# plt.plot(error[i, :], label='exp '+str(i))
plt.title('Accuracy of the Ensemble')
plt.xlabel('Number of meters per model ($m^d$)')
plt.ylabel('Prediction error MAE($y, \hat y$)')
plt.legend()
plt.show()
plt.savefig('accuracy_ensemble.pgf', bbox_inches='tight')
#for i in range(models_len):
# print('MAE exp ' + str(i) + '= ' + str(np.mean(error[i, :])))
k = 9
plt.figure(5)
plt.clf()
for i in range(models_len):
'''
if i in [0, 1]:
c = n_vec[i] - 1
else:
c=1
'''
c = 1
pred_i = predictions[i][k].reshape(-1) * c
mae_i = fl.MAE(fl.y_test, pred_i)
plt.plot(pred_i, label='exp ' + str(i) + ', MAE=' + str(mae_i))
plt.plot(fl.y_test, '--', label='Test')
plt.legend()
plt.show()
t0 = time.perf_counter()
for k in range(reps):
print('\t\tk=' + str(k))
if strategic_attack:
meters_model = np.load(dir_models + 'meters_' + str(k) +
'.npy',
allow_pickle=True)
meters_a = random.sample(set(meters_model[0]), m_a)
else:
meters_a = random.sample(set(range(m)), m_a)
fa.get_prediction = get_prediction_a
fa.get_grad = get_grad_a
y_test, hat_y, hat_y_a, bias_opt = fl.find_attack(
dir_models, max_num_models, 1, meters_a, unique_bias)
impact[k, i, j] = fl.MAE(hat_y, hat_y_a)
pred_error[k, i, j] = fl.MAE(hat_y, y_test)
t_f = time.perf_counter()
print('\t***Train time: ' + str((t_f - t0) / 60.0))
type_exp += '_nominal'
dir_results = './results/'
np.save(dir_results + 'impact' + type_exp + '.npy', impact)
np.save(dir_results + 'pred_error' + type_exp + '.npy', pred_error)
t0 = time.perf_counter()
print('\t model: ' + str(i))
for k in range(max_num_models):
print('\tk=' + str(k))
if i in [0, 1]:
fa.get_prediction = get_prediction_a
fa.get_grad = get_grad_a
elif i in [2, 3]:
fa.get_prediction = get_prediction_b
fa.get_grad = get_grad_b
y_test, hat_y = fa.get_forecast(dir_rep, k)
predictions_i.append(hat_y)
error[i, k] = fl.MAE(y_test, hat_y)
#pdb.set_trace()
all_predictions.append(predictions_i)
t_f = time.perf_counter()
print('\t***Train time: ' + str((t_f - t0) / 60.0))
type_exp += '_test'
dir_results = './'
np.save(dir_results + 'MAE' + type_exp + '.npy', error)
np.save(dir_results + 'predictions' + type_exp + '.npy', all_predictions)
