def test_training(self, array_type):
input_array = array_type(self.training_data)
KrigingClass = KrigingModel(input_array[0:3], regularization=False)
np.random.seed(0)
p = 2
bh_results = KrigingClass.parameter_optimization(p)
# Calculate other variables and parameters
optimal_theta, optimal_reg_param, optimal_mean, optimal_variance, optimal_cov_mat, opt_cov_inv, optimal_ymu = KrigingClass.optimal_parameter_evaluation(
bh_results.x, p)
# Training performance
training_ss_error, rmse_error, y_training_predictions = KrigingClass.error_calculation(optimal_theta, p,
optimal_mean, opt_cov_inv,
optimal_ymu,
KrigingClass.x_data_scaled,
KrigingClass.y_data)
r2_training = KrigingClass.r2_calculation(KrigingClass.y_data, y_training_predictions)
np.random.seed(0)
results = KrigingClass.training()
np.testing.assert_array_equal(results.optimal_weights, optimal_theta)
np.testing.assert_array_equal(results.regularization_parameter, optimal_reg_param)
np.testing.assert_array_equal(results.optimal_mean, optimal_mean)
np.testing.assert_array_equal(results.optimal_variance, optimal_variance)
np.testing.assert_array_equal(results.optimal_covariance_matrix, optimal_cov_mat)
np.testing.assert_array_equal(results.optimal_y_mu, optimal_ymu)
np.testing.assert_array_equal(results.output_predictions, y_training_predictions)
np.testing.assert_array_equal(results.training_R2, r2_training)
np.testing.assert_array_equal(results.training_rmse, rmse_error)
np.testing.assert_array_equal(results.optimal_p, p)
np.testing.assert_array_equal(results.x_data, KrigingClass.x_data)
np.testing.assert_array_equal(results.x_data_scaled, KrigingClass.x_data_scaled)
np.testing.assert_array_equal(results.x_data_min, KrigingClass.x_data_min)
np.testing.assert_array_equal(results.x_data_max, KrigingClass.x_data_max)
def test_error_calculation(self, array_type):
input_array = array_type(self.training_data)
KrigingClass = KrigingModel(input_array[0:3], regularization=False)
p = 2
var_vector = np.array([1, 2, 1.00000000e-06])
theta, reg_param, mean, variance, cov_mat, cov_inv, y_mu = KrigingClass.optimal_parameter_evaluation(
var_vector, p)
y_prediction_exp = np.zeros((KrigingClass.x_data_scaled.shape[0], 1))
for i in range(0, KrigingClass.x_data_scaled.shape[0]):
cmt = (np.matmul(((np.abs(KrigingClass.x_data_scaled[i, :] -
KrigingClass.x_data_scaled))**p),
theta)).transpose()
cov_matrix_tests = np.exp(-1 * cmt)
y_prediction_exp[i, 0] = mean + np.matmul(
np.matmul(cov_matrix_tests.transpose(), cov_inv), y_mu)
ss_error, rmse_error, y_prediction = KrigingClass.error_calculation(
theta, p, mean, cov_inv, y_mu, KrigingClass.x_data_scaled,
KrigingClass.y_data)
np.testing.assert_array_equal(y_prediction, y_prediction_exp)
assert np.sum((KrigingClass.y_data - y_prediction_exp)**
2) / KrigingClass.x_data_scaled.shape[0] == ss_error
assert np.sqrt(
np.sum((KrigingClass.y_data - y_prediction_exp)**2) /
KrigingClass.x_data_scaled.shape[0]) == rmse_error
def test_r2_calculation(self, array_type):
input_array = array_type(self.training_data)
KrigingClass = KrigingModel(input_array[0:3], regularization=False)
p = 2
var_vector = np.array([1, 2, 1.00000000e-06])
theta, reg_param, mean, variance, cov_mat, cov_inv, y_mu = KrigingClass.optimal_parameter_evaluation(var_vector, p)
ss_error, rmse_error, y_prediction = KrigingClass.error_calculation(theta, p, mean, cov_inv, y_mu,
KrigingClass.x_data_scaled, KrigingClass.y_data)
r_square = KrigingClass.r2_calculation(KrigingClass.y_data, y_prediction)
assert 0.999999999999 == r_square