def __init__(self, problem_idx):
SyntheticEnv.__init__(self, problem_idx)
self.name = 'synthetic_griewank'
if problem_idx == 0:
self.dim_x = 3
self.feasible_action_value_threshold = -2
elif problem_idx == 1:
self.dim_x = 10
self.feasible_action_value_threshold = -50
elif problem_idx == 2:
self.dim_x = 20
self.feasible_action_value_threshold = -50
self.feasible_reward = 100
self.reward_function = lambda sol: -benchmarks.griewank(sol)[0]
def get_objective_function(sol):
if obj_fcn == 'shekel':
return shekel(sol, A, C)[0]
elif obj_fcn == 'schwefel':
return -benchmarks.schwefel(sol)[0]
elif obj_fcn == 'griewank':
return -benchmarks.griewank(sol)[0]
elif obj_fcn == 'rastrigin':
return -benchmarks.rastrigin(sol)[0]
elif obj_fcn == 'ackley':
return -benchmarks.ackley(sol)[0]
elif obj_fcn == 'rosenbrock':
return -benchmarks.rosenbrock(sol)[0]
elif obj_fcn == 'schaffer':
return -benchmarks.schaffer(sol)[0]
else:
print "wrong function name"
sys.exit(-1)
def test_function(param):
"""
Griewank testing function (the minus is to convert the problem into a maximization)
"""
return -1 * griewank((param["x0"], param["x1"], param["x2"]))[0],
def griewank_on_cube(u: [float]) -> float:
# https://deap.readthedocs.io/en/master/api/benchmarks.html#deap.benchmarks.griewank
u_squished = [1200 * (ui**1.1 - 0.5) for ui in u]
return benchmarks.griewank(u_squished)[0]
def evaluate(self, x):
return griewank(x)[0]
def griewank_arg0(sol):
return benchmarks.griewank(sol)[0]
def griewank_arg0(sol):
return benchmarks.griewank(sol)[0]
_delta = self._fBest_DoE - self.fopt
if self.iter_count > 1 and \
np.mean(self._hist_EI[0:min(3, self.iter_count - 1)]) < 0.01 * _delta:
self.stop_dict['low-EI'] = np.mean(self._hist_EI)
if self.eval_count >= (self.max_FEs / 2):
self.stop_dict['max_FEs'] = self.eval_count
return super().check_stop()
np.random.seed(42)
dim = 2
max_FEs = 100
obj_fun = lambda x: benchmarks.griewank(x)[0]
lb, ub = -600, 600
search_space = ContinuousSpace([lb, ub]) * dim
mean = constant_trend(dim, beta=None)
# autocorrelation parameters of GPR
thetaL = 1e-10 * (ub - lb) * np.ones(dim)
thetaU = 10 * (ub - lb) * np.ones(dim)
theta0 = np.random.rand(dim) * (thetaU - thetaL) + thetaL
model = GaussianProcess(corr='squared_exponential',
theta0=theta0,
thetaL=thetaL,
thetaU=thetaU,
nugget=1e-6,
def fitness(x):
return benchmarks.griewank(x)[0]