def test_qEI(self):
for double in (True, False):
self._setUp(double=double)
qEI = qExpectedImprovement(self.model_st, best_f=0.0)
candidates, _ = optimize_acqf(
acq_function=qEI,
bounds=self.bounds,
q=3,
num_restarts=10,
raw_samples=20,
options={"maxiter": 5},
)
self.assertTrue(torch.all(-EPS <= candidates))
self.assertTrue(torch.all(candidates <= 1 + EPS))
qEI = qExpectedImprovement(self.model_fn, best_f=0.0)
candidates, _ = optimize_acqf(
acq_function=qEI,
bounds=self.bounds,
q=3,
num_restarts=10,
raw_samples=20,
options={"maxiter": 5},
)
self.assertTrue(torch.all(-EPS <= candidates))
self.assertTrue(torch.all(candidates <= 1 + EPS))
candidates_batch_limit, _ = optimize_acqf(
acq_function=qEI,
bounds=self.bounds,
q=3,
num_restarts=10,
raw_samples=20,
options={"maxiter": 5, "batch_limit": 5},
)
self.assertTrue(torch.all(-EPS <= candidates_batch_limit))
self.assertTrue(torch.all(candidates_batch_limit <= 1 + EPS))
def find_eta_c(self, rho_t):
"""
Find the minimum of the posterior mean, i.e., min_x mu(x|D) s.t. Pr(g(x) <= 0) > rho_t, where D is the data set D={Y,X}, mu(x|D)
is the posterior mean of the GP queried at location x, and rho_t depends on the current budget of failures.
"""
self.my_print("Finding min_x mu(x|D) s.t. Pr(g(x) <= 0) > 0.99")
self.gp_mean_obj_cons.rho_t = rho_t
options = {
"batch_limit": 1,
"maxiter": 200,
"ftol": 1e-6,
"method": self.method_opti
}
x_eta_c, _ = optimize_acqf(acq_function=self.gp_mean_obj_cons,
bounds=self.bounds,
q=1,
num_restarts=self.Nrestarts_eta_c,
raw_samples=500,
return_best_only=True,
options=options)
self.my_print("Done!")
# Revaluate the mean (because the one returned might be affected by the constant that is added to ensure non-negativity)
eta_c = self.model_list[0](x_eta_c).mean.view(1)
return x_eta_c, eta_c
def optimize_acqf_and_get_observation(acq_func, args):
"""
Optimizes the acquisition function, and returns a new candidate and a noisy observation.
botorch defaults: num_restarts=10, raw_samples=256, batch_limit=5, maxiter=200
"""
batch_initial_conditions = gen_one_shot_kg_initial_conditions(
acq_function=acq_func,
bounds=bo_bounds,
q=1,
num_restarts=args.acqf_opt_num_restarts,
raw_samples=args.acqf_opt_raw_samples,
options={"batch_limit": args.acqf_opt_batch_limit,
"maxiter": args.acqf_opt_maxiter},
)
# optimize acquisition function
candidates, _ = optimize_acqf(
acq_function=acq_func,
bounds=bo_bounds,
q=1,
num_restarts=args.acqf_opt_num_restarts,
raw_samples=args.acqf_opt_raw_samples, # used for intialization heuristic
options={"batch_limit": args.acqf_opt_batch_limit,
"maxiter": args.acqf_opt_maxiter},
batch_initial_conditions=batch_initial_conditions
)
# proposed evaluation
new_theta = candidates.detach()
# observe new noisy function evaluation
new_G, new_G_sem = composite_simulation(new_theta.squeeze())
return new_theta, new_G, new_G_sem
def argmax_posterior_mean(cands: to.Tensor, cands_values: to.Tensor,
ddp_space: BoxSpace, num_restarts: int,
num_samples: int) -> to.Tensor:
"""
Compute the GP input with the maximal posterior mean.
