def test_likelihood_ratio_pac():
# Test that likelihood_ratio returns a near zero p_value with PAC
sdar = fast_fitted_model(StableDAR, sigin=_sigin)
dar = fast_fitted_model(DAR, sigin=_sigin)
ar = fast_fitted_model(AR, sigin=_sigin)
har = fast_fitted_model(HAR, sigin=_sigin)
for h0 in [ar, har]:
for h1 in [dar, sdar]:
p_value = h1.likelihood_ratio(h0)
assert_greater(1e-3, p_value)
def test_likelihood_ratio_noise():
# Test that likelihood_ratio returns a positive p_value with noise
sdar = fast_fitted_model(StableDAR, sigin=_noise)
dar = fast_fitted_model(DAR, sigin=_noise)
ar = fast_fitted_model(AR, sigin=_noise)
har = fast_fitted_model(HAR, sigin=_noise)
for h0 in [ar, har]:
for h1 in [dar, sdar]:
p_value = h1.likelihood_ratio(h0)
assert_greater(p_value, 0)
def test_dehumming_improve_snr():
# Test that dehumming removes the added noise
rng = np.random.RandomState(0)
fs = 250.
enf = 50.
clean = rng.randn(512)
noisy = clean.copy()
# add ENF noise
time = np.arange(512) / float(fs)
for harmonic in (1, 2):
noisy += np.sin(time * 2 * np.pi * enf * harmonic) / harmonic
for dim in [(1, 512), (512, ), (512, 1)]:
dehummed = dehummer(
noisy.reshape(*dim), fs, enf=enf)
assert_greater(norm(noisy - clean), norm(dehummed.ravel() - clean))
assert_array_equal(dehummed.shape, dim)
# test that the dehumming does not accept 2d arrays
assert_raises(ValueError, dehummer, noisy.reshape(2, 256), fs)
def test_surrogates():
# Test the surrogates comodulogram
for method in ALL_PAC_METRICS:
msg = 'with method=%s' % method
if method in BICOHERENCE_PAC_METRICS or method == 'jiang':
continue
n_surrogates = 10
est = ComodTest(method=method, n_surrogates=n_surrogates).fit(signal)
assert_array_equal(est.comod_.shape, (n_low, n_high), err_msg=msg)
assert_array_equal(est.surrogates_.shape,
(n_surrogates, n_low, n_high),
err_msg=msg)
# z-score
z_score = est.comod_z_score_
assert_array_equal(z_score.shape, (n_low, n_high), err_msg=msg)
if method != 'jiang': # 'jiang' method does not estimate CFC but CFD
assert_greater(z_score[1, 1], z_score[-1, -1], msg=msg)
# surrogate_max
surrogate_max = est.surrogate_max_
assert_array_equal(surrogate_max.shape, (n_surrogates, ))
assert_greater(est.comod_[1, 1], surrogate_max.max(), msg=msg)
assert_greater(surrogate_max.max(), est.comod_[-1, -1], msg=msg)
# Smoke test with contours in the plotting function
est.plot(contour_level=0.01, contour_method='comod_max')
est.plot(contour_level=3, contour_method='z_score')
plt.close('all')
def test_delay_shape():
est = fast_delay()
assert_equal(est.neg_log_likelihood_.shape, est.delays_ms_.shape)
assert_greater(est.neg_log_likelihood_.shape[0], 1)
i_best = np.nanargmin(est.neg_log_likelihood_)
assert_equal(est.best_delay_ms_, est.delays_ms_[i_best])