def _stack_list(data, data_labels, w):
"""Construct a numpy array by stacking arrays in a list
Parameter
----------
data : 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.
data_labels : list of arrays of int, element i has shape=[samples_i]
Each element in the list contains the labels for the samples in
data.
w : list of array, element i has shape=[voxels_i, features]
The orthogonal transforms (mappings) :math:`W_i` for each subject.
Returns
-------
data_stacked : 2D array, shape=[samples, features]
The data samples from all subjects are stacked into a single
2D array, where "samples" is the sum of samples_i.
labels_stacked : array, shape=[samples,]
The labels from all subjects are stacked into a single
array, where "samples" is the sum of samples_i.
weights : array, shape=[samples,]
The number of samples of the subject that are related to that
sample. They become a weight per sample in the MLR loss.
"""
labels_stacked = utils.concatenate_list(data_labels)
weights = np.empty((labels_stacked.size, ))
data_shared = [None] * len(data)
curr_samples = 0
for s in range(len(data)):
if data[s] is not None:
subject_samples = data[s].shape[1]
curr_samples_end = curr_samples + subject_samples
weights[curr_samples:curr_samples_end] = subject_samples
data_shared[s] = w[s].T.dot(data[s])
curr_samples += data[s].shape[1]
data_stacked = utils.concatenate_list(data_shared, axis=1).T
return data_stacked, labels_stacked, weights
def _stack_list(data, data_labels, w):
"""Construct a numpy array by stacking arrays in a list
Parameter
----------
data : 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.
data_labels : list of arrays of int, element i has shape=[samples_i]
Each element in the list contains the labels for the samples in
data.
w : list of array, element i has shape=[voxels_i, features]
The orthogonal transforms (mappings) :math:`W_i` for each subject.
Returns
-------
data_stacked : 2D array, shape=[samples, features]
The data samples from all subjects are stacked into a single
2D array, where "samples" is the sum of samples_i.
labels_stacked : array, shape=[samples,]
The labels from all subjects are stacked into a single
array, where "samples" is the sum of samples_i.
weights : array, shape=[samples,]
The number of samples of the subject that are related to that
sample. They become a weight per sample in the MLR loss.
"""
labels_stacked = utils.concatenate_list(data_labels)
weights = np.empty((labels_stacked.size,))
data_shared = [None] * len(data)
curr_samples = 0
for s in range(len(data)):
if data[s] is not None:
subject_samples = data[s].shape[1]
curr_samples_end = curr_samples + subject_samples
weights[curr_samples:curr_samples_end] = subject_samples
data_shared[s] = w[s].T.dot(data[s])
curr_samples += data[s].shape[1]
data_stacked = utils.concatenate_list(data_shared, axis=1).T
return data_stacked, labels_stacked, weights
def test_concatenate_list():
from brainiak.utils.utils import concatenate_list
import numpy as np
l = [None] * 5
l[1] = np.array([0, 1, 2])
l[3] = np.array([3, 4])
r = concatenate_list(l, axis=0)
assert np.all(np.arange(5) == r), "Invalid concatenation of a list of arrays"
def test_concatenate_list():
from brainiak.utils.utils import concatenate_list
import numpy as np
l = [None] * 5
l[1] = np.array([0, 1, 2])
l[3] = np.array([3, 4])
r = concatenate_list(l, axis=0)
assert np.all(
np.arange(5) == r), "Invalid concatenation of a list of arrays"
def _init_classes(self, y):
"""Map all possible classes to the range [0,..,C-1]
Parameters
----------
y : list of arrays of int, each element has shape=[samples_i,]
Labels of the samples for each subject
Returns
-------
new_y : list of arrays of int, each element has shape=[samples_i,]
Mapped labels of the samples for each subject
..note::
The mapping of the classes is saved in the attribute classes_.
"""
self.classes_ = unique_labels(utils.concatenate_list(y))
new_y = [None] * len(y)
for s in range(len(y)):
new_y[s] = np.digitize(y[s], self.classes_) - 1
return new_y
def _init_classes(self, y):
"""Map all possible classes to the range [0,..,C-1]
Parameters
----------
y : list of arrays of int, each element has shape=[samples_i,]
Labels of the samples for each subject
Returns
-------
new_y : list of arrays of int, each element has shape=[samples_i,]
Mapped labels of the samples for each subject
Note
----
The mapping of the classes is saved in the attribute classes_.
"""
self.classes_ = unique_labels(utils.concatenate_list(y))
new_y = [None] * len(y)
for s in range(len(y)):
new_y[s] = np.digitize(y[s], self.classes_) - 1
return new_y
