def split_channel(self, by):
""" Returns a list TemporalDataset splited by channels
Args:
by(String): the descriptor by which the splitting is made
Returns:
list of TemporalDataset,
split by the selected channel_descriptor
"""
unique_values, inverse = get_unique_inverse(
self.channel_descriptors[by])
dataset_list = []
for i_v, v in enumerate(unique_values):
selection = np.where(inverse == i_v)[0]
measurements = self.measurements[:, selection, :]
descriptors = self.descriptors.copy()
descriptors[by] = v
obs_descriptors = self.obs_descriptors
channel_descriptors = subset_descriptor(self.channel_descriptors,
selection)
time_descriptors = self.time_descriptors
dataset = TemporalDataset(measurements=measurements,
descriptors=descriptors,
obs_descriptors=obs_descriptors,
channel_descriptors=channel_descriptors,
time_descriptors=time_descriptors,
check_dims=False)
dataset_list.append(dataset)
return dataset_list
def split_obs(self, by):
""" Returns a list Datasets splited by obs
Args:
by(String): the descriptor by which the splitting is made
Returns:
list of Datasets, splitted by the selected obs_descriptor
"""
unique_values, inverse = get_unique_inverse(self.obs_descriptors[by])
dataset_list = []
for i_v, _ in enumerate(unique_values):
selection = np.where(inverse == i_v)[0]
measurements = self.measurements[selection, :]
descriptors = self.descriptors.copy()
obs_descriptors = subset_descriptor(self.obs_descriptors,
selection)
channel_descriptors = self.channel_descriptors
dataset = Dataset(measurements=measurements,
descriptors=descriptors,
obs_descriptors=obs_descriptors,
channel_descriptors=channel_descriptors,
check_dims=False)
dataset_list.append(dataset)
return dataset_list
def cov_from_unbalanced(dataset, obs_desc, dof=None, method='shrinkage_diag'):
"""
Estimates a covariance matrix from an unbalanced dataset, i.e. from a
dataset that contains different numbers of samples for different
stimuli.
Args:
dataset(data.Dataset):
rsatoolbox Dataset object
dof(int or list of int): degrees of freedom for covariance estimation
defaults to n_measurements - n_stimuli, should be corrected
if this is not the case
method(str): which estimate to use:
'diag': provides a diagonal matrix, i.e. univariate noise normalizer
'full': computes the sample covariance without shrinkage
'shrinkage_eye': shrinks the data covariance towards a multiple of the identity.
'shrinkage_diag': shrinks the covariance matrix towards the diagonal covariance matrix.
Returns:
numpy.ndarray (or list): sigma_p: covariance matrix over channels
"""
if isinstance(dataset, Iterable):
cov_mat = []
for i, dat in enumerate(dataset):
if dof is None:
cov_mat.append(
cov_from_unbalanced(dat, obs_desc=obs_desc, method=method))
elif isinstance(dof, Iterable):
cov_mat.append(
cov_from_unbalanced(dat,
obs_desc=obs_desc,
method=method,
dof=dof[i]))
else:
cov_mat.append(
cov_from_unbalanced(dat,
obs_desc=obs_desc,
method=method,
dof=dof))
else:
assert "Dataset" in str(
type(dataset)), "Provided object is not a dataset"
assert obs_desc in dataset.obs_descriptors.keys(), \
"obs_desc not contained in the dataset's obs_descriptors"
matrix = dataset.measurements
means, values, _ = average_dataset_by(dataset, obs_desc)
values, inverse = get_unique_inverse(dataset.obs_descriptors[obs_desc])
matrix -= means[inverse]
# calculate sample covariance matrix s
if dof is None:
dof = matrix.shape[0] - len(values)
cov_mat = _estimate_covariance(matrix, dof, method)
return cov_mat
def average_dataset_by(dataset, by):
"""
computes the average of a dataset per value of a descriptor
Args:
dataset(rsatoolbox.data.Dataset): the dataset to operate on
by(String): which obs_descriptor to split by
Returns:
numpy.ndarray: average: average activation vector
"""
unique_values, inverse = get_unique_inverse(dataset.obs_descriptors[by])
average = np.nan * np.empty(
(len(unique_values), dataset.measurements.shape[1]))
n_obs = np.nan * np.empty(len(unique_values))
for i_v, _ in enumerate(unique_values):
measurements = dataset.measurements[inverse == i_v, :]
average[i_v] = np.mean(measurements, axis=0)
n_obs[i_v] = measurements.shape[0]
return average, unique_values, n_obs