lon_bin = 20 + np.floor((4./36)*lon_m)
lat_bin = 60 + np.floor((4./18)*lat_m)
R_bin = 90 + np.floor(2*(R_m - 10))
dev_id_features = device_features[dev_id_int, :]
prob_lon = device_features[dev_id_int, lon_bin]
prob_lat = device_features[dev_id_int, lat_bin]
prob_R = device_features[dev_id_int, R_bin]
#seq_id_features = seq_features[seq_id_int, :]
#prob_x = np.sum((dev_id_features[:30]/np.sqrt(np.sum(dev_id_features[:30]**2)))*(seq_id_features[:30]/np.sqrt(np.sum(seq_id_features[:30]**2))))
#prob_y = np.sum((dev_id_features[30:60]/np.sqrt(np.sum(dev_id_features[30:60]**2)))*(seq_id_features[30:60]/np.sqrt(np.sum(seq_id_features[:30]**2))))
#prob_z = np.sum((dev_id_features[60:90]/np.sqrt(np.sum(dev_id_features[60:90]**2)))*(seq_id_features[60:90]/np.sqrt(np.sum(seq_id_features[:30]**2))))
#prob_x = gaussian_probability(seq_features[seq_id_int, 0], device_features[dev_id_int, 0], device_features[dev_id_int, 3])
#prob_y = gaussian_probability(seq_features[seq_id_int, 1], device_features[dev_id_int, 1], device_features[dev_id_int, 4])
#prob_z = gaussian_probability(seq_features[seq_id_int, 2], device_features[dev_id_int, 2], device_features[dev_id_int, 5])
posterior = prob_lat*prob_lon*prob_R*train_lengths[device_ids.index(dev_id_int)]
#posterior = prob_lat*prob_lon*train_lengths[device_ids.index(dev_id_int)]
submission_file.write(question_id + "," + str(posterior)+'\n')
if int(question_id)%1000 == 0:
print question_id
z_m = np.clip(seq_features[seq_id_int, 2], -15., 14.9)
x_bin = 20 + np.floor((4./3)*x_m)
y_bin = 60 + np.floor((4./3)*y_m)
z_bin = 100 + np.floor((4./3)*z_m)
dev_id_features = device_features[dev_id_int, :]
prob_x = device_features[dev_id_int, x_bin]
prob_y = device_features[dev_id_int, y_bin]
prob_z = device_features[dev_id_int, z_bin]
#seq_id_features = seq_features[seq_id_int, :]
#prob_x = np.sum((dev_id_features[:30]/np.sqrt(np.sum(dev_id_features[:30]**2)))*(seq_id_features[:30]/np.sqrt(np.sum(seq_id_features[:30]**2))))
#prob_y = np.sum((dev_id_features[30:60]/np.sqrt(np.sum(dev_id_features[30:60]**2)))*(seq_id_features[30:60]/np.sqrt(np.sum(seq_id_features[:30]**2))))
#prob_z = np.sum((dev_id_features[60:90]/np.sqrt(np.sum(dev_id_features[60:90]**2)))*(seq_id_features[60:90]/np.sqrt(np.sum(seq_id_features[:30]**2))))
#prob_x = gaussian_probability(seq_features[seq_id_int, 0], device_features[dev_id_int, 0], device_features[dev_id_int, 3])
#prob_y = gaussian_probability(seq_features[seq_id_int, 1], device_features[dev_id_int, 1], device_features[dev_id_int, 4])
#prob_z = gaussian_probability(seq_features[seq_id_int, 2], device_features[dev_id_int, 2], device_features[dev_id_int, 5])
posterior = prob_x*prob_y*prob_z*train_lengths[device_ids.index(dev_id_int)]
submission_file.write(question_id + "," + str(posterior)+'\n')
if int(question_id)%1000 == 0:
print question_id