:param cands: candidates a.k.a. x
:param cands_values: observed values a.k.a. y
:param ddp_space: space of the domain distribution parameters, indicates the lower and upper bound
:param num_restarts: number of restarts for the optimization of the acquisition function
:param num_samples: number of samples for the optimization of the acquisition function
:return: un-normalized candidate with maximum posterior value a.k.a. x
"""
if not isinstance(cands, to.Tensor):
raise pyrado.TypeErr(given=cands, expected_type=to.Tensor)
if not isinstance(cands_values, to.Tensor):
raise pyrado.TypeErr(given=cands_values, expected_type=to.Tensor)
if not isinstance(ddp_space, BoxSpace):
raise pyrado.TypeErr(given=ddp_space, expected_type=BoxSpace)
# Normalize the input data and standardize the output data
uc_projector = UnitCubeProjector(
to.from_numpy(ddp_space.bound_lo).to(dtype=to.get_default_dtype()),
to.from_numpy(ddp_space.bound_up).to(dtype=to.get_default_dtype()),
)
cands_norm = uc_projector.project_to(cands)
cands_values_stdized = standardize(cands_values)
if cands_norm.shape[0] > cands_values.shape[0]:
print_cbt(
f"There are {cands.shape[0]} candidates but only {cands_values.shape[0]} evaluations. Ignoring "
f"the candidates without evaluation for computing the argmax.",
"y",
)
cands_norm = cands_norm[:cands_values.shape[0], :]
# Create and fit the GP model
gp = SingleTaskGP(cands_norm, cands_values_stdized)
gp.likelihood.noise_covar.register_constraint("raw_noise",
GreaterThan(1e-5))
mll = ExactMarginalLogLikelihood(gp.likelihood, gp)
fit_gpytorch_model(mll)
# Find position with maximal posterior mean
cand_norm, _ = optimize_acqf(
acq_function=PosteriorMean(gp),
bounds=to.stack(
[to.zeros(ddp_space.flat_dim),
to.ones(ddp_space.flat_dim)]).to(dtype=to.float32),
q=1,
num_restarts=num_restarts,
raw_samples=num_samples,
)
cand_norm = cand_norm.to(dtype=to.get_default_dtype())
cand = uc_projector.project_back(cand_norm.detach())
print_cbt(f"Converged to argmax of the posterior mean: {cand.numpy()}",
"g",
bright=True)
return cand
def run():
train_x, train_obj, train_con, best_observed_value_nei = generate_initial_data(n=10)
# define models for objective and constraint
model_obj = FixedNoiseGP(train_x, train_obj, train_yvar.expand_as(train_obj)).to(train_x)
model_con = FixedNoiseGP(train_x, train_con, train_yvar.expand_as(train_con)).to(train_x)
# combine into a multi-output GP model
model = ModelListGP(model_obj, model_con)
mll = SumMarginalLogLikelihood(model.likelihood, model)
fit_gpytorch_model(mll)
acqui_gpmean_cons = GPmeanConstrained(model=model,objective=constrained_obj)
# Forward:
# X = torch.rand(size=(1,6))
# acqui_gpmean_cons.forward(X)
method_opti = "SLSQP" # constraints
# method_opti = "COBYLA" # constraints
# method_opti = "L-BFGS-B"
# Below, num_restarts must be equal to q, otherwise, it fails...
options = {"batch_limit": 1,"maxiter": 200,"ftol":1e-6,"method":method_opti}
x_eta_c, eta_c = optimize_acqf(acq_function=acqui_gpmean_cons,bounds=bounds,q=1,num_restarts=1,
raw_samples=500,return_best_only=True,options=options)
pdb.set_trace()
def optimize_acqf_and_get_observation(acq_func, obj_func, time_list,
global_start_time):
"""Optimizes the acquisition function, and returns a new candidate and observation."""
# optimize
candidates, _ = optimize_acqf(
acq_function=acq_func,
bounds=standard_bounds,
q=1,
num_restarts=20,
raw_samples=1024, # used for initialization heuristic
options={
"batch_limit": 5,
"maxiter": 200,
"nonnegative": True
},
sequential=True,
)
# observe new values
new_x = candidates.detach()
new_obj = []
new_con = []
for x in new_x:
res = obj_func(x)
y = res['objs']
c = res['constraints']
new_obj.append(y)
new_con.append(c)
global_time = time.time() - global_start_time
time_list.append(global_time)
new_obj = torch.tensor(new_obj, **tkwargs).reshape(new_x.shape[0], -1)
new_con = torch.tensor(new_con, **tkwargs).reshape(new_x.shape[0], -1)
print(f'evaluate {new_x.shape[0]} configs on real objective')
return new_x, new_obj, new_con
def test(dim=1):
assert dim == 1
X0 = torch.Tensor([[0.93452506],
[0.18872502],
[0.89790337],
[0.95841797],
[0.82335255],
[0.45000000],
[0.50000000]])
Y0 = torch.Tensor([-0.4532849,-0.66614552,-0.92803395,0.08880341,-0.27683621,1.000000,1.500000])
# Y0 = Y0[:,None]
Neval = Y0.shape[0]
train_x = X0
train_y = Y0[:,None]
Nrestarts = 10
model = SingleTaskGP(train_x, train_y)
EI = ExpectedImprovement(model, best_f=0.2)
print("[get_next_point()] Computing next candidate by maximizing the acquisition function ...")
