def log_lik(subject_cov, group_cov, sigma, whiten=True):
if whiten:
whitening = spd_manifold.inv_sqrtm(group_cov)
subject_cov = np.dot(np.dot(whitening, subject_cov), whitening)
group_cov = np.eye(n_rois)
return (-n_rois**2 * np.log(sigma) - 1 / (2 * sigma**2) *
(np.sum((group_cov - subject_cov)**2) +
np.sum(np.diag(group_cov - subject_cov)**2)))
def log_lik(subject_cov, group_cov, sigma, whiten=True):
if whiten:
whitening = spd_manifold.inv_sqrtm(group_cov)
subject_cov = np.dot(np.dot(whitening, subject_cov), whitening)
group_cov = np.eye(n_rois)
return ( -n_rois**2*np.log(sigma)
- 1/(2*sigma**2) * (
np.sum((group_cov - subject_cov)**2)
+ np.sum(np.diag(group_cov - subject_cov)**2)
))
def fit(self, group_covs):
if self.whiten:
self.mean_cov = mean_cov = spd_manifold.log_mean(group_covs)
whitening = spd_manifold.inv_sqrtm(mean_cov)
group_covs = [
np.dot(np.dot(whitening, g), whitening) for g in group_covs
]
mean_cov = np.eye(n_rois)
else:
self.mean_cov = mean_cov = np.mean(group_covs, axis=0)
self.sigma = 1. / n_rois * np.sqrt(1. / len(group_covs) * np.sum(
np.sum((mean_cov - g)**2) + np.sum(np.diag(mean_cov - g)**2)
for g in group_covs))
return self
def fit(self, group_covs):
if self.whiten:
self.mean_cov = mean_cov = spd_manifold.log_mean(group_covs)
whitening = spd_manifold.inv_sqrtm(mean_cov)
group_covs = [np.dot(np.dot(whitening, g), whitening)
for g in group_covs]
mean_cov = np.eye(n_rois)
else:
self.mean_cov = mean_cov = np.mean(group_covs, axis=0)
self.sigma = 1./n_rois* np.sqrt(1./len(group_covs)*
np.sum(
np.sum((mean_cov - g)**2)
+ np.sum(np.diag(mean_cov - g)**2)
for g in group_covs
))
return self
