def feature_extraction(data, sfreq): #What features do I want??? #Kurtosis #line-length #entropy #Perform feature extraction for 2 second segments kurt = np.zeros((data.shape[0],data.shape[1]//2//sfreq)) s = np.zeros((data.shape[0],data.shape[1]//2//sfreq)) H = np.zeros((data.shape[0],data.shape[1]//2//sfreq)) segments = np.zeros((data.shape[0],data.shape[1]//2//sfreq,sfreq*2)) for n in range(0,data.shape[1]//2//sfreq): cur_data = data[:,n*sfreq*2:(n+1)*sfreq*2] segments[:,n,:] = cur_data kurt[:,n] = univariate.compute_kurtosis(cur_data) s[:,n] = univariate.compute_line_length(cur_data) H[:,n] = univariate.compute_spect_entropy(sfreq, cur_data, psd_method='welch') #Average across channels kurt_avg = np.mean(kurt,axis=0) s_avg = np.mean(s,axis=0)*np.power(10,6) H_avg = np.mean(H,axis=0) return kurt_avg, s_avg, H_avg
def test_spect_entropy():
expected = -(0.005 / 0.00625) * log(0.005 / 0.00625, 2.) - \
(0.00125 / 0.00625) * log(0.00125 / 0.00625, 2.)
assert_almost_equal(compute_spect_entropy(sfreq, data_sin), expected)
def test_shape_output_spect_entropy():
for j in range(n_epochs):
feat = compute_spect_entropy(sfreq, data[j, :, :])
assert_equal(feat.shape, (n_channels, ))