def test_SRM():
rot = np.array([[-0.89433495, -0.44719485, -0.01348182],
[-0.43426149, 0.87492975, -0.21427761],
[-0.10761949, 0.18578133, 0.97667976]])
data2 = np.dot(data1, rot)
result = align([data1, data2], align='SRM')
assert np.allclose(result[0], result[1], rtol=1)
def test_procrustes():
data1 = load('spiral')
rot = np.array([[-0.89433495, -0.44719485, -0.01348182],
[-0.43426149, 0.87492975, -0.21427761],
[-0.10761949, 0.18578133, 0.97667976]])
data2 = np.dot(data1, rot)
result = align([data1, data2])
assert np.allclose(result[0], result[1])
def test_hyper():
rot = np.array([[-0.89433495, -0.44719485, -0.01348182],
[-0.43426149, 0.87492975, -0.21427761],
[-0.10761949, 0.18578133, 0.97667976]])
data2 = np.dot(data1, rot)
result = align([data1, data2], align='hyper')
assert np.allclose(
result[0], result[1], rtol=1
) #note: this tolerance is probably too high, but fails at anything lower
def test_procrustes():
data = sio.loadmat('examples/sample_data/test_data.mat')
data1 = data['spiral']
rot = np.array([[-0.89433495, -0.44719485, -0.01348182],
[-0.43426149, 0.87492975, -0.21427761],
[-0.10761949, 0.18578133, 0.97667976]])
data2 = np.dot(data1, rot)
result = align([data1,data2])
assert np.allclose(result[0],result[1])
def test_srm():
data = sio.loadmat('examples/sample_data/test_data.mat')
data1 = data['spiral']
rot = np.array([[-0.89433495, -0.44719485, -0.01348182],
[-0.43426149, 0.87492975, -0.21427761],
[-0.10761949, 0.18578133, 0.97667976]])
data2 = np.dot(data1, rot)
result = align([data1,data2], method='SRM')
assert np.allclose(result[0],result[1], rtol=1) #note: this tolerance is probably too high, but fails at anything lower
def test_align_shapes():
# Should return data with the same shape as input data
aligned = align(weights)
assert all(al.shape == wt.shape for al, wt in zip(aligned, weights))
def test_align_geo():
aligned = align(geo)
assert np.allclose(aligned[0], aligned[1])
def main():
print('\nRUNNNING ANALYSIS \n')
print('\tRunning Alignment: ', ALIGN)
# Get list of available files
# Note: this currently excludes first subject, because they are weird.
dat_files = [file for file in os.listdir(DAT_PATH)[1:] if 'epo.fif' in file]
# Create time definition vector
times = np.arange(-1, 1, 1/100)
###################################################################################################
for filter_setting in FILTER_SETTINGS:
# Load (or reload) all subject data
all_subjs = [read_epochs(os.path.join(DAT_PATH, f_name),
preload=True, verbose=None) for f_name in dat_files]
print('\tRunning across {:1d} subjects.'.format(len(all_subjs)))
f_label, f_low, f_high = filter_setting
print('\t\tRunning filters:', f_label)
all_data, all_labels = [], []
for subj in all_subjs:
# Enforce a minimum number of trials - skip subj if not met
if len(subj) < N_EPOCHS:
continue
t_data, t_labels = extract_data(subj, l_freq=f_low, h_freq=f_high, resample=True)
all_data.append(t_data)
all_labels.append(t_labels)
###################################################################################################
## HYPERALIGNMENT
# Data organization - extract matrices, and flatten to continuous data
all_data_2d = [make_2d(dat) for dat in all_data]
# Do alignment
# Note: this also switches orientation (takes the transpose) to match hypertools
# This is beacause align assumes data in orientation of [n_samples x n_channels]
aligned_data = align([dat.T for dat in all_data_2d], align=ALIGN)
aligned_data = [dat.T for dat in aligned_data]
aligned_data = [make_3d(dat, N_EPOCHS) for dat in aligned_data]
###################################################################################################
## CLASSIFICATION
# Run within subject classification - non time resolved
within_scores = within_subj_classification(all_data, all_labels)
# Get average results - within and across subjects
within_subj_avgs = np.mean(within_scores, 1)
within_glob_avg = np.mean(within_subj_avgs)
# Within subject classification - time resolved
ts_within_scores = time_within_subj_classification(all_data, all_labels)
# Collapse across k-folds
ts_within_scores = np.mean(ts_within_scores, 2)
# Run prediction between subjects - on unaligned data
btwn_scores = btwn_subj_classication(all_data, all_labels)
# Get average results
avg_btwn_scores = np.mean(btwn_scores)
# Run prediction between subjects - on unaligned data and time-resolved
ts_btwn_scores = time_btwn_subj_classification(all_data, all_labels)
# Check within subject prediction of aligned data
within_al1_scores = within_subj_classification(aligned_data, all_labels)
# Run prediction between subjects - on aligned data
btwn_al_scores = btwn_subj_classication(aligned_data, all_labels)
# Get average results
avg_btwn_al_scores = np.mean(btwn_al_scores)
# Run prediction between subjects - on aligned data, across time points
ts_btwn_al_scores = time_btwn_subj_classification(aligned_data, all_labels)
# Calculate average across time points
avg_time_class_btwn_al = np.mean(ts_btwn_al_scores, 0)
# Set up data for plotting
results = [ts_within_scores, ts_btwn_scores, ts_btwn_al_scores]
labels = ['Within', 'Between', 'Aligned']
# Make an amazing plot
plot_results(times, results, labels, save_fig=True, save_name=f_label + '_' + ALIGN)
def test_align_reduce5d_shape():
# Should return data with same rowcnt but reduced to 5 columns
aligned = align(weights, ndims=5)
assert all(al.shape == (wt.shape[0], 5)
for al, wt in zip(aligned, weights))
