def detection_with_bothwindows(eigen_file="UCI_eigs_slices.pkl",
timestamps=195,
percent_ranked=0.05,
window1=10,
window2=30,
initial_window=30,
difference=True):
principal = True
spectrums = normal_util.load_object(eigen_file)
spectrums = np.asarray(spectrums)
spectrums = spectrums.real
spectrums = spectrums.reshape((timestamps, -1))
spectrums = normalize(spectrums, norm='l2')
print("short window is " + str(window1))
print("long window is " + str(window2))
print("initial window is " + str(initial_window))
print(spectrums.shape)
(z_shorts, z_longs, z_scores,
anomalies) = change_detection_two_windows(spectrums,
principal=principal,
percent_ranked=percent_ranked,
window1=window1,
window2=window2,
initial_window=initial_window,
difference=difference)
print("found anomalous time stamps are")
print(anomalies)
events = anomalies
return (z_shorts, z_longs, z_scores, events)
def detection_with_shortwindow(eigen_file="UCI_eigs_slices.pkl",
timestamps=195,
percent_ranked=0.05,
window=10,
initial_window=15,
difference=True):
principal = True
spectrums = normal_util.load_object(eigen_file)
spectrums = np.asarray(spectrums).real
spectrums = spectrums.reshape((timestamps, -1))
spectrums = normalize(spectrums, norm='l2')
print("window is " + str(window))
print("initial window is " + str(initial_window))
print(spectrums.shape)
(z_scores, anomalies) = change_detection(spectrums,
principal=principal,
percent_ranked=percent_ranked,
window=window,
initial_window=initial_window,
difference=difference)
print("found anomalous time stamps are")
print(anomalies)
return z_scores
def visiualize_vecs_UCI(eigen_file, vec_file, eigen_name, vec_name):
Temporal_eigenvalues = normal_util.load_object(eigen_file)
activity_vecs = normal_util.load_object(vec_file)
limit = 5
for i in range(0, len(Temporal_eigenvalues)):
Temporal_eigenvalues[i] = Temporal_eigenvalues[i][0:limit]
for i in range(0, len(activity_vecs)):
activity_vecs[i] = activity_vecs[i].flatten()[0:limit]
graph_name = "UCI"
# normal_util.plot_laplacian_spectrum(diag_vecs, graph_name)
normal_util.plot_activity_intensity(
np.asarray(Temporal_eigenvalues).real, eigen_name)
normal_util.plot_activity_intensity(
np.asarray(activity_vecs).real, vec_name)
def plot_anomaly_and_spectro(fname,
scores,
score_labels,
events,
eigen_file="USLegis_L_singular.pkl"):
labels = list(range(97, 109, 1))
scores[0] = set_non_negative(scores[0])
fig, axs = plt.subplots(2)
plt.rcParams.update({'figure.autolayout': True})
diag_vecs = normal_util.load_object(eigen_file)
diag_vecs = np.transpose(np.asarray(diag_vecs)) #let time be x-axis
diag_vecs = np.flip(diag_vecs, 0)
max_time = len(scores[0])
t = list(range(0, max_time))
colors = ['#fdbb84', '#43a2ca', '#bc5090', '#e5f5e0', '#fa9fb5', '#c51b8a']
for i in range(len(scores)):
axs[0].plot(t,
scores[0],
color=colors[i],
ls='solid',
label=score_labels[i])
for event in events:
max_score = 0
for i in range(len(scores)):
if scores[i][event] > max_score:
max_score = scores[i][event]
axs[0].annotate(
str(labels[event]), # this is the text
(event, max_score), # this is the point to label
textcoords="offset points", # how to position the text
xytext=(0, -12), # distance from text to points (x,y)
ha='center') # horizontal alignment can be left, right or center
addLegend = True
for event in events:
#plt.axvline(x=event,color='k', linestyle="--", linewidth=0.5)
max_score = 0
for i in range(len(scores)):
if scores[i][event] > max_score:
max_score = scores[i][event]
if (addLegend):
axs[0].plot(event,
max_score,
marker="*",
color='#de2d26',
ls='solid',
label="detected anomalies")
addLegend = False
else:
axs[0].plot(event,
max_score,
marker="*",
color='#de2d26',
ls='solid')
#US senate
for i in range(len(labels)):
labels[i] = str(labels[i])
axs[0].set_xlabel('Congress')
axs[0].set_ylabel('anomaly score')
plt.tight_layout()
axs[0].legend()
axs[0].set_xticks(t)
axs[0].set_xticklabels(labels)
axs[1].set_xlabel('Congress')
axs[1].set_ylabel('rank')
axs[1].set_xticks(t)
axs[1].set_xticklabels(labels)
axs[1].imshow(diag_vecs, aspect='auto')
plt.savefig(fname + 'anomalySpectro.pdf')
def load_mp():
fname = "datasets/canVote_processed/mp_dict.pkl"
MP_dict = normal_util.load_object(fname)
return MP_dict
def plot_spectrum(pkl_name, graph_name):
eigen_slices = normal_util.load_object(pkl_name)
normal_util.plot_activity_intensity(eigen_slices, graph_name)
