idx = 0
for i in range(len(data_EmotionTest)):
tdelta = data_EmotionTest.iloc[i]["Time_End"] - data_EmotionTest.iloc[i]["Time_Start"]
time_end = data_EmotionTest.iloc[i]["Time_End"]
valence = data_EmotionTest.iloc[i]["Valence"]
arousal = data_EmotionTest.iloc[i]["Arousal"]
emotion = data_EmotionTest.iloc[i]["Emotion"]
for j in np.arange(0, (tdelta // split_time), 0.4):
end = time_end - (j * split_time)
start = time_end - ((j + 1) * split_time)
ecg = data[(data["timestamp"].values >= start) & (data["timestamp"].values <= end)]
status = 0
# extract ecg features
time_domain = featuresExct.extractTimeDomain(ecg['ecg'].values)
freq_domain = featuresExct.extractFrequencyDomain(ecg['ecg'].values)
nonlinear_domain = featuresExct.extractNonLinearDomain(ecg['ecg'].values)
if time_domain.shape[0] != 0 and freq_domain.shape[0] != 0 and nonlinear_domain.shape[0] != 0:
featuresEachMin = np.concatenate([time_domain, freq_domain, nonlinear_domain])
if np.sum(np.isinf(featuresEachMin)) == 0 | np.sum(np.isnan(featuresEachMin)) == 0:
np.save(path_results + "ecg_" + str(idx) + ".npy", featuresEachMin)
status = 1
else:
status = 0
# add object to dataframes
emotionTestResult = emotionTestResult.append(
{"Idx": str(idx), "Start": str(start), "End": str(end), "Valence": valence, "Arousal": arousal, "Emotion": emotion, "Status": status},
ignore_index=True)
idx += 1
#features extractor
featuresExct = ECGFeatures(200)
#road test features
road_test_features = []
for j in np.arange(0, (tdelta // SPLIT_TIME), 0.2):
end = time_end - ((j) * SPLIT_TIME)
start = time_end - ((j + 1) * SPLIT_TIME) - EXTENTION_TIME
ecg = ecg_road_data[(ecg_road_data["timestamp"].values >= start)
& (ecg_road_data["timestamp"].values <= end)]
ecg_values = featuresExct.filterECG(ecg['ecg'].values)
ecg_values = signal.resample(ecg_values,
int((len(ecg_values) / FS_ECG) * 200))
time_domain = featuresExct.extractTimeDomain(ecg_values)
freq_domain = featuresExct.extractFrequencyDomain(ecg_values)
nonlinear_domain = featuresExct.extractNonLinearDomain(ecg_values)
if time_domain.shape[0] != 0 and freq_domain.shape[
0] != 0 and nonlinear_domain.shape[0] != 0:
concatenate_features = np.concatenate(
[time_domain, freq_domain, nonlinear_domain])
road_test_features.append(concatenate_features)
road_test_features = np.concatenate([road_test_features])
# concatenate features and normalize them
X = np.concatenate([features, road_test_features])
scaler = StandardScaler()
scaler.fit(X)
X_norm = scaler.transform(X)
])
# extract resp features
resp_time_features = resp_features_exct.extractTimeDomain(
resp)
resp_freq_features = resp_features_exct.extractFrequencyDomain(
resp)
resp_nonlinear_features = resp_features_exct.extractNonLinear(
resp)
resp_features = np.concatenate([
resp_time_features,
resp_features_exct.extractFrequencyDomain(resp),
resp_features_exct.extractNonLinear(resp)
])
# extract ecg features
ecg_time_domain = ecg_features_exct.extractTimeDomain(
ecg_values)
ecg_freq_domain = ecg_features_exct.extractFrequencyDomain(
ecg_values)
ecg_nonlinear_domain = ecg_features_exct.extractNonLinearDomain(
ecg_values)
ecg_features = np.concatenate([
ecg_time_domain, ecg_freq_domain,
ecg_nonlinear_domain
])
features = np.concatenate([
eda_features, ppg_features, resp_features,
ecg_resp_features, ecg_features, eeg_features
])
np.save(
# take 2.5 sec after end
# end = time_end - ((j - 1) * SPLIT_TIME) + EXTENTION_TIME
# start = time_end - (j * SPLIT_TIME)
end = time_end - ((j) * SPLIT_TIME)
start = time_end - ((j + 1) * SPLIT_TIME) - EXTENTION_TIME
resp = resp_data[(resp_data.iloc[:, 0].values >= start) & (
resp_data.iloc[:, 0].values <= end)]["resp"].values
ecg_resp = ecg_resp_data[