options={"batch_limit": 50,"maxiter": 200,"ftol":1e-9,"method":"L-BFGS-B","iprint":2,"maxls":30,"disp":True}
x_next,alpha_next = optimize_acqf(acq_function=EI,bounds=torch.Tensor([[0.0]*dim,[1.0]*dim],device=device),q=1,num_restarts=Nrestarts,
raw_samples=500,return_best_only=True,options=options)
print("x_next:",x_next)
print("alpha_next:",alpha_next)
def find_eta(self):
"""
Find the minimum of the posterior mean, i.e., min_x mu(x|D), where D is the data set D={Y,X}, and mu(x|D)
is the posterior mean of the GP queried at location x. For this, we do local optimization with random
restarts.
"""
logger.info("Finding min_x mu(x|D)...")
options = {
"batch_limit": 50,
"maxiter": 200,
"ftol": 1e-6,
"method": self.method_opti
}
x_eta, _ = optimize_acqf(acq_function=self.gp_mean,
bounds=self.bounds,
q=1,
num_restarts=self.Nrestarts_eta,
raw_samples=500,
return_best_only=True,
options=options)
logger.info("Done!")
# Revaluate the mean (because the one returned might be affected by the constant that is added to ensure non-negativity)
eta = self.model(x_eta).mean.view(1)
return x_eta, eta
def get_risky_evaluation(self):
# Gather fmin samples, using the Frechet distribution:
fmin_samples = get_fmin_samples_from_gp(
model=self.model_list[0],
Nsamples=self.Nsamples_fmin,
eta=self.eta_c) # This assumes self.eta has been updated
self.update_u_vec(fmin_samples)
self.which_mode = "risky"
self.my_print(
"[get_risky_evaluation()] Computing next candidate by maximizing the acquisition function ..."
)
options = {
"batch_limit": 50,
"maxiter": 300,
"ftol": 1e-9,
"method": self.method_risky,
"iprint": 2,
"maxls": 20,
"disp": self.disp_info_scipy_opti
}
x_next, alpha_next = optimize_acqf(acq_function=self,
bounds=self.bounds,
q=1,
num_restarts=self.Nrestarts_risky,
raw_samples=500,
return_best_only=True,
options=options)
self.my_print("Done!")
return x_next, alpha_next
def current_best(self, past_only: bool = False, **kwargs) -> Tuple[Tensor, Tensor]:
"""
Get the current best solution and value
:param past_only: If True, optimization is over previously evaluated points only.
