def extractThetaAlphaBeta(self, x):
'''
:param x: eeg signal
:return: theta, alpha, and beta of the signal
'''
filtered = self.preProcessing(x)
theta = butterBandpassFilter(filtered, lowcut=4, highcut=7, fs=self.fs, order=2)
alpha = butterBandpassFilter(filtered, lowcut=8, highcut=15, fs=self.fs, order=2)
beta = butterBandpassFilter(filtered, lowcut=16, highcut=31, fs=self.fs, order=2)
return theta, alpha, beta
def extractMFCCFeatures(self, x, winlen=2.0, stride=0.5, min_len=465):
'''
Compute melch frequency spectrum of EDA
:param x: input signal
:param winlen: windows length for framming. The default is 2.0 sec since the gradual changes in principle EDA towards stimulus is between 1.0 and 3.0 secs
:param stride: string length for framming. The default is 0.5 sec
:return: normalized melc coefficient
'''
x_len = len(x)
len_diff = min_len - x_len
if len_diff > 0:
x = np.append(x, x[x_len - len_diff:])
# print(x.shape)
filtered = butterBandpassFilter(x,
lowcut=0.03,
highcut=5.,
order=4,
fs=self.fs)
melcfet = mfcc(filtered,
samplerate=self.fs,
winlen=winlen,
winstep=stride)
melcfet -= (np.mean(melcfet, axis=0) + 1e-18)
# print(np.squeeze(np.squeeze(melcfet)).flatten().shape)
return np.squeeze(np.squeeze(melcfet)).flatten()
def preProcessing(self, x, n=50):
lc = 4
hc = 55
filtered = butterBandpassFilter(x, lowcut=lc, highcut=hc, fs=self.fs)
smoothed = avgSlidingWindow(filtered, n=n)
return smoothed
def extractCVXEDA(self, x):
if np.std(x) < 1e-5:
return np.array([])
try:
filtered = butterBandpassFilter(x,
lowcut=0.03,
highcut=5.,
order=4,
fs=self.fs)
yn = (filtered - filtered.mean()) / filtered.std()
[r, p, t, l, d, _, _] = cvxEDA(yn, 1. / self.fs)
# r features
r_mean = np.mean(r)
r_std = np.std(r)
r_max = np.max(r)
r_min = np.min(r)
# p features
p_mean = np.mean(p)
p_std = np.std(p)
p_max = np.max(p)
p_min = np.min(p)
# t features
t_mean = np.mean(t)
t_std = np.std(t)
t_max = np.max(t)
t_min = np.min(t)
# l features
l_mean = np.mean(l)
l_std = np.std(l)
l_max = np.max(l)
l_min = np.min(l)
# l features
d_mean = np.mean(d)
d_std = np.std(d)
d_max = np.max(d)
d_min = np.min(d)
return np.array([
r_mean, r_std, r_max, r_min, p_mean, p_std, p_max, p_min,
t_mean, t_std, t_max, t_min, l_mean, l_std, l_max, l_min,
d_mean, d_std, d_max, d_min
])
except:
return np.array([])
def extractThetaAlphaBeta(self, x):
'''
ref:https://www.journals.elsevier.com/clinical-neurophysiology/view-for-free/guidelines-of-the-ifcn-2nd-ed-published-1999
:param x: eeg signal
:return: theta, alpha, and beta of the signal
'''
filtered = self.preProcessing(x)
theta = butterBandpassFilter(filtered,
lowcut=4,
highcut=8,
fs=self.fs,
order=2)
alpha_low = butterBandpassFilter(filtered,
lowcut=8,
highcut=10,
fs=self.fs,
order=2)
alpha_high = butterBandpassFilter(filtered,
lowcut=10,
highcut=14,
fs=self.fs,
order=2)
beta = butterBandpassFilter(filtered,
lowcut=14,
highcut=25,
fs=self.fs,
order=2)
gamma_low = butterBandpassFilter(filtered,
lowcut=25,
highcut=40,
fs=self.fs,
order=2)
gamma_high = butterBandpassFilter(filtered,
lowcut=40,
highcut=100,
fs=self.fs,
order=2)
return theta, alpha_low, alpha_high, beta, gamma_low, gamma_high
featuresExct = GSRFeatures()
for i in (0, 1, 7, 9):
time_start = timeToInt(game_results.iloc[i]["Time_Start"])
time_end = timeToInt(game_results.iloc[i]["Time_End"])
skin_conductance = gsr_data[
(gsr_data["GSR_Timestamp_Shimmer_CAL"].values >= time_start)
& (gsr_data["GSR_Timestamp_Shimmer_CAL"].values <= time_end
)]["GSR_GSR_Skin_Conductance_CAL"].values
ppg = gsr_data[(gsr_data["GSR_Timestamp_Shimmer_CAL"].values >= time_start)
& (gsr_data["GSR_Timestamp_Shimmer_CAL"].values <= time_end
)]["GSR_PPG_A13_CAL"].values
skin_conductance_f = butterBandpassFilter(skin_conductance,
0.03,
5.0,
15.0,
order=5)
ts, ppg_f = featuresExct.computeHeartBeat(ppg.flatten(), 15.0)
plt.plot(skin_conductance_f)
plt.ylim([0, 1])
plt.ylabel("Skin conductance (µS)")
plt.savefig(save_path + str(i) + "_GSR.png")
plt.show()
plt.plot(ts, ppg_f)
plt.ylim([40, 100])
plt.ylabel("Heart rate (bpm)")
plt.savefig(save_path + str(i) + "_PPG.png")
plt.show()