,
n_intervals=10,
n_features=2,
n_lags=n_lags,
verbose=False,
seed=seed,
exposure_type='multiple_exposures')
features, _, labels, censoring, coeffs = sim.simulate()
lrn = ConvSCCS(n_lags=n_lags,
penalized_features=[],
tol=0,
max_iter=10,
random_state=seed)
estimated_coeffs, _ = lrn.fit(features, labels, censoring)
np.testing.assert_almost_equal(np.hstack(estimated_coeffs),
np.hstack(coeffs),
decimal=1)
# print("C_tv : %f \n" % cv_track.best_model['C_tv'])
# print("C_group_l1 : %f \n" % cv_track.best_model['C_group_l1'])
# cv_track.plot_cv_report(35, 45)
# plt.show()
# confidence_intervals = cv_track.best_model['confidence_intervals']
# using the parameters resulting from cross-validation
learner = ConvSCCS(n_lags=n_lags.astype('uint64'),
penalized_features=np.arange(n_features),
random_state=42,
C_tv=270.2722840570933,
C_group_l1=5216.472772625124)
_, confidence_intervals = learner.fit(features,
labels,
censoring,
confidence_intervals=True,
n_samples_bootstrap=20)
# Plot estimated parameters
# get bootstrap confidence intervals
refitted_coeffs = confidence_intervals['refit_coeffs']
lower_bound = confidence_intervals['lower_bound']
upper_bound = confidence_intervals['upper_bound']
n_rows = int(np.ceil(n_features / 2))
remove_last_plot = (n_features % 2 != 0)
fig, axarr = plt.subplots(n_rows, 2, sharex=True, sharey=True, figsize=(10, 6))
y = confidence_intervals['refit_coeffs']
lb = confidence_intervals['lower_bound']