def test_standardize():
rng = np.random.RandomState(42)
n_features = 10
n_samples = 17
# Create random signals with offsets
a = rng.random_sample((n_samples, n_features))
a += np.linspace(0, 2., n_features)
# transpose array to fit _standardize input.
# Without trend removal
b = nisignal._standardize(a, standardize='zscore')
stds = np.std(b)
np.testing.assert_almost_equal(stds, np.ones(n_features))
np.testing.assert_almost_equal(b.sum(axis=0), np.zeros(n_features))
# With trend removal
a = np.atleast_2d(np.linspace(0, 2., n_features)).T
b = nisignal._standardize(a, detrend=True, standardize=False)
np.testing.assert_almost_equal(b, np.zeros(b.shape))
length_1_signal = np.atleast_2d(np.linspace(0, 2., n_features))
np.testing.assert_array_equal(
length_1_signal,
nisignal._standardize(length_1_signal, standardize='zscore'))
def test_standardize():
rand_gen = np.random.RandomState(0)
n_features = 10
n_samples = 17
# Create random signals with offsets
a = rand_gen.random_sample((n_samples, n_features))
a += np.linspace(0, 2., n_features)
# transpose array to fit _standardize input.
# Without trend removal
b = nisignal._standardize(a, normalize=True)
energies = (b ** 2).sum(axis=0)
np.testing.assert_almost_equal(energies, np.ones(n_features))
np.testing.assert_almost_equal(b.sum(axis=0), np.zeros(n_features))
# With trend removal
a = np.atleast_2d(np.linspace(0, 2., n_features)).T
b = nisignal._standardize(a, detrend=True, normalize=False)
np.testing.assert_almost_equal(b, np.zeros(b.shape))
length_1_signal = np.atleast_2d(np.linspace(0, 2., n_features))
np.testing.assert_array_equal(length_1_signal,
nisignal._standardize(length_1_signal,
normalize=True))
def test_standardize():
rand_gen = np.random.RandomState(0)
n_features = 10
n_samples = 17
# Create random signals with offsets
a = rand_gen.random_sample((n_samples, n_features))
a += np.linspace(0, 2., n_features)
# transpose array to fit _standardize input.
# Without trend removal
b = nisignal._standardize(a, normalize=True)
energies = (b ** 2).sum(axis=0)
np.testing.assert_almost_equal(energies, np.ones(n_features))
np.testing.assert_almost_equal(b.sum(axis=0), np.zeros(n_features))
# With trend removal
a = np.atleast_2d(np.linspace(0, 2., n_features)).T
b = nisignal._standardize(a, detrend=True, normalize=False)
np.testing.assert_almost_equal(b, np.zeros(b.shape))
length_1_signal = np.atleast_2d(np.linspace(0, 2., n_features))
np.testing.assert_array_equal(length_1_signal,
nisignal._standardize(length_1_signal,
normalize=True))
def transform(self, X, vectorize=True, confounds=None):
"""Apply transform to covariances matrices to get the connectivity
matrices for the chosen kind.
Parameters
----------
X : list of numpy.ndarray with shapes (n_samples, n_features)
The input subjects time series.
vectorize : bool default=True
If True, flattened lower triangular part of the connectivity
matrices will be computed and returned.
confounds: CSV file or array-like, optional
This parameter is passed to signal.clean. Please see the related
documentation for details.
shape: (number of scans, number of confounds)
Returns
-------
output : numpy.ndarray, shape (n_samples, n_features, n_features)
The transformed connectivity matrices.
"""
if self.kind == 'correlation':
covariances_std = [
self.cov_estimator_.fit(
signal._standardize(x, detrend=False,
normalize=True)).covariance_ for x in X
]
connectivities = [_cov_to_corr(cov) for cov in covariances_std]
else:
covariances = [self.cov_estimator_.fit(x).covariance_ for x in X]
if self.kind == 'covariance':
connectivities = covariances
elif self.kind == 'tangent':
connectivities = [
_map_eigenvalues(
np.log,
self.whitening_.dot(cov).dot(self.whitening_))
for cov in covariances
]
elif self.kind == 'precision':
connectivities = [linalg.inv(cov) for cov in covariances]
elif self.kind == 'partial correlation':
connectivities = [
_prec_to_partial(linalg.inv(cov)) for cov in covariances
]
else:
raise ValueError('Allowed connectivity kinds are '
'"correlation", '
'"partial correlation", "tangent", '
'"covariance" and "precision", got kind '
'"{}"'.format(self.kind))
connectivities = np.array(connectivities)
if vectorize:
connectivities = sym_to_vec(connectivities, confounds=confounds)
return connectivities
def transform(self, X, vectorize=True, confounds=None):
"""Apply transform to covariances matrices to get the connectivity
matrices for the chosen kind.
Parameters
----------
X : list of numpy.ndarray with shapes (n_samples, n_features)
The input subjects time series.
vectorize : bool default=True
If True, flattened lower triangular part of the connectivity
matrices will be computed and returned.
confounds: CSV file or array-like, optional
This parameter is passed to signal.clean. Please see the related
documentation for details.
shape: (number of scans, number of confounds)
Returns
-------
output : numpy.ndarray, shape (n_samples, n_features, n_features)
The transformed connectivity matrices.
"""
print("Connectome matrices estimation")
if self.kind == 'correlation':
covariances_std = [self.cov_estimator_.fit(
signal._standardize(x, detrend=False, normalize=True)
).covariance_ for x in X]
connectivities = [_cov_to_corr(cov) for cov in covariances_std]
else:
covariances = [self.cov_estimator_.fit(x).covariance_ for x in X]
if self.kind == 'covariance':
connectivities = covariances
elif self.kind == 'tangent':
connectivities = [_map_eigenvalues(np.log, self.whitening_.dot(
cov).dot(self.whitening_))
for cov in covariances]
elif self.kind == 'precision':
connectivities = [linalg.inv(cov) for cov in covariances]
elif self.kind == 'partial correlation':
connectivities = [_prec_to_partial(linalg.inv(cov))
for cov in covariances]
else:
raise ValueError('Allowed connectivity kinds are '
'"correlation", '
'"partial correlation", "tangent", '
'"covariance" and "precision", got kind '
'"{}"'.format(self.kind))
connectivities = np.array(connectivities)
if vectorize:
connectivities = sym_to_vec(connectivities, confounds=confounds)
return connectivities
