m_log_weights_np = D["log_weights"]
v_A = np.zeros(len(tmp))
for d_timestep in range(2*6,len(tmp)):
d_row = int(tmp[d_timestep,0]*0.01/LAT_RESO)
d_col = int((tmp[d_timestep,1]-300)*0.01/LON_RESO)
d_ll_t = np.mean(m_log_weights_np[d_timestep,:,:])
d_qualtile = (d_ll_t - m_map_ll_mean[d_row,d_col])/m_map_ll_std[d_row,d_col]
if (not np.isfinite(d_qualtile)):
v_A[d_timestep] = 2
elif (d_qualtile <= -1.282):
v_A[d_timestep] = 1
v_A = v_A[12:]
v_anomalies = np.zeros(len(v_A))
for d_i_4h in range(0,len(v_A)+1-24):
v_A_4h = v_A[d_i_4h:d_i_4h+24]
v_anomalies_i = contrario_utils.contrario_detection(v_A_4h,CONTRARIO_EPS)
v_anomalies[d_i_4h:d_i_4h+24][v_anomalies_i==1] = 1
## Plotting
tmp = tmp[12:]
v_lat = (tmp[:,0]/300.0)*LAT_RANGE + LAT_MIN
v_lon = (tmp[:,1]-300.0)/300.0*LON_RANGE + LON_MIN
plt.plot(v_lon,v_lat,color='g',linewidth=0.8)
if len(contrario_utils.nonzero_segments(v_anomalies)) > 0:
D["anomaly_idx"] = v_anomalies
l_dict_anomaly.append(D)
# for l_segment in nonzero_segments(v_anomalies):
# plt.plot(v_lon[l_segment],v_lat[l_segment],color='r',linewidth=0.8)
##
#plt.close()
d_logprob_t = np.mean(m_log_weights_np[d_timestep, :])
# KDE
l_local_log_prod = Map_logprob[str(d_row) + "," + str(d_col)]
if len(l_local_log_prod) < 2:
v_A[d_timestep] = 2
else:
kernel = stats.gaussian_kde(l_local_log_prod)
cdf = kernel.integrate_box_1d(-np.inf, d_logprob_t)
if cdf < 0.1:
v_A[d_timestep] = 1
v_A = v_A[12:]
v_anomalies = np.zeros(len(v_A))
for d_i_4h in range(0, len(v_A) + 1 - 24):
v_A_4h = v_A[d_i_4h:d_i_4h + 24]
v_anomalies_i = contrario_utils.contrario_detection(
v_A_4h, config.contrario_eps)
v_anomalies[d_i_4h:d_i_4h + 24][v_anomalies_i == 1] = 1
if len(contrario_utils.nonzero_segments(v_anomalies)) > 0:
D["anomaly_idx"] = v_anomalies
l_dict_anomaly.append(D)
except:
n_error += 1
print("Number of processed tracks: ", len(l_dict))
print("Number of abnormal tracks: ", len(l_dict_anomaly))
print("Number of errors: ", n_error)
# Save to disk
n_anomalies = len(l_dict_anomaly)
save_filename = os.path.basename(config.trainingset_name)\
+"-" + os.path.basename(config.trainingset_name)\