def test_nuts(self):
n_samples = 100000
n_adapt = 10000
x_init = np.random.rand(n)
samples = mcmc.nuts(n_samples, x_init, log_pdf, grad_log_pdf, n_adapt=n_adapt)
assert samples.shape == (n_samples, n)
cov = np.cov(samples[n_adapt:, :].T)
assert np.allclose(cov, true_cov, atol=0.1, rtol=0.1)
def acquire(self, n, t=None):
"""Acquire a batch of acquisition points.
Parameters
----------
n : int
Number of acquisitions.
t : int, optional
Current iteration, (unused).
Returns
-------
array_like
Coordinates of the yielded acquisition points.
"""
if n > self._n_samples:
raise ValueError("The number of acquisitions, n, has to be lower"
"than the number of the samples (%d).".format(
self._n_samples))
logger.debug('Acquiring the next batch of %d values', n)
gp = self.model
# Updating the ABC threshold.
self.eps = np.percentile(gp.Y, self.quantile_eps * 100)
def _evaluate_gradient_logpdf(theta):
denominator = self.evaluate(theta)
if denominator == 0:
return -np.inf
pt_eval = self.evaluate_gradient(theta) / denominator
return pt_eval.ravel()
def _evaluate_logpdf(theta):
val_pdf = self.evaluate(theta)
if val_pdf == 0:
return -np.inf
return np.log(val_pdf)
# Obtaining the RandMaxVar acquisition.
for i in range(self._limit_faulty_init + 1):
if i > self._limit_faulty_init:
raise SystemExit("Unable to find a suitable initial point.")
# Proposing the initial point.
theta_init = np.zeros(shape=len(gp.bounds))
for idx_param, range_bound in enumerate(gp.bounds):
theta_init[idx_param] = self.random_state.uniform(
range_bound[0], range_bound[1])
# Refusing to accept a faulty initial point.
if np.isinf(_evaluate_logpdf(theta_init)):
continue
# Sampling the acquisition using the chosen sampler.
if self.name_sampler == 'metropolis':
if self._sigma_proposals_metropolis is None:
# Setting the default values of the sigma proposals to 1/10
# of each parameters interval's length.
sigma_proposals = []
for bound in self.model.bounds:
length_interval = bound[1] - bound[0]
sigma_proposals.append(length_interval / 10)
self._sigma_proposals_metropolis = sigma_proposals
samples = mcmc.metropolis(
self._n_samples,
theta_init,
_evaluate_logpdf,
sigma_proposals=self._sigma_proposals_metropolis,
seed=self.seed)
elif self.name_sampler == 'nuts':
samples = mcmc.nuts(self._n_samples,
theta_init,
_evaluate_logpdf,
_evaluate_gradient_logpdf,
seed=self.seed)
else:
raise ValueError(
"Incompatible sampler. Please check the options in the documentation."
)
# Using the last n points of the MH chain for the acquisition batch.
batch_theta = samples[-n:, :]
break
return batch_theta