:param kwargs: ignored
:return: Current best solution and value
"""
inner = PosteriorMean(self.model)
if past_only:
with torch.no_grad():
values = inner(self.X.reshape(-1, 1, self.dim_x))
best = torch.argmax(values)
current_best_sol = self.X[best]
current_best_value = -values[best]
else:
current_best_sol, current_best_value = optimize_acqf(
acq_function=inner,
bounds=self.bounds,
q=1,
num_restarts=self.num_restarts,
raw_samples=self.num_restarts * self.raw_multiplier,
)
# negated again to report the correct value
if self.verbose:
print(
"Current best solution, value: ", current_best_sol, -current_best_value
)
return current_best_sol, -current_best_value
def get_candidate(model, acq, full_train_Y, q, bounds, dim):
if acq == 'EI':
if q == 1:
EI = ExpectedImprovement(model, full_train_Y.max().item())
else:
EI = qExpectedImprovement(model, full_train_Y.max().item())
bounds_t = torch.FloatTensor([[bounds[0]] * dim, [bounds[1]] * dim])
candidate, acq_value = optimize_acqf(
EI,
bounds=bounds_t,
q=q,
num_restarts=15,
raw_samples=5000,
)
elif acq == 'TS':
sobol = SobolEngine(dim, scramble=True)
n_candidates = min(5000, max(20000, 2000 * dim))
pert = sobol.draw(n_candidates)
X_cand = (bounds[1] - bounds[0]) * pert + bounds[0]
thompson_sampling = MaxPosteriorSampling(model=model,
replacement=False)
candidate = thompson_sampling(X_cand, num_samples=q)
else:
raise NotImplementedError('Only TS and EI are implemented')
return candidate, EI if acq == 'EI' else None
def get_bayes_pop(self, n_suggestions):
"""
Parameters
----------
n_suggestions: Number of new suggestions/trial solutions to generate using BO
Returns
The new set of trial solutions obtained by optimizing the acquisition function
-------
"""
try:
candidates, _ = optimize_acqf(
acq_function=self.acquisition,
bounds=self.min_max_bounds,
q=n_suggestions,
num_restarts=10,
raw_samples=512, # used for initialization heuristic
sequential=True)
bayes_pop = unnormalize(candidates, self.torch_bounds).numpy()
except Exception as e:
print('Error in get_bayes_pop(): {}'.format(e))
population = self.search_space.unwarp(
bayes_pop) # Translate the solution back to the original space
return population
def bo_loop(gp_model: SingleTaskGP, acq_func_id: str,
acq_func_kwargs: Dict[str, Any], acq_func_opt_kwargs: Dict[str,
Any],
bounds: Tensor, tkwargs: Dict[str, Any], q: int, num_restarts: int,
raw_initial_samples, seed: int,
num_MC_sample_acq: int) -> Iterable[Any]:
# seed everything
np.random.seed(seed)
torch.manual_seed(seed)
# put on proper device
# we want to maximize
fmax = torch.quantile(gp_model.train_targets, .9).item()
print(f"Using good point cutoff {fmax:.2f}")
device = gp_model.train_inputs[0].device
bounds = bounds.to(**tkwargs)
gp_model.eval()
acq_func_kwargs['best_f'] = fmax
acq_func = query_acq_func(acq_func_id=acq_func_id,
acq_func_kwargs=acq_func_kwargs,
gp_model=gp_model,
q=q,
num_MC_samples_acq=num_MC_sample_acq
) # if q is 1 use analytic acquisitions
acq_func.to(**tkwargs)
options = {
'batch_limit': 100
} if acq_func_opt_kwargs == {} else acq_func_opt_kwargs
print("Start acquisition function optimization...")
if q == 1:
# use optimize_acq (with LBFGS)
candidate, acq_value = optimize_acqf(acq_function=acq_func,
bounds=bounds,
q=q,
num_restarts=num_restarts,
raw_samples=raw_initial_samples,
return_best_only=True,
options=options)
else:
candidate, acq_value = optimize_acqf_torch(
acq_function=acq_func,
bounds=bounds,
q=q,
num_restarts=num_restarts,
raw_samples=raw_initial_samples,
return_best_only=True,
options=options,
)
print(f"Acquired {candidate} with acquisition value {acq_value}")
return candidate.to(device=device)
def optimize_loop(model, loss, X_train, y_train, bounds, n_samples=10):
best_value = y_train.max()
acq_func = qExpectedImprovement(model, best_f=best_value)
X_new, acq_value = optimize_acqf(acq_func,
bounds=bounds,
q=20,
num_restarts=10,
raw_samples=76)
#X_new = X_new.view((n_samples,-1))
print(X_new)
return X_new
def optimize_acqf_and_get_observation(self, acq, seed):
"""Optimizes the acquisition function, and returns a new candidate."""
init = initializers.gen_batch_initial_conditions(
acq, self.bounds, options={"seed": seed}, **self.optim_kwargs)
# optimize
candidate, acq_value = optimize_acqf(acq,
bounds=self.bounds,
batch_initial_conditions=init,
**self.optim_kwargs)
# observe new value
new_x = candidate.detach() #self.scale_to_bounds(candidate.detach())
return new_x
def optimize_acqui_use_restarts_as_batch(self, options):
logger.info("[get_next_point()] Optimizing acquisition function ...")
