def ex1_2_3_mean_square_error():
'''"# Empirically, hamilton segmenter performed better for RRI
ecg_signal_resampled = resample(ecg_signal,
len(ppg_signal)) # we need to resample the eeg signal to rate of the ppg signal
assert len(ecg_signal_resampled) == len(ppg_signal)
ecg_peak_indices = ecg_processing.hamilton_segmenter(ecg_signal_resampled, ppg_sf)[
'rpeaks'] # now it has the same sf as the ppg signal
ecg_peak_times = ecg_peak_indices / ppg_sf
rri = np.diff(ecg_peak_times)
'''
# Empirically, Elgendi performed better for RRI
ecg_signal_resampled = resample(ecg_signal, len(ppg_signal)) # we need to resample the eeg signal to rate of the ppg signal
assert len(ecg_signal_resampled) == len(ppg_signal)
ecg_sf = ppg_sf
ecg_peak_times = ex1_2_4_elgendi()
rri = np.diff(ecg_peak_times)
# Empirically, find_peaks performed better for PPI
filtered_ppg_signal = ppg_processing.bvp(ppg_signal, ppg_sf, show=False)['filtered']
ppg_peak_indices, _ = find_peaks(filtered_ppg_signal, prominence=35, distance=0.5 * ppg_sf)
ppg_peak_times = ppg_peak_indices / ppg_sf
ppi = np.diff(ppg_peak_times)
# we need to trim the bigger one, if the algorithms did not find the same amount of peaks
if len(rri) > len(ppi):
rri = rri[:len(ppi)]
if len(ppi) > len(rri):
ppi = ppi[:len(rri)]
mse = np.square(np.subtract(rri, ppi)).mean()
print("MSE:", mse)
def _process_signal(self, signal, freq):
biosppy_processed = bvp(signal, freq, show=self.plot)
y_filtered = biosppy_processed['filtered']
if isinstance(y_filtered, list):
y = np.asarray(y_filtered)
else:
y = y_filtered
measures = hb.process(y, freq, calc_freq=True)
return measures
def ex1_2_2_scipy():
# Find peaks
print(ppg_sf)
filtered_ppg_signal = ppg_processing.bvp(ppg_signal, ppg_sf, show=False)['filtered']
ppg_peak_indices, _ = find_peaks(filtered_ppg_signal, prominence=35, distance=0.5 * ppg_sf)
ppg_peak_times = ppg_peak_indices / ppg_sf
plot_peaks(filtered_ppg_signal, ppg_times, ppg_peak_times, ppg_peak_indices, ppg_envelopes, ppg_label, 'findpeaks',
'results_findpeaks', (-300, 300))
plot_peak_intervals(ppg_peak_times, ppg_label, 'findpeaks', 'PPI', 'results_findpeaks')
def load_feel_data(sub_id): # TODO machine learning to maximize this shit
feel_loc = "../Data/FeelData/EEGStudy1/Feel_Data/"
feel_data = loadmat(feel_loc + "Subject_" + str(sub_id) + ".mat")
ts = feel_data['numStart'][0]
bvp_data = [i[0] for i in feel_data['denB']]
temp_data = [i[0] for i in feel_data['denT']]
gsr_data = [i[0] for i in feel_data['denG']]
secs = int(ts[5])
ms = 1000 * (int(ts[5]) - secs)
start_time = datetime.datetime(int(ts[0]), int(ts[1]), int(ts[2]),
int(ts[3]), int(ts[4]), secs, ms)
time_data = [
datetime.time(int(t_raw[0]), int(t_raw[1]), int(t_raw[2]))
for t_raw in feel_data['ACQ']
]
bvp_sig = bvp.bvp(signal=bvp_data, sampling_rate=20, show=True)
def ex1_2_2_biosspy():
# Find onsets
# It uses Zong et al. approach, which skips corrupted signal parts
result = ppg_processing.bvp(ppg_signal, ppg_sf, show=False)
filtered_ppg_signal, ppg_onset_indices, heart_rate_ts, heart_rate = result['filtered'], result['onsets'], result[
'heart_rate_ts'], result['heart_rate']
ppg_onset_times = ppg_onset_indices / ppg_sf
plot_onsets(filtered_ppg_signal, ppg_times, ppg_onset_times, ppg_envelopes, ppg_label, 'bvp',
'results_bvp', (-300, 300))
plot_peak_intervals(ppg_onset_times, ppg_label, 'bvp', 'Onset intervals', 'results_bvp')
# Plot heart rate, just for fun
fig = plt.figure(figsize=(16, 4))
plt.plot(heart_rate_ts, heart_rate)
mean = np.mean(heart_rate)
plt.plot((0, heart_rate_ts[-1]), [mean, mean], 'r-', label='Mean')
plt.xlabel('Time [s]')
plt.ylabel('Instantaneous Heart Rate [bpm]')
plt.grid()
plt.title("Heart rate derived from PPG")
fig.tight_layout()
plt.legend(loc='upper right')
