class HRMonitor:
"""
Encapsulated class attributes (with default values)
"""
__hr = 0 # the current heart rate
__time = None # CircularList containing the time vector
__ppg = None # CircularList containing the raw signal
__filtered = None # CircularList containing filtered signal
__num_samples = 0 # The length of data maintained
__new_samples = 0 # How many new samples exist to process
__fs = 0 # Sampling rate in Hz
__thresh = 0.6 # Threshold from Tutorial 2
"""
Initialize the class instance
"""
def __init__(self, num_samples, fs, times=[], data=[]):
self.__hr = 0
self.__num_samples = num_samples
self.__fs = fs
self.__time = CircularList(times, num_samples)
self.__ppg = CircularList(data, num_samples)
self.__filtered = CircularList([], num_samples)
"""
Add new samples to the data buffer
Handles both integers and vectors!
"""
def add(self, t, x):
if isinstance(t, np.ndarray):
t = t.tolist()
if isinstance(x, np.ndarray):
x = x.tolist()
self.__time.add(t)
self.__ppg.add(x)
self.__new_samples += len(x)
"""
Compute the average heart rate over the peaks
"""
def compute_heart_rate(self, peaks):
t = np.array(self.__time)
return 60 / np.mean(np.diff(t[peaks]))
"""
Process the new data to update step count
"""
def process(self):
# Grab only the new samples into a NumPy array
x = np.array(self.__ppg[-self.__new_samples:])
# Filter the signal (feel free to customize!)
x = filt.detrend(x, 25)
x = filt.moving_average(x, 5)
x = filt.gradient(x)
x = filt.normalize(x)
# Store the filtered data
self.__filtered.add(x.tolist())
# Find the peaks in the filtered data
_, peaks = filt.count_peaks(x, self.__thresh, 1)
# Update the step count and reset the new sample count
self.__hr = self.compute_heart_rate(peaks)
self.__new_samples = 0
# Return the heart rate, peak locations, and filtered data
return self.__hr, peaks, np.array(self.__filtered)
"""
Clear the data buffers and step count
"""
def reset(self):
self.__steps = 0
self.__time.clear()
self.__ppg.clear()
self.__filtered = np.zeros(self.__num_samples)
class Pedometer:
"""
Encapsulated class attributes (with default values)
"""
__steps = 0 # the current step count
__l1 = None # CircularList containing L1-norm
__filtered = None # CircularList containing filtered signal
__num_samples = 0 # The length of data maintained
__new_samples = 0 # How many new samples exist to process
__fs = 0 # Sampling rate in Hz
__b = None # Low-pass coefficients
__a = None # Low-pass coefficients
__thresh_low = 3 # Threshold from Tutorial 2
__thresh_high = 15 # Threshold from Tutorial 2
__xi = None # Initial conditions for filter
__yi = None # Initial conditions for filter
"""
Initialize the class instance
"""
def __init__(self, num_samples, fs, data=None):
self.__steps = 0
self.__num_samples = num_samples
self.__fs = fs
self.__l1 = CircularList(data, num_samples)
self.__filtered = CircularList([], num_samples)
self.__b, self.__a = filt.create_filter(3, 1.2, "lowpass", fs)
self.__xi = np.zeros(4)
self.__yi = np.zeros(4)
"""
Add new samples to the data buffer
Handles both integers and vectors!
"""
def add(self, ax, ay, az):
l1 = filt.l1_norm(ax, ay, az)
if isinstance(ax, int):
num_add = 1
else:
num_add = len(ax)
l1 = l1.tolist()
self.__l1.add(l1)
self.__new_samples += num_add
"""
Process the new data to update step count
"""
def process(self):
# Grab only the new samples into a NumPy array
x = np.array(self.__l1[-self.__new_samples:])
# Filter the signal (detrend, LP, MA, etc…)
x = filt.detrend(x)
# x = filt.filter(self.__b, self.__a, x)
x, self.__xi, self.__yi = filt.filter_ic(self.__b, self.__a, x,
self.__xi, self.__yi)
x = filt.gradient(x)
x = filt.moving_average(x, 25)
# Store the filtered data
self.__filtered.add(x.tolist())
# Count the number of peaks in the filtered data
count, peaks = filt.count_peaks(x, self.__thresh_low,
self.__thresh_high)
# Update the step count and reset the new sample count
self.__steps += count
self.__new_samples = 0
# Return the step count, peak locations, and filtered data
return self.__steps, peaks, np.array(self.__filtered)
"""
Clear the data buffers and step count
"""
def reset(self):
self.__steps = 0
self.__l1.clear()
self.__filtered = np.zeros(self.__num_samples)
class HRMonitor:
"""
Encapsulated class attributes (with default values)
"""
__hr = 0 # the current heart rate
__time = None # CircularList containing the time vector
__ppg = None # CircularList containing the raw signal
__filtered = None # CircularList containing filtered signal
__num_samples = 0 # The length of data maintained
__new_samples = 0 # How many new samples exist to process
__fs = 0 # Sampling rate in Hz
__thresh = 0.6 # Threshold from Tutorial 2
__directory = "/Users/akshaygopalkrishnan/Desktop/ECE 16/Python/Lab 7/data/data"
"""
Initialize the class instance
"""
def __init__(self, num_samples, fs, times=[], data=[]):
self.__hr = 0
self.__num_samples = num_samples
self.__fs = fs
self.__time = CircularList(data, num_samples)
self.__ppg = CircularList(data, num_samples)
self.__filtered = CircularList([], num_samples)
self.__gmm = GMM(n_components=2)
"""
Add new samples to the data buffer
Handles both integers and vectors!
