def __init__(self, model: Union[IModel, IDifferentiable, IEntropySearchModel], cost_model: IModel,
space: ParameterSpace, sampler: McmcSampler = None, num_samples: int = 400,
num_representer_points: int = 50, proposal_function: Acquisition = None,
burn_in_steps: int = 50) -> None:
"""
Entropy Search acquisition function approximates the distribution of the global
minimum and tries to decrease its entropy. See this paper for more details:
P. Hennig and C. J. Schuler
Entropy search for information-efficient global optimization
Journal of Machine Learning Research, 13, 2012
:param model: GP model to compute the distribution of the minimum dubbed pmin.
:param space: Domain space which we need for the sampling of the representer points
:param sampler: mcmc sampler for representer points
:param num_samples: integer determining how many samples to draw for each candidate input
:param num_representer_points: integer determining how many representer points to sample
:param proposal_function: Function that defines an unnormalized log proposal measure from which to sample the
representer points. The default is expected improvement.
:param burn_in_steps: integer that defines the number of burn-in steps when sampling the representer points
"""
super().__init__()
self.cost_model = cost_model
self.es = EntropySearch(model, space, sampler,
num_samples, num_representer_points,
proposal_function, burn_in_steps)
def entropy_search_acquisition(gpy_model, continuous_space):
sampler = AffineInvariantEnsembleSampler(continuous_space)
return EntropySearch(gpy_model, continuous_space, sampler, num_representer_points=10)
with_gradients = False
elif args.model_type == "gp":
model = BOGP(X_init=X_init, Y_init=Y_init)
if args.acquisition_type == "ei":
acquisition = ExpectedImprovement(model)
elif args.acquisition_type == "pi":
acquisition = ProbabilityOfImprovement(model)
elif args.acquisition_type == "nlcb":
acquisition = NegativeLowerConfidenceBound(model)
elif args.acquisition_type == "logei":
acquisition = LogExpectedImprovement(model)
elif args.acquisition_type == "entropy_search":
model = BOGP(X_init=X_init, Y_init=Y_init)
acquisition = EntropySearch(model, space=space)
# if with_gradients:
# acquisition_optimizer = AcquisitionOptimizer(space)
# else:
acquisition_optimizer = DirectOptimizer(space)
candidate_point_calculator = Sequential(acquisition, acquisition_optimizer)
bo = BayesianOptimizationLoop(model=model, space=space, X_init=X_init, Y_init=Y_init, acquisition=acquisition,
candidate_point_calculator=candidate_point_calculator)
bo.run_loop(user_function=obj, stopping_condition=FixedIterationsStoppingCondition(args.num_iterations))
curr_inc = np.inf
traj = []