def test_pso():
opp = op.OptimizationProblem('min', lambda x: x[:, 0]**2 - 4 * x[:, 0] + 4,
[0], [6])
# check pso with given 1d minimization problem we should be close to true solution
global_optimizer = go.GlobalOptimizer()
x_star, f_star = global_optimizer.pso(opp)
x_star = np.round(x_star, 2)
assert abs(2 - x_star[0]) <= 1e-1
# check pso with given 1d maximization problem we should be close to true solution
opp.mode = 'max'
global_optimizer = go.GlobalOptimizer()
x_star, f_star = global_optimizer.pso(opp)
x_star_true = 6
assert abs(x_star_true - x_star[0]) <= 1e-1
# check pso with 2d rosenbrock minimization problem
opp = op.OptimizationProblem(
'min', lambda x: (1 - x[:, 0])**2 + 100 * (x[:, 1] - x[:, 0]**2)**2,
[-5, -5], [10, 10])
global_optimizer = go.GlobalOptimizer()
x_star, f_star = global_optimizer.pso(opp)
x_star = np.round(x_star, 2)
x_star_true = np.array([1, 1])
err = np.round(max(abs(x_star_true - x_star)), 2)
assert err <= 1e-1
def test_ga():
opp = op.OptimizationProblem('min', lambda x: x[:, 0]**2 - 4 * x[:, 0] + 4,
[0], [6])
# check pso with given 1d minimization problem we should be close to true solution
global_optimizer = go.GlobalOptimizer()
x_star, f_star = global_optimizer.ga(opp)
x_star_true = 2
err = np.round(max(abs(x_star_true - x_star)), 2)
assert err <= 1e-1
# check pso with given 1d maximization problem we should be close to true solution
opp.mode = 'max'
global_optimizer = go.GlobalOptimizer()
x_star, f_star = global_optimizer.ga(opp)
x_star = np.round(x_star, 2)
x_star_true = 6
err = np.round(max(abs(x_star_true - x_star)), 2)
assert err <= 1e-1
# check pso with 2d rosenbrock minimization problem
opp = op.OptimizationProblem(
'min', lambda x: np.sin(x[:, 0] + x[:, 1]) +
(x[:, 0] - x[:, 1])**2 - 1.5 * x[:, 0] + 2.5 * x[:, 1] + 1, [-1.5, -3],
[4, 4])
global_optimizer = go.GlobalOptimizer()
x_star, f_star = global_optimizer.ga(opp)
x_star = np.round(x_star, 2)
x_star_true = np.array([-0.54, -1.54])
err = np.round(max(abs(x_star_true - x_star)), 2)
assert err <= 1e-1
def test_optimization_problem():
optimization_problem = op.OptimizationProblem(
'min', lambda x: x[:, 0]**2 - 4 * x[:, 0] + 4, [0], [6])
# # just check things and functions
assert optimization_problem.mode == 'min'
assert optimization_problem.lower_constraints == np.array([0])
assert optimization_problem.upper_constraints == np.array([6])
random_input = np.random.randn(2, 2)
assert optimization_problem.f(random_input)[0] == (
lambda x: x[:, 0]**2 - 4 * x[:, 0] + 4)(random_input)[0]
def fit(self, X, y, optimizer={}):
self.X_train = X.copy()
self.y_train = y.copy()
self.d = len(X[0])
lower_constraints = np.random.uniform(-1, 0, self.d)
upper_constraints = np.random.uniform(0, 1, self.d)
global_optimizer = go.GlobalOptimizer()
p = op.OptimizationProblem('min',
self.cost_function,
lower_constraints,
upper_constraints,
feasibility=False)
# default params all algorithms
w = 0.6
w_init = 0.9
N = 30
max_iter = 200
tol = 1e-4
trajectory = True
mutation = True
F = 2
CR = 0.7
self.lambda_ = 0
algo = 'pso'
if 'algo' in optimizer:
algo = optimizer['algo']
if 'w' in optimizer:
w = optimizer['w']
if 'w_init' in optimizer:
w_init = optimizer['w_init']
if 'N' in optimizer:
N = optimizer['N']
if 'max_iter' in optimizer:
max_iter = optimizer['max_iter']
if 'tol' in optimizer:
tol = optimizer['tol']
if 'trajectory' in optimizer:
trajectory = optimizer['trajectory']
if 'mutation' in optimizer:
mutation = optimizer['mutation']
if 'F' in optimizer:
F = optimizer['F']
if 'CR' in optimizer:
CR = optimizer['CR']
if 'lambda_' in optimizer:
self.lambda_ = optimizer['lambda_']
if algo == 'pso':
x_star, J_star, x_path, y_path = global_optimizer.pso(
p, trajectory=True, max_iter=max_iter, N=N, tol=tol, w=w)
elif algo == 'apso':
x_star, J_star, x_path, y_path = global_optimizer.adaptive_pso(
p,
trajectory=True,
max_iter=max_iter,
N=N,
tol=tol,
w_init=w_init)
elif algo == 'de':
x_star, J_star, x_path, y_path = global_optimizer.de(
p,
trajectory=True,
max_iter=max_iter,
N=N,
tol=tol,
F=F,
CR=CR)
elif algo == 'ga':
x_star, J_star, x_path, y_path = global_optimizer.ga(
p,
trajectory=True,
max_iter=max_iter,
N=N,
tol=tol,
mutation=mutation)
self.weights = x_star
self.cost_function_values = y_path