x_next, alpha_next = optimize_acqf(acq_function=self,
bounds=self.bounds,
q=1,
num_restarts=self.Nrestarts,
raw_samples=500,
return_best_only=True,
options=options)
return x_next, alpha_next
def _update_current_best(self) -> None:
"""
Updates the current best solution and corresponding value
"""
pm = PosteriorMean(self.model)
self.current_best_sol, self.current_best_val = optimize_acqf(
pm,
Tensor([[0], [1]]).repeat(1, self.dim),
q=1,
num_restarts=self.num_restarts,
raw_samples=self.raw_samples,
)
def test_EI(self):
for double in (True, False):
self._setUp(double=double)
EI = ExpectedImprovement(self.model_st, best_f=0.0)
candidates, _ = optimize_acqf(
acq_function=EI,
bounds=self.bounds,
q=1,
num_restarts=10,
raw_samples=20,
options={"maxiter": 5},
)
self.assertTrue(-EPS <= candidates <= 1 + EPS)
EI = ExpectedImprovement(self.model_fn, best_f=0.0)
candidates, _ = optimize_acqf(
acq_function=EI,
bounds=self.bounds,
q=1,
num_restarts=10,
raw_samples=20,
options={"maxiter": 5},
)
self.assertTrue(-EPS <= candidates <= 1 + EPS)
def select_next_points_botorch(observed_X: List[List[float]],
observed_y: List[float]) -> np.ndarray:
"""Generate the next sample to evaluate with XTB
Uses BOTorch to pick the next sample using Expected Improvement
Args:
observed_X: Observed coordinates
observed_y: Observed energies
Returns:
Next coordinates to try
"""
# Clip the energies if needed
observed_y = np.clip(observed_y, -np.inf,
2 + np.log10(np.clip(observed_y, 1, np.inf)))
# Convert inputs to torch arrays
train_X = torch.tensor(observed_X, dtype=torch.float)
train_y = torch.tensor(observed_y, dtype=torch.float)
train_y = train_y[:, None]
train_y = standardize(-1 * train_y)
# Make the GP
gp = SingleTaskGP(train_X,
train_y,
covar_module=gpykernels.ScaleKernel(
gpykernels.ProductStructureKernel(
num_dims=train_X.shape[1],
base_kernel=gpykernels.PeriodicKernel(
period_length_prior=NormalPrior(360, 0.1)))))
mll = ExactMarginalLogLikelihood(gp.likelihood, gp)
fit_gpytorch_model(mll)
# Solve the optimization problem
# Following boss, we use Eq. 5 of https://arxiv.org/pdf/1012.2599.pdf with delta=0.1
n_sampled, n_dim = train_X.shape
kappa = np.sqrt(
2 *
np.log10(np.power(n_sampled, n_dim / 2 + 2) * np.pi**2 /
(3.0 * 0.1))) # Results in more exploration over time
ei = UpperConfidenceBound(gp, kappa)
bounds = torch.zeros(2, train_X.shape[1])
bounds[1, :] = 360
candidate, acq_value = optimize_acqf(ei,
bounds=bounds,
q=1,
num_restarts=64,
raw_samples=64)
return candidate.detach().numpy()[0, :]
def _sample(self, candidates: Optional[np.array] = None) -> np.array:
if len(self.X_observed) < self.num_initial_random_draws:
return self.initial_sampler.sample(candidates=candidates)
else:
z_observed = torch.Tensor(self.transform_outputs(self.y_observed.numpy()))
with torch.no_grad():
# both (n, 1)
#mu_pred, sigma_pred = self.thompson_sampling.prior(self.X_observed)
mu_pred, sigma_pred = self.initial_sampler.prior.predict(self.X_observed)
mu_pred = torch.Tensor(mu_pred)
sigma_pred = torch.Tensor(sigma_pred)
# (n, 1)
r_observed = residual_transform(z_observed, mu_pred, sigma_pred)
# build and fit GP on residuals
gp = SingleTaskGP(
train_X=self.X_observed,
train_Y=r_observed,
likelihood=GaussianLikelihood(noise_constraint=GreaterThan(1e-3)),
)
mll = ExactMarginalLogLikelihood(gp.likelihood, gp)
fit_gpytorch_model(mll)
acq = ShiftedExpectedImprovement(
model=gp,
best_f=z_observed.min(dim=0).values,
mean_std_predictor=self.initial_sampler.prior.predict,
maximize=False,
)
if candidates is None:
candidate, acq_value = optimize_acqf(
acq,
bounds=self.bounds_tensor,
q=1,
num_restarts=5,
raw_samples=100,
)
# import matplotlib.pyplot as plt
# x = torch.linspace(-1, 1).unsqueeze(dim=-1)
# x = torch.cat((x, x * 0), dim=1)
# plt.plot(x[:, 0].flatten().tolist(), acq(x.unsqueeze(dim=1)).tolist())
# plt.show()
return candidate[0]
else:
# (N,)
ei = acq(torch.Tensor(candidates).unsqueeze(dim=-2))
