def _sssrm(self, data_align, data_sup, labels):
"""Block-Coordinate Descent algorithm for fitting SS-SRM.
Parameters
----------
data_align : list of 2D arrays, element i has shape=[voxels_i, n_align]
Each element in the list contains the fMRI data for alignment of
one subject. There are n_align samples for each subject.
data_sup : list of 2D arrays, element i has shape=[voxels_i, samples_i]
Each element in the list contains the fMRI data of one subject for
the classification task.
labels : list of arrays of int, element i has shape=[samples_i]
Each element in the list contains the labels for the data samples
in data_sup.
Returns
-------
w : list of array, element i has shape=[voxels_i, features]
The orthogonal transforms (mappings) :math:`W_i` for each subject.
s : array, shape=[features, samples]
The shared response.
"""
classes = self.classes_.size
# Initialization:
self.random_state_ = np.random.RandomState(self.rand_seed)
random_states = [
np.random.RandomState(self.random_state_.randint(2**32))
for i in range(len(data_align))
]
# Set Wi's to a random orthogonal voxels by TRs
w, _ = srm._init_w_transforms(data_align, self.features, random_states)
# Initialize the shared response S
s = SSSRM._compute_shared_response(data_align, w)
# Initialize theta and bias
theta, bias = self._update_classifier(data_sup, labels, w, classes)
# calculate and print the objective function
if logger.isEnabledFor(logging.INFO):
objective = self._objective_function(data_align, data_sup, labels,
w, s, theta, bias)
logger.info('Objective function %f' % objective)
# Main loop:
for iteration in range(self.n_iter):
logger.info('Iteration %d' % (iteration + 1))
# Update the mappings Wi
w = self._update_w(data_align, data_sup, labels, w, s, theta, bias)
# Output the objective function
if logger.isEnabledFor(logging.INFO):
objective = self._objective_function(data_align, data_sup,
labels, w, s, theta, bias)
logger.info('Objective function after updating Wi %f' %
objective)
# Update the shared response S
s = SSSRM._compute_shared_response(data_align, w)
# Output the objective function
if logger.isEnabledFor(logging.INFO):
objective = self._objective_function(data_align, data_sup,
labels, w, s, theta, bias)
logger.info('Objective function after updating S %f' %
objective)
# Update the MLR classifier, theta and bias
theta, bias = self._update_classifier(data_sup, labels, w, classes)
# Output the objective function
if logger.isEnabledFor(logging.INFO):
objective = self._objective_function(data_align, data_sup,
labels, w, s, theta, bias)
logger.info('Objective function after updating MLR %f' %
objective)
return w, s, theta, bias
def _sssrm(self, data_align, data_sup, labels):
"""Block-Coordinate Descent algorithm for fitting SS-SRM.
Parameters
----------
data_align : list of 2D arrays, element i has shape=[voxels_i, n_align]
Each element in the list contains the fMRI data for alignment of
one subject. There are n_align samples for each subject.
data_sup : list of 2D arrays, element i has shape=[voxels_i, samples_i]
Each element in the list contains the fMRI data of one subject for
the classification task.
labels : list of arrays of int, element i has shape=[samples_i]
Each element in the list contains the labels for the data samples
in data_sup.
Returns
-------
w : list of array, element i has shape=[voxels_i, features]
The orthogonal transforms (mappings) :math:`W_i` for each subject.
s : array, shape=[features, samples]
The shared response.
"""
classes = self.classes_.size
# Initialization:
np.random.seed(self.rand_seed)
# Set Wi's to a random orthogonal voxels by TRs
w, _ = srm._init_w_transforms(data_align, self.features)
# Initialize the shared response S
s = SSSRM._compute_shared_response(data_align, w)
# Initialize theta and bias
theta, bias = self._update_classifier(data_sup, labels, w, classes)
# calculate and print the objective function
if logger.isEnabledFor(logging.INFO):
objective = self._objective_function(data_align, data_sup, labels,
w, s, theta, bias)
logger.info('Objective function %f' % objective)
# Main loop:
for iteration in range(self.n_iter):
logger.info('Iteration %d' % (iteration + 1))
# Update the mappings Wi
w = self._update_w(data_align, data_sup, labels, w, s, theta, bias)
# Output the objective function
if logger.isEnabledFor(logging.INFO):
objective = self._objective_function(data_align, data_sup,
labels, w, s, theta, bias)
logger.info('Objective function after updating Wi %f'
% objective)
# Update the shared response S
s = SSSRM._compute_shared_response(data_align, w)
# Output the objective function
if logger.isEnabledFor(logging.INFO):
objective = self._objective_function(data_align, data_sup,
labels, w, s, theta, bias)
logger.info('Objective function after updating S %f'
% objective)
# Update the MLR classifier, theta and bias
theta, bias = self._update_classifier(data_sup, labels, w, classes)
# Output the objective function
if logger.isEnabledFor(logging.INFO):
objective = self._objective_function(data_align, data_sup,
labels, w, s, theta, bias)
logger.info('Objective function after updating MLR %f'
% objective)
return w, s, theta, bias