if lapl:
x = x.dot(montage.make_laplacian_proj('EEG'))
del df
x *= 1e6
x_split = np.split(x[:x.shape[0] - x.shape[0] % 10], 10)
X_split = []
Y_split = []
delay = 0
n_taps = 200
band = (8, 12)
for x in x_split:
X, Y = get_XY(x, band_hilbert(x[:, channels.index('P4')], fs, band),
n_taps, delay)
X_split.append(X)
Y_split.append(Y)
del x
corrs = np.zeros((10 - 3 - 2) * 3)
mses = np.zeros((10 - 3 - 2) * 3)
corrs_one = np.zeros((10 - 3 - 2) * 3)
mses_one = np.zeros((10 - 3 - 2) * 3)
for k in range(5):
X_train = np.concatenate(X_split[k:3 + k])
Y_train = np.concatenate(Y_split[k:3 + k])
X_val = X_split[k + 4]
nperseg=n_fft,
return_onesided=False,
noverlap=int(n_fft * 0.9))
mask = np.abs(spec).mean(1)
w = np.diag(mask)
for j, d in enumerate(delays):
print('delay', d)
# freq. domain ideal filter
H = get_ideal_H(n_fft, fs, band, d)
b_cfir_f = cLS(F, H, 0)
b_cfir_wf = cLS(np.dot(w, F), np.dot(w, H), 0)
X_train, Y_train = get_XY(x_train, y_train, n_taps, d)
b_cfir_t = cLS(X_train, Y_train, 0)
if d >= 0:
tar = np.abs(y_test)[:-d]
b_fir_band = firwin(d, band / fs * 2, pass_zero=False)
b_fir_smooth = firwin(d, 2 / fs * 2)
rec_fir = lfilter(b_fir_smooth, [1],
np.abs(lfilter(b_fir_band, [1], x_test)))[d:]
rec_cfir_f = np.abs(lfilter(b_cfir_f, [1], x_test))[d:]
rec_cfir_wf = np.abs(lfilter(b_cfir_wf, [1], x_test))[d:]
rec_cfir_t = np.abs(lfilter(b_cfir_t, [1], x_test))[d:]
corrs_fir[k, j] = np.corrcoef(rec_fir, tar)[0, 1]
F = np.array([
np.exp(-2j * np.pi / n_fft * k * np.arange(n_taps))
for k in np.arange(n_fft)
])
fig, axes = plt.subplots(3, 2, sharex=True, sharey=True)
for j, d in enumerate([200, 0, 100, -50, 50, 'rand']):
if d != 'rand':
#if d<999:
#plt.close()
H = get_ideal_H(n_fft, fs, band, d)
b_cfir_f = cLS(F, H, 0)
X_train, Y_train = get_XY(x[:fs * 60 * 2],
band_hilbert(x[:fs * 60 * 2], fs, (8, 12)),
n_taps, d)
b_cfir_f = cLS(X_train, Y_train, 0)
if d > 0:
b_fir_band = firwin(d, band / fs * 2, pass_zero=False)
b_fir_smooth = firwin(d, 2 / fs * 2)
rec_fir = lfilter(b_fir_smooth, [1],
np.abs(lfilter(b_fir_band, [1], x)))
#plt.plot(b_cfir_f)
#plt.show()
#plt.plot(x)
rec = np.abs(lfilter(b_cfir_f, [1], x))
else:
rec = np.random.normal(size=len(x))
corrs_rls = np.zeros((len(indx) - 1, len(delays))) * np.nan
min_phase_delays = np.zeros((len(indx) - 1, len(delays))) * np.nan
corrs_min_phase = np.zeros((len(indx) - 1, len(delays))) * np.nan
d = 0
for k, kk in enumerate(indx[2:]):
print('****', k)
x_train = x[indx[k - 1]:indx[k - 1] + int(step)]
x_std = x_train.std()
x_mean = x_train.mean()
x_train = (x_train - x_mean) / x_std
y_train = band_hilbert(x_train, 500, band)
x_test = x[kk:kk + int(step)]
y_test = band_hilbert(x_test, 500, band)
X_train, Y_train = get_XY(x_train, y_train, n_taps, d)
b_cfir_t = cLS(X_train, Y_train, 0)
rec_cfir_t = np.abs(lfilter(b_cfir_t, [1], x_test))[d:]
X_train, Y_train = get_XY(x_train, y_train, n_taps, 50)
b_cfir_t = cLS(X_train, Y_train, 0)
rec_cfir_t50 = np.abs(lfilter(b_cfir_t, [1], x_test))
X_train, Y_train = get_XY(x_train, y_train, n_taps, 100)
b_cfir_t = cLS(X_train, Y_train, 0)
rec_cfir_t100 = np.abs(lfilter(b_cfir_t, [1], x_test))
tar = np.abs(y_test)[:-d if d > 0 else None]
plt.plot(np.arange(len(rec_cfir_t100)) / fs,
rec_cfir_t100 / rec_cfir_t100.std(),