return torch.Tensor(candidates[ei.argmax()])
def gen(self, num_points=1, **kwargs):
self.model.eval()
train_x = self.model.train_inputs[0]
acq = self._get_acquisition_fn()
new_candidate, batch_acq_values = optimize_acqf(
acq_function=acq,
bounds=torch.tensor(np.c_[self.lb, self.ub]).T.to(train_x),
q=num_points,
num_restarts=self.restarts,
raw_samples=self.samps,
)
return new_candidate.numpy()
def optimize_acqf_and_get_observation(acq_func, dataset):
"""Optimizes the acquisition function, and returns a new candidate and a noisy observation."""
# optimize
candidates, _ = optimize_acqf(
acq_function=acq_func,
bounds=bounds,
q=BATCH_SIZE,
num_restarts=10,
raw_samples=512, # used for intialization heuristic
options={"batch_limit": 5, "maxiter": 200},
)
# observe new values
new_x = candidates.detach()
exact_obj = call_external(new_x, BATCH_SIZE, 5, dataset).unsqueeze(-1) # add output dimension
new_obj = exact_obj + NOISE_SE * torch.randn_like(exact_obj)
return new_x, new_obj
def optimize_acqf_and_get_observation(acq_func, mean_y=None, std_y=None):
"""optimize acquisition function and evaluate new candidates
"""
# optimize
candidates, _ = optimize_acqf(
acq_function=acq_func,
bounds=bounds,
q=BATCH_SIZE,
num_restarts=10,
raw_samples=512, # used for intialization heuristic
)
# observe new values
new_x = candidates.detach()
new_y = evaluate_y(new_x, mean_y=mean_y, std_y=std_y)
return new_x, new_y
def propose_candidates(args, prev_res):
## train a surrogate model
gp = SingleTaskGP(prev_res['hparams'], prev_res['acc'])
mll = ExactMarginalLogLikelihood(gp.likelihood, gp)
fit_gpytorch_model(mll)
## construct an acquisition function and optimize it
UCB = UpperConfidenceBound(gp, beta=0.1)
candidates, acq_value = optimize_acqf(UCB,
bounds=bounds,
q=args.n_cand,
num_restarts=5,
raw_samples=20)
return candidates
def optimize_loop(model, loss, X_train, y_train, X_test, y_test, bounds):
best_value = y_train.min()
acq_func = ExpectedImprovement(model, best_f=best_value, maximize=False)
X_new, acq_value = optimize_acqf(acq_func, bounds=bounds, q=1, \
num_restarts=100, raw_samples=100)
X_new = X_new.view((1, 1))
y_new = (eval_true_func(X_new) - y_mean) / y_std
# concatenate new points to training set
X_train_new = torch.cat((X_train, X_new))
y_train_new = torch.cat((y_train, y_new))
plot_acq_func(acq_func, X_test=X_test, X_train=X_train_new, X_new=X_new)
# condition model on new observation
gpr_model, gpr_mll = get_gpr_model(X_new, y_new, model=model)
# plot model performance
plot_testing(gpr_model, X_test=X_test, X_train=X_train_new, \
y_train=y_train_new, y_test=y_test, X_new=X_new, y_new=y_new)
return gpr_model, gpr_mll, X_train_new, y_train_new
def _maximize_kg(self) -> None:
"""
maximizes the KG acquisition function and stores the resulting value and
the candidate
"""
acq_func = qKnowledgeGradient(
model=self.model, current_value=self.current_best_val
)
# acq_func = qExpectedImprovement(model=self.model, best_f=self.current_best_val)
# acq_func = ExpectedImprovement(model=self.model, best_f=self.current_best_val)
self.next_candidate, self.kg_value = optimize_acqf(
acq_func,
Tensor([[0], [1]]).repeat(1, self.dim),
q=1,
num_restarts=self.num_restarts,
raw_samples=self.raw_samples,
)
def optimize_acqf_and_get_observation(acq_func):
"""Optimizes the acquisition function,
and returns a new candidate and a noisy observation"""
candidates, _ = optimize_acqf(
acq_function=acq_func,
bounds=torch.stack([
torch.zeros(d, dtype=dt),
torch.ones(d, dtype=dt),
]),
q=1,
num_restarts=10,
raw_samples=200,
)
x = unnormalize(candidates.detach(), bounds=bounds)
print('Hyper-parameter: ' + str(x))
obj = self.train(x.view(-1)).unsqueeze(-1)
print(print('Error: ' + str(obj)))
return x, obj
def optimize_acqf_and_get_observation(acq_func):
"""Optimizes the acquisition function, and returns a new candidate and a noisy observation."""