"""
def add(self, t, x):
if isinstance(t, np.ndarray):
t = t.tolist()
if isinstance(x, np.ndarray):
x = x.tolist()
if isinstance(x, int):
self.__new_samples += 1
else:
self.__new_samples += len(x)
self.__time.add(t)
self.__ppg.add(x)
"""
Compute the average heart rate over the peaks
"""
def compute_heart_rate(self, peaks):
t = np.array(self.__time)
if len(np.diff(t[peaks])) > 0:
return 60 / np.mean(np.diff(t[peaks]))
else:
return 0
"""
Removes outlier peaks from the filtered data
@:param peaks: The location of each peak
@:return the peaks with outliers removed
"""
def remove_outliers(self, peaks):
if len(peaks) > 1:
t = np.array(self.__time)
# Calculate peak difference and average/standard deviation
peak_diff = np.diff(t[peaks])
avg_peak_diff = np.mean(peak_diff)
std_peak_diff = np.std(peak_diff)
for i in range(len(peak_diff)):
# If the peak difference is less than 2 deviations from the average, remove the peak (outlier)
if peak_diff[i] < (avg_peak_diff - (1 * std_peak_diff)):
peaks.pop(i + 1)
return peaks
# Filter the signal (as in the prior lab)
def train_process(self, x):
x = filt.detrend(x, 25)
x = filt.moving_average(x, 5)
x = filt.gradient(x)
return filt.normalize(x)
# Retrieve a list of the names of the subjects
def get_subjects(self, directory):
filepaths = glob.glob(directory + "/*")
return [filepath.split("/")[-1] for filepath in filepaths]
# Estimate the heart rate from the user-reported peak count
def get_hr(self, filepath, num_samples, fs):
count = int(filepath.split("_")[-1].split(".")[0])
seconds = num_samples / fs
return count / seconds * 60 # 60s in a minute
# Estimate the sampling rate from the time vector
def estimate_fs(self, times):
return 1 / np.mean(np.diff(times))
# Retrieve a data file, verifying its FS is reasonable
def get_data(self, directory, subject, trial, fs):
search_key = "%s/%s/%s_%02d_*.csv" % (directory, subject, subject,
trial)
filepath = glob.glob(search_key)[0]
t, ppg = np.loadtxt(filepath, delimiter=',', unpack=True)
t = (t - t[0]) / 1e3
hr = self.get_hr(filepath, len(ppg), fs)
fs_est = self.estimate_fs(t)
if (fs_est < fs - 1 or fs_est > fs):
print("Bad data! FS=%.2f. Consider discarding: %s" %
(fs_est, filepath))
return t, ppg, hr, fs_est
"""
Trains the GMM model on offline data
@:return: the trained GMM model
"""
def train(self):
print("Training GMM model... ")
subjects = self.get_subjects(self.__directory)
train_data = np.array([])
for subject in subjects:
for trial in range(1, 11):
t, ppg, hr, fs_est = self.get_data(self.__directory, subject,
trial, self.__fs)
train_data = np.append(train_data, self.train_process(ppg))
# Train the GMM
train_data = train_data.reshape(-1,
1) # convert from (N,1) to (N,) vector
self.__gmm = GMM(n_components=2).fit(train_data)
"""
Estimate the heart rate given GMM output labels
"""
def estimate_hr(self, labels, num_samples, fs):
peaks = np.diff(labels, prepend=0) == 1
count = sum(peaks)
seconds = num_samples / fs
hr = count / seconds * 60 # 60s in a minute
return hr, peaks
"""
Uses the GMM model to estimate the heart rate
@:param filtered: the filtered data
@:param fs: the sampling frequency
@:return: the estimated heart rate and estimated time of each peak
"""
def predict(self):
# Grab only the new samples into a NumPy array
x = np.array(self.__ppg[-self.__new_samples:])
filtered_arr = self.train_process(x)
self.__filtered.add(filtered_arr.tolist())
labels = self.__gmm.predict(np.array(self.__filtered).reshape(-1, 1))
self.__new_samples = 0
hr_est, est_peaks = self.estimate_hr(labels, len(self.__filtered),
self.__fs)
return hr_est, est_peaks, np.array(self.__filtered)
"""
Process the new data to update step count
"""
def process(self):
# Grab only the new samples into a NumPy array
x = np.array(self.__ppg[-self.__new_samples:])
# Filter the signal (feel free to customize!)
x = filt.detrend(x, 25)
x = filt.moving_average(x, 5)
x = filt.gradient(x)
x = filt.normalize(x)
# Store the filtered data
self.__filtered.add(x.tolist())
# Find the peaks in the filtered data
_, peaks = filt.count_peaks(self.__filtered, self.__thresh, 1)
peaks = self.remove_outliers(peaks)
# Update the step count and reset the new sample count
self.__hr = self.compute_heart_rate(peaks)
self.__new_samples = 0
# Return the heart rate, peak locations, and filtered data
return self.__hr, peaks, np.array(self.__filtered)
"""
Clear the data buffers and step count
"""
def reset(self):
self.__steps = 0
self.__time.clear()
self.__ppg.clear()
self.__filtered = np.zeros(self.__num_samples)
