def apply(self, chunk: np.ndarray):
if len(chunk) <= self.upd_n_samples:
x = self.buffer.update_buffer(chunk)
if self.samples_counter > self.buffer.buffer.shape[0]:
y = band_hilbert(x[:self.ada_n_taps], self.fs,
self.band)[self.ada_n_taps // 2]
self.rls.adapt(
y, x[self.ada_n_taps // 2 + self.delay - self.n_taps +
1:self.ada_n_taps // 2 + 1 + self.delay])
self.samples_counter += len(chunk)
return sg.lfilter(self.rls.w[::-1], [1.], x)[-len(chunk):]
else:
return rt_emulate(self, chunk, self.upd_n_samples)
fb_type = df.query('block_number==6')['block_name'].values[0]
# rename FB blocks to "FB"
df['block_name'] = df['block_name'].apply(lambda x: 'FB'
if x == fb_type in x else x)
# select FB blocks
df = df.loc[df['block_name'].isin(['Baseline'])]
# remove eyes artifacts ICA
df[channels] = df[channels].values.dot(eye_rejection_matrix)
raw_p4 = df['P4'].values
ba = sg.butter(1, np.array([1, 50]) / fs * 2, 'band')
env = np.abs(band_hilbert(sg.filtfilt(*ba, raw_p4), FS, band))
online_env = CFIRBandEnvelopeDetector(band,
FS,
IdentityFilter(),
n_taps=FS,
delay_ms=100).apply(raw_p4)
delay, corr = opt_delay_corr(env, online_env)
delays.append(delay)
corrds.append(corr)
print(subj_id)
import pylab as plt
plt.scatter(corrds, delays, alpha=0.7, s=50)
# plt.scatter(corrds2, delays2, alpha=0.7, s=50)
plt.plot()
import pandas as pd
dataset = "alpha2-delay-subj-21_12-06_12-15-09"
eeg_df = pd.read_pickle('data/rest_state_probes.pkl').query(
'dataset=="{}"'.format(dataset))
x = eeg_df['eeg'].iloc[20000:30000].values
#x = np.random.normal(size=5000)
band = [8, 12]
fs = 500
delay = 0
from release.utils import magnitude_spectrum
_, weights = magnitude_spectrum(eeg_df['eeg'].iloc[10000:20000].values, fs,
2000)
#weights = np.median(np.abs(sg.stft(eeg_df['eeg'].iloc[10000:20000].values, fs, nperseg=2000, nfft=2000, return_onesided=False))[2], 1)
y = np.roll(np.abs(band_hilbert(x, fs, band)), delay)
# rect_filter_y = RectEnvDetector(band, fs, delay, 150).apply(x)
whilbert_filter_y = np.abs(
WHilbertFilter(band, fs, delay, 500, 2000).apply(x))
cfir_filter_y = np.abs(
CFIRBandEnvelopeDetector(band, fs, delay, 500, 2000, weights).apply(x))
rlscfir_filter_y = np.abs(
AdaptiveCFIRBandEnvelopeDetector(band,
fs,
delay,
500,
2000,
weights=None,
max_chunk_size=1).apply(x))
ffiltar_filter_y = np.abs(
# subj eeg
data = probes_df.query('subj_id=="{}" '.format(subj_id))
data = data.loc[data.block_number.isin(FB_ALL)]
# subj fb type
fb_type = datasets_df.query(
'subj_id=={}'.format(subj_id))['fb_type'].values[0]
# subj band
band = np.array(
datasets_df.query('subj_id=={}'.format(subj_id))['band'].values[0])
# compute envelope
env = np.abs(band_hilbert(data['P4'].values, FS, band))
online_env = data['online_envelope']
delay, corr = opt_delay_corr(env, online_env)
delays.append(delay)
corrds.append(corr)
online_env = CFIRBandEnvelopeDetector(band,
FS,
IdentityFilter(),
n_taps=FS,
delay_ms=60).apply(data['P4'].values)