# optimize
candidates, _ = optimize_acqf(
acq_function=acq_func,
bounds=bounds,
q=BATCH_SIZE,
num_restarts=10,
raw_samples=500, # used for intialization heuristic
options={
"batch_limit": 5,
"max_iter": 200,
})
# observe new values
new_x = candidates.detach()
exact_obj = neg_hartmann6(new_x).unsqueeze(-1) # add output dimension
exact_con = outcome_constraint(new_x).unsqueeze(-1) # add output dimension
new_obj = exact_obj + NOISE_SE * torch.randn_like(exact_obj)
new_con = exact_con + NOISE_SE * torch.randn_like(exact_con)
return new_x, new_obj, new_con
def argmax_posterior_mean(cands: to.Tensor, cands_values: to.Tensor,
uc_normalizer: UnitCubeProjector,
num_restarts: int,
num_samples: int) -> to.Tensor:
"""
Compute the GP input with the maximal posterior mean.
:param cands: candidates a.k.a. x
:param cands_values: observed values a.k.a. y
:param uc_normalizer: unit cube normalizer used during the experiments (can be recovered form the bounds)
:param num_restarts: number of restarts for the optimization of the acquisition function
:param num_samples: number of samples for the optimization of the acquisition function
:return: un-normalized candidate with maximum posterior value a.k.a. x
"""
# Normalize the input data and standardize the output data
cands_norm = uc_normalizer.project_to(cands)
cands_values_stdized = standardize(cands_values)
# Create and fit the GP model
gp = SingleTaskGP(cands_norm, cands_values_stdized)
gp.likelihood.noise_covar.register_constraint('raw_noise',
GreaterThan(1e-5))
mll = ExactMarginalLogLikelihood(gp.likelihood, gp)
fit_gpytorch_model(mll)
# Find position with maximal posterior mean
cand_norm, acq_value = optimize_acqf(
acq_function=PosteriorMean(gp),
bounds=to.stack([
to.zeros_like(uc_normalizer.bound_lo),
to.ones_like(uc_normalizer.bound_up)
]),
q=1,
num_restarts=num_restarts,
raw_samples=num_samples)
cand = uc_normalizer.project_back(cand_norm.detach())
print_cbt(f'Converged to argmax of the posterior mean\n{cand.numpy()}',
'g',
bright=True)
return cand
def optimize_acqf_and_get_observation(acq_func):
"""
Optimizes the acquisition function, and returns a new candidate and a noisy observation.
:param acq_func: Acquisition function to use.
:return: new_x: new candidate
new_train_obj: noisy observation of g(x)
"""
# optimize
candidates, _ = optimize_acqf(
acq_function=acq_func,
bounds=bounds,
q=BATCH_SIZE,
num_restarts=10,
raw_samples=512, # used for initialization heuristic
options={"batch_limit": 5, "maxiter": 200},
)
# observe new values
new_x = candidates.detach()
exact_obj, _, _ = exact_sin_objective(new_x, plot_sample=None, fplot=False)
new_train_obj = train_sin_objective(new_x, fplot=False)
return new_x, new_train_obj