void TestObjectiveFunctional::test_calculate_potential_evaluation(void)
{
message += "test_calculate_potential_evaluation\n";
MultilayerPerceptron mlp;
Vector<double> parameters;
MockObjectiveFunctional mof(&mlp);
double evaluation;
double potential_evaluation;
// Only neural parameters
mlp.set_network_architecture(1,1,1);
mlp.set_independent_parameters_number(0);
evaluation = mof.calculate_evaluation();
parameters = mlp.get_parameters();
potential_evaluation = mof.calculate_potential_evaluation(parameters);
assert_true(evaluation == potential_evaluation, LOG);
// Only independent parameters
mlp.set_network_architecture(0,0,0);
mlp.set_independent_parameters_number(1);
evaluation = mof.calculate_evaluation();
parameters = mlp.get_parameters();
potential_evaluation = mof.calculate_potential_evaluation(parameters);
assert_true(evaluation == potential_evaluation, LOG);
// Both neural and independent parameters
mlp.set_network_architecture(1,1,1);
mlp.set_independent_parameters_number(1);
evaluation = mof.calculate_evaluation();
parameters = mlp.get_parameters();
potential_evaluation = mof.calculate_potential_evaluation(parameters);
assert_true(evaluation == potential_evaluation, LOG);
}
void TestObjectiveFunctional::test_calculate_independent_parameters_gradient_central_differences(void)
{
message += "test_calculate_independent_parameters_gradient_central_differences\n";
MultilayerPerceptron mlp;
MockObjectiveFunctional mof(&mlp);
int independent_parameters_number;
Vector<double> independent_parameters_gradient;
// Test
mlp.set();
independent_parameters_number = mlp.get_independent_parameters_number();
independent_parameters_gradient = mof.calculate_independent_parameters_gradient_central_differences();
assert_true(independent_parameters_gradient.get_size() == independent_parameters_number, LOG);
// Test
mlp.set(1,1,1);
independent_parameters_number = mlp.get_independent_parameters_number();
independent_parameters_gradient = mof.calculate_independent_parameters_gradient_central_differences();
assert_true(independent_parameters_gradient.get_size() == independent_parameters_number, LOG);
// Test
mlp.set(1);
independent_parameters_number = mlp.get_independent_parameters_number();
independent_parameters_gradient = mof.calculate_independent_parameters_gradient_central_differences();
assert_true(independent_parameters_gradient.get_size() == independent_parameters_number, LOG);
}
void TestEvolutionaryAlgorithm::test_set_default(void)
{
message += "test_set_default\n";
MultilayerPerceptron mlp;
MockObjectiveFunctional mof(&mlp);
EvolutionaryAlgorithm ea(&mof);
// Test
ea.set_default();
assert_true(ea.get_population_size() == 0, LOG);
// Test
mlp.set(1);
ea.set_default();
assert_true(ea.get_population_size() == 10, LOG);
}
void NormalizedSquaredErrorTest::test_calculate_performance(void)
{
message += "test_calculate_performance\n";
Vector<double> parameters;
NeuralNetwork nn(1,1);
MultilayerPerceptron* mlpp = nn.get_multilayer_perceptron_pointer();
mlpp->get_layer_pointer(0)->set_activation_function(Perceptron::Linear);
mlpp->initialize_biases(0.0);
mlpp->initialize_synaptic_weights(1.0);
DataSet ds(1,1,2);
Matrix<double> new_data(2, 2);
new_data[0][0] = -1.0;
new_data[0][1] = -1.0;
new_data[1][0] = 1.0;
new_data[1][1] = 1.0;
ds.set_data(new_data);
NormalizedSquaredError nse(&nn, &ds);
assert_true(nse.calculate_performance() == 0.0, LOG);
// Test
nn.set(1, 1);
nn.randomize_parameters_normal();
parameters = nn.arrange_parameters();
ds.set(1, 1, 2);
ds.randomize_data_normal();
assert_true(nse.calculate_performance() == nse.calculate_performance(parameters), LOG);
}
void TestObjectiveFunctional::test_calculate_potential_Hessian(void)
{
message += "test_calculate_potential_Hessian\n";
MultilayerPerceptron mlp;
int parameters_number;
Vector<double> parameters;
MockObjectiveFunctional mof(&mlp);
Matrix<double> potential_Hessian;
mlp.set_network_architecture(1, 1, 1);
mlp.set_independent_parameters_number(0);
mlp.initialize_parameters(0.0);
parameters_number = mlp.get_parameters_number();
parameters = mlp.get_parameters();
potential_Hessian = mof.calculate_potential_Hessian(parameters);
assert_true(potential_Hessian.get_rows_number() == parameters_number, LOG);
assert_true(potential_Hessian.get_columns_number() == parameters_number, LOG);
mlp.set_network_architecture(0, 0, 0);
mlp.set_independent_parameters_number(1);
mlp.initialize_parameters(0.0);
parameters_number = mlp.get_parameters_number();
parameters = mlp.get_parameters();
potential_Hessian = mof.calculate_potential_Hessian(parameters);
assert_true(potential_Hessian.get_rows_number() == parameters_number, LOG);
assert_true(potential_Hessian.get_columns_number() == parameters_number, LOG);
mlp.set_network_architecture(1, 1, 1);
mlp.set_independent_parameters_number(1);
mlp.initialize_parameters(0.0);
parameters_number = mlp.get_parameters_number();
parameters = mlp.get_parameters();
potential_Hessian = mof.calculate_potential_Hessian(parameters);
assert_true(potential_Hessian.get_rows_number() == parameters_number, LOG);
assert_true(potential_Hessian.get_columns_number() == parameters_number, LOG);
}
void TestObjectiveFunctional::test_calculate_potential_gradient(void)
{
message += "test_calculate_potential_gradient\n";
MultilayerPerceptron mlp;
MockObjectiveFunctional mof(&mlp);
Vector<double> parameters;
// Only neural parameters
mlp.set_network_architecture(1, 1, 1);
mlp.set_independent_parameters_number(0);
mlp.initialize_parameters(0.0);
parameters = mlp.get_parameters();
assert_true(mof.calculate_potential_gradient(parameters) == 0.0, LOG);
// Only independent parameters
mlp.set_network_architecture(0, 0, 0);
mlp.set_independent_parameters_number(1);
mlp.initialize_parameters(0.0);
parameters = mlp.get_parameters();
assert_true(mof.calculate_potential_gradient(parameters) == 0.0, LOG);
// Both neural and independent parameters
mlp.set_network_architecture(1, 1, 1);
mlp.set_independent_parameters_number(1);
mlp.initialize_parameters(0.0);
parameters = mlp.get_parameters();
assert_true(mof.calculate_potential_gradient(parameters) == 0.0, LOG);
}