def __init__(self, surrogate, target, momentum, num_steps_min=10, num_steps_max=100, step_size_min=0.05,
step_size_max=0.3, adaptation_schedule=None, acc_star=0.7):
"""
"""
HMCBase.__init__(self, target, momentum, num_steps_min, num_steps_max, step_size_min, step_size_max,
adaptation_schedule, acc_star)
assert_implements_log_pdf_and_grad(surrogate)
self.surrogate = surrogate
self.target = surrogate
self.orig_target = target
def get_hmc_kernel():
num_steps_min, num_steps_max, step_size_min, step_size_max = get_hmc_parameters(
)
target, momentum = get_target_momentum()
hmc = HMCBase(target, momentum, num_steps_min, num_steps_max,
step_size_min, step_size_max)
return hmc
def sample_banana2(N, D, target):
momentum = IsotropicZeroMeanGaussian(D=D, sigma=1.0)
num_steps = 1000
step_size = 0.2
hmc = HMCBase(target, momentum, num_steps, num_steps, step_size, \
step_size, adaptation_schedule=None)
start_samples = sample_banana(N, D)
# simulate trajectory from starting point, note _proposal_trajectory is a "hidden" method
Qs_total = []
for i in xrange(N):
current = start_samples[i]
current_log_pdf = target.log_pdf(current)
Qs, acc_probs, log_pdf_q = hmc._proposal_trajectory(
current, current_log_pdf)
Qs_total.append(Qs[-1])
return np.asarray(Qs_total)
def accept_prob_log_pdf(self, current, q, p0_log_pdf, p_log_pdf, current_log_pdf):
# same as super-class, but with original target
kernel_target = self.target
self.target = self.orig_target
acc_prob, log_pdf_q = HMCBase.accept_prob_log_pdf(self, current, q, p0_log_pdf, p_log_pdf, current_log_pdf)
# restore target
self.target = kernel_target
return acc_prob, log_pdf_q
def __init__(self,
surrogate,
target,
momentum,
num_steps_min=10,
num_steps_max=100,
step_size_min=0.05,
step_size_max=0.3,
adaptation_schedule=None,
acc_star=0.7):
"""
"""
HMCBase.__init__(self, target, momentum, num_steps_min, num_steps_max,
step_size_min, step_size_max, adaptation_schedule,
acc_star)
assert_implements_log_pdf_and_grad(surrogate)
self.surrogate = surrogate
self.target = surrogate
self.orig_target = target
def accept_prob_log_pdf(self, current, q, p0_log_pdf, p_log_pdf,
current_log_pdf):
# same as super-class, but with original target
kernel_target = self.target
self.target = self.orig_target
acc_prob, log_pdf_q = HMCBase.accept_prob_log_pdf(
self, current, q, p0_log_pdf, p_log_pdf, current_log_pdf)
# restore target
self.target = kernel_target
return acc_prob, log_pdf_q
def test_hmc_base_init_execute():
num_steps_min, num_steps_max, step_size_min, step_size_max = get_hmc_parameters(
)
target, momentum = get_target_momentum()
HMCBase(target, momentum, num_steps_min, num_steps_max, step_size_min,
step_size_max)
T = 100
k = 0
while k < num_steps + 1:
# KSD
Qs_batch = Qs_total[:, k, :]
ksd_total.append(compute_ksd(Qs_batch, target))
k += T
ksd_total = np.asarray(ksd_total)
results.append([
Qs_total, acc_probs_total, accor_total, ksd_total, ess_total,
mean_x1_total
])
# now run HMC
momentum = IsotropicZeroMeanGaussian(D=D, sigma=1.0)
hmc = HMCBase(target, momentum, num_steps, num_steps, step_size, \
step_size, adaptation_schedule=None)
# simulate trajectory from starting point, note _proposal_trajectory is a "hidden" method
Qs_total = []
acc_probs_total = []
accor_total = []
ksd_total = []
ess_total = []
mean_x1_total = []
np.random.seed(seed + 1)
for i in xrange(M):
current = start_samples[i]
current_log_pdf = target.log_pdf(current)
Qs, acc_probs, log_pdf_q = hmc._proposal_trajectory(
current, current_log_pdf)
# compute auto correlation on first dim