(ecg_resp_data.iloc[:, 0].values >= start) &
(ecg_resp_data.iloc[:, 0].values <= end)]["ecg"].values
status = 0
# extract ecg features
time_domain = ecg_features_exct.extractTimeDomain(ecg_resp)
freq_domain = ecg_features_exct.extractFrequencyDomain(
ecg_resp)
nonlinear_domain = ecg_features_exct.extractNonLinearDomain(
ecg_resp)
# extract resp features
resp_time_features = resp_features_exct.extractTimeDomain(
resp)
resp_freq_features = resp_features_exct.extractFrequencyDomain(
resp)
resp_nonlinear_features = resp_features_exct.extractNonLinear(
resp)
if resp_time_features.shape[
0] != 0 and resp_freq_features.shape[
0] != 0 and resp_nonlinear_features.shape[
0] != 0 and time_domain.shape[
class DataFetchRoad:
def __init__(self, gps_file, ecg_file, mask_file, ecg_n=45, split_time = 45, stride=0.2):
self.gps_file = gps_file
self.ecg_file = ecg_file
self.mask_file = mask_file
self.ecg_n = ecg_n
self.stride = stride
self.featuresExct = ECGFeatures(FS_ECG)
# normalization ecg features
self.ecg_mean = np.array([8.69006926e+02, 5.36894839e+01, 5.13280806e+01, 3.60941889e+01,
5.12644394e+01, 3.50184864e+01, 7.05304536e+01, 4.00354951e+00,
1.93190808e+01, 2.37990616e+01, 4.74128624e+01, 1.79100193e+01,
3.61127269e+01, 8.89897875e+00, 1.83440847e+01, 9.47694202e-03,
2.99294977e-02, 8.32468992e-02, 2.66982160e-01, 1.05872866e+02,
8.24120855e+02, 1.61071678e+03, 2.27217251e+03, 3.79534713e+00,
2.62838417e+01, 3.42443566e+01, 3.56764545e+01, 3.36520073e+00,
5.63147720e+00, 6.03246072e+00, 6.01429266e+00, 5.03771520e+01,
4.96228480e+01, 1.88115762e+00, 1.67060736e+00, 1.47731287e-01,
3.62288239e+01, 6.01129089e+01, 2.21647750e+00, 1.17607048e+04,
4.32725792e-01])
self.ecg_std = np.array([1.12849622e+02, 7.22749812e+00, 4.13239686e+01, 3.76508218e+01,
6.06763869e+01, 4.94652050e+01, 9.08751400e+00, 2.91083561e+00,
1.74801557e+01, 1.09281612e+01, 2.35429989e+01, 1.10034329e+01,
2.36269451e+01, 9.38428850e+00, 2.01946583e+01, 1.47242980e-03,
1.22646518e-02, 2.66829649e-02, 7.71102801e-02, 8.09024660e+02,
3.84167641e+03, 1.11956783e+04, 1.61020832e+04, 3.85171123e+00,
1.77196093e+01, 1.81367242e+01, 2.12639285e+01, 1.62165833e+00,
1.37502961e+00, 1.34862368e+00, 1.51030700e+00, 2.27359887e+01,
2.27359887e+01, 2.64173948e+00, 5.57174922e-01, 8.24678134e-02,
4.28983320e+01, 4.14642962e+01, 9.86755656e-01, 4.09682862e+04,
3.83966021e-01])
self.data_set = self.readData()
self.test_n = len(self.data_set)
def fetch(self):
i = 0
while i < len(self.data_set):
data_i = self.data_set[i]
yield data_i
i+=1
def readData(self):
data_set = []
gps_data = pd.read_csv(self.gps_file)
ecg_data = pd.read_csv(self.ecg_file)
ecg_data.loc[:, 'timestamp'] = ecg_data.loc[:, 'timestamp'].apply(timeToInt)
gps_data.loc[:, 'timestamp'] = gps_data.loc[:, 'timestamp'].apply(timeToInt)
for j in range(1, len(gps_data)):
start = gps_data.loc[j]["timestamp"]
end = start + (SPLIT_TIME+4)
ecg = ecg_data[(ecg_data["timestamp"].values >= start) & (ecg_data["timestamp"].values <= end)]["ecg"].values
# print(len(ecg))
if len(ecg) >= self.ecg_n:
ecg = ecg[:self.ecg_n]
#extract ECG features
time_domain = self.featuresExct.extractTimeDomain(ecg)
freq_domain = self.featuresExct.extractFrequencyDomain(ecg)
nonlinear_domain = self.featuresExct.extractNonLinearDomain(ecg)
concatenate_features = (np.concatenate([time_domain, freq_domain, nonlinear_domain]) - self.ecg_mean) / self.ecg_std
data_set.append(concatenate_features)
#raw ecg
# ecg = (ecg - 2140.397356669409) / 370.95493558685325
# ecg = (ecg - 2048.485947046843) / 156.5629266658633
# ecg = ecg / (4095 - 0)
# ecg = signal.resample(ecg, 200 * SPLIT_TIME)
# data_set.append(ecg)
# print(ecg)
return data_set