def get_circadian_mvmt():
X = data_utils.get_column("loc_x")
Y = data_utils.get_column("loc_y")
fft_x = np.fft.rfft(X)
fft_y = np.fft.rfft(Y)
ps_x = np.sum(np.abs(fft_x)**2)
ps_y = np.sum(np.abs(fft_y)**2)
return math.log(ps_x + ps_y) - math.log(24)
def _get_phone_use_duration():
T = data_utils.get_column("time_recorded")
P = data_utils.get_column("phone_on")
time_used = 0.0
last_time = T[0]
phone_used_last_time = P[0]
for time, p in zip(T, P):
if p == 1 and phone_used_last_time:
time_used += (time - last_time)
last_time = time
phone_used_last_time = p
return time_used / ((T[-1] - T[0]) / (60 * 60 * 24))
def _get_phone_use_freq(threshold=3):
P = data_utils.get_column("phone_on").flatten()
freq = 0
last_status = 0
for i, status in enumerate(P):
if status == 1 and all(P[max(0, i - threshold + 1):i] == 1): freq += 1
last_status = status
return freq
def compute_optimal_location_clusters(max_clusters=10, return_labels=False):
X = data_utils.get_column("loc_x, loc_y")
best_inertia = float('inf')
best_centers = []
if return_labels: best_labels = []
for i in range(1, max_clusters):
km = KMeans(n_clusters=i, random_state=42)
km.fit(X)
if km.inertia_ < best_inertia:
best_inertia = km.inertia_
best_centers = km.cluster_centers_
if return_labels: best_labels = km.labels_
if return_labels:
return X, best_labels, best_centers
else:
return best_centers
def _compute_location_variance():
x = data_utils.get_column("loc_x")
y = data_utils.get_column("loc_y")
return math.log(np.std(x)**2 + np.std(y)**2)