delay2, corr2 = opt_delay_corr(env, online_env)
delays2.append(delay2)
corrds2.append(corr2)
# subj eeg
data = probes_df.query('subj_id=="{}" '.format(subj_id))
# subj fb type
fb_type = datasets_df.query(
'subj_id=={}'.format(subj_id))['fb_type'].values[0]
# subj band
band = np.array(
datasets_df.query('subj_id=={}'.format(subj_id))['band'].values[0])
# block numbers utils
block_numbers = data['block_number'].values
ch_data = data.loc[data.block_number.isin(FB_ALL), channel].values
env = np.abs(band_hilbert(ch_data, FS, band))
# env = np.roll(np.abs(CFIRBandEnvelopeDetector(band, FS, 250).apply(ch_data)), -250)
photo_data = data.loc[data.block_number.isin(FB_ALL), 'PHOTO'].values
# photo_data = env
# compute envelope
spindles, spindle_starts = find_spindles(
env, photo_data > threshold_factor * np.median(photo_data))
spindle = np.mean(spindles[spindle_starts < len(env) // 2], 0)
norm = np.median(spindle[:FS])
spindle1 = spindle / norm
stats_df = stats_df.append(
pd.DataFrame({
'subj_id': subj_id,
'fb_type': fb_type,
'spindle': spindle1,
# GFP threshold artifact segments
# th = np.abs(df[channels[:-1]]).rolling(int(fs), center=True).mean().mean(1)
# df = df.loc[th < 100e-6]
# define SNR
x = df['P4'].values[5 * FS:]
# drop noisy datasets
if len(x) < N_SAMPLES_TRAIN + N_SAMPLES_TEST + 10 * FS: continue
# find individual alpha and snr
band, snr = individual_max_snr_band(x, FS)
print(j_dataset, band, snr)
# save x
an = band_hilbert(x, fs,
band)[5 * FS:N_SAMPLES_TRAIN + N_SAMPLES_TEST + 5 * FS]
x = x[5 * FS:N_SAMPLES_TRAIN + N_SAMPLES_TEST + 5 * FS]
# train band
band_train, snr_train = individual_max_snr_band(x[:N_SAMPLES_TRAIN], FS)
print(j_dataset, band_train, snr_train, '\n')
# save info
band, snr = individual_max_snr_band(x[:N_SAMPLES_TRAIN], FS)
eeg_df = eeg_df.append(pd.DataFrame(
{
'sim': 0,
'dataset': dataset,
'snr': snr,
'band_left_train': band_train[0],
'band_right_train': band_train[1],
plt.close()
# whilbert
delay = 100
filt = WHilbertFilter(band, FS, delay, 200, 2000)
y = filt.apply(eeg)
plt.figure(figsize=(4, 2))
plt.plot(y[slc].real, cm['b'], linestyle='-', alpha=1)
plt.plot(y[slc].imag, cm['b'], linestyle='--', alpha=0.5)
setup_gca()
plt.savefig(fdir + 'whilb_an.png', dpi=500, transparent=True)
plt.close()
win_slc = slice(550, 750)
win_x = band_hilbert(eeg[slc][win_slc], FS, band, 2000)
plt.figure(figsize=(4, 2))
plt.plot(eeg[slc], alpha=0)
#plt.plot(np.arange(wt)[win_slc], win_x.imag, cm['b'], linestyle='--', alpha=0.5)
plt.plot(np.arange(wt)[win_slc], win_x.real, cm['b'])
# plt.plot(np.arange(wt)[win_slc][100], win_x.real[100], 'k', marker='o', markersize=10)
# plt.plot(np.arange(wt)[win_slc][100], win_x.imag[100], 'k', marker='o', markersize=10, alpha=0.5)
setup_gca()
plt.savefig(fdir + 'whilb_filt_win.png', dpi=500, transparent=True)
plt.close()
plt.figure(figsize=(4, 2))
plt.plot(eeg[slc], alpha=0)
plt.plot(np.arange(wt)[win_slc],
win_x.imag,