fig.savefig("results_bvp/heart_rate")
def parsePhysiological(self, log, participantid):
print "Participant: ", participantid
rawSCArr = []
rawHRArr = []
hrUnfiltered = []
scUnfiltered = []
scFiltered = []
hrFiltered = []
timestamps = []
heartrates = {}
skinconductances = {}
for ts in self.timestamps:
heartrates[ts] = {
'measurements': [],
'value': 0
}
skinconductances[ts] = {
'measurements': [],
'value': 0
}
hrRange = (1000/samplingRate)*hrInterval #range for BPM calculation in Number of measurements (array positions)
for k,v in enumerate(log):
if k>2:
ts = int(float(v[0]))
timestamps.append(ts)
hrUnfiltered.append(float(v[13]))
scUnfiltered.append(float(v[8]))
numberOfMeasurements = len(hrUnfiltered)
# print numberOfMeasurements, "--", participantid
scFiltered = scUnfiltered
for k,v in enumerate(hrUnfiltered):
# print k , " - ", numberOfMeasurements, " - ", participantid
measurement = None
sample = []
if k>hrRange/2 and k<numberOfMeasurements-hrRange/2:
sample = hrUnfiltered[k-hrRange/2:k+hrRange/2]
elif k<hrRange/2:
sample = hrUnfiltered[0:hrRange]
else:
sample = hrUnfiltered[numberOfMeasurements-hrRange:numberOfMeasurements-1]
try:
bvpArr = bvp.bvp(sample,20,show=False)
measurement = np.nanmean(bvpArr["heart_rate"])
hrFiltered.append(measurement)
except:
# print("could not compute bpm")
measurement = hrFiltered[k-1]
hrFiltered.append(measurement)
# print measurement
for k,v in enumerate(hrFiltered):
modulo = timestamps[k]%500
tsRounded = timestamps[k]-modulo
if tsRounded in self.timestamps:
heartrates[tsRounded]['measurements'].append(v)
skinconductances[tsRounded]['measurements'].append(scFiltered[k])
for m in heartrates:
heartrates[m]['value'] = np.nanmean(heartrates[m]['measurements'])
skinconductances[m]['value'] = np.nanmean(skinconductances[m]['measurements'])
# print heartrates
# self.rawSkinconductances[participantid] = rawSCArr
# self.rawHeartrates[participantid] = rawHRArr
self.heartrates[participantid] = heartrates
self.skinconductances[participantid] = skinconductances
print "YOLO"
print len(self.heartrates[participantid])
print len(self.timestamps)
# timestamp = timestamp0 + timestamp1*256 + timestamp2*65536
timestamp = int(round(time.time() * 1000))
# print timestamp
# allHR.append(PPG_raw)
sc = {"data": [{"timestamp": timestamp, "value": GSR_ms}]}
#calculate bpm for last X seconds: at 20HZ, that's every X*20 measurements.
numMeasurements = 1200 #60 sec -> each timestamp will calculate for last 60sec
allHR.append(PPG_raw)
if len(allHR) >= numMeasurements:
# print allHR[len(allHR)-200:len(allHR)-1]
x = bvp.bvp(allHR[len(allHR) - numMeasurements:len(allHR) - 1],
20,
show=False)
# print "heart rate:"
# print x['heart_rate']
# allHR = []
bpm = np.mean(x['heart_rate'])
# print bpm
hr = {"data": [{"timestamp": timestamp, "value": bpm}]}
hrReq = requests.post('http://localhost:3000/postHR', json=hr)
# print json.dumps(obj)
# print hr
scReq = requests.post('http://localhost:3000/postSC', json=sc)
# print hrReq.json()
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("csv_file", type=str, help="File constaining the raw data")
parser.add_argument('-s',
'--sampling-rate',
type=int,
default=1000,
help="The sample rate")
parser.add_argument('-m',
'--min-amplitude',
type=float,
default=0.1,
help="Min amplitude (for EDA)")
args = parser.parse_args()
signals = np.loadtxt(args.csv_file, delimiter=',')
t = signals[:, 0]
bvp_raw = signals[:, 1]
eda_raw = signals[:, 2]
bvp_signal = bvp.bvp(bvp_raw, sampling_rate=args.sampling_rate)
eda_signal = eda.eda(eda_raw,
sampling_rate=args.sampling_rate,
min_amplitude=args.min_amplitude)
print(bvp_signal)
print(eda_signal)
#
# with open(csv_file_name, 'w') as csv_file:
# csv_reader = csv.reader(csv_file, delimiter=',')
# for row in csv_reader:
