def proposal(self, current, current_log_pdf, **kwargs):
if current_log_pdf is None:
current_log_pdf = self.target_log_pdf(current)
forward_drift = self._compute_drift(current)
forward_mu = current + forward_drift
forward_drift_norm = np.linalg.norm(forward_drift)
logger.debug("Norm of forward drift: %.3f" % forward_drift_norm)
self.forward_drift_norms += [forward_drift_norm]
proposal = sample_gaussian(N=1,
mu=forward_mu,
Sigma=self.L_C,
is_cholesky=True,
cov_scaling=self.step_size)[0]
forward_log_prob = self.proposal_log_pdf(current,
proposal[np.newaxis, :],
drift=forward_drift)[0]
backward_drift = self._compute_drift(proposal)
backward_log_prob = self.proposal_log_pdf(proposal,
current[np.newaxis, :],
drift=backward_drift)[0]
proposal_log_pdf = self.target_log_pdf(proposal)
result_kwargs = {}
return proposal, proposal_log_pdf, current_log_pdf, forward_log_prob, backward_log_prob, result_kwargs
def proposal(self, current):
"""
Returns a sample from the proposal centred at current, and its log-probability
"""
if self.schedule is None and (self.mu is None or self.C is None):
raise ValueError("Kameleon has not seen data yet." \
"Either call set_batch_covariance() or set update schedule")
L_R = self._construct_proposal_covariance(current)
proposal = sample_gaussian(N=1,
mu=current,
Sigma=L_R,
is_cholesky=True)[0]
proposal_log_prob = log_gaussian_pdf(proposal,
current,
L_R,
is_cholesky=True)
# probability of proposing current when would be sitting at proposal
L_R_inv = self._construct_proposal_covariance(proposal)
proopsal_log_prob_inv = log_gaussian_pdf(current,
proposal,
L_R_inv,
is_cholesky=True)
return proposal, proposal_log_prob, proopsal_log_prob_inv
def proposal(self, current, current_log_pdf, **kwargs):
if self.Z is None and self.schedule is None:
raise ValueError("%s has not seen data yet. Call set_batch()" %
self.__class__.__name__)
if current_log_pdf is None:
current_log_pdf = self.target_log_pdf(current)
L_R = self._construct_proposal_covariance(current)
proposal = sample_gaussian(N=1,
mu=current,
Sigma=L_R,
is_cholesky=True)[0]
proposal_log_prob = log_gaussian_pdf(proposal,
current,
L_R,
is_cholesky=True)
proposal_log_pdf = self.target_log_pdf(proposal)
# probability of proposing y when would be sitting at proposal
L_R_inv = self._construct_proposal_covariance(proposal)
proopsal_log_prob_inv = log_gaussian_pdf(current,
proposal,
L_R_inv,
is_cholesky=True)
log_acc_prob = proposal_log_pdf - current_log_pdf + proopsal_log_prob_inv - proposal_log_prob
log_acc_prob = np.min([0, log_acc_prob])
results_kwargs = {}
return proposal, proposal_log_pdf, current_log_pdf, proposal_log_prob, proopsal_log_prob_inv, results_kwargs
def proposal(self, current, current_log_pdf, **kwargs):
if current_log_pdf is None:
current_log_pdf = self.target_log_pdf(current)
forward_drift = self._compute_drift(current)
forward_mu = current + forward_drift
forward_drift_norm = np.linalg.norm(forward_drift)
logger.debug("Norm of forward drift: %.3f" % forward_drift_norm)
self.forward_drift_norms += [forward_drift_norm]
proposal = sample_gaussian(N=1, mu=forward_mu, Sigma=self.L_C,
is_cholesky=True, cov_scaling=self.step_size)[0]
forward_log_prob = self.proposal_log_pdf(current, proposal[np.newaxis, :],
drift=forward_drift)[0]
backward_drift = self._compute_drift(proposal)
backward_log_prob = self.proposal_log_pdf(proposal, current[np.newaxis, :],
drift=backward_drift)[0]
proposal_log_pdf = self.target_log_pdf(proposal)
result_kwargs = {}
return proposal, proposal_log_pdf, current_log_pdf, forward_log_prob, backward_log_prob, result_kwargs
def proposal(self, current, current_log_pdf, **kwargs):
if current_log_pdf is None:
current_log_pdf = self.target_log_pdf(current)
proposal = sample_gaussian(N=1, mu=current, Sigma=self.L_C,
is_cholesky=True, cov_scaling=self.step_size)[0]
forw_backw_log_prob = self.proposal_log_pdf(current, proposal[np.newaxis, :])[0]
proposal_log_pdf = self.target_log_pdf(proposal)
results_kwargs = {}
# probability of proposing current when would be sitting at proposal is symmetric
return proposal, proposal_log_pdf, current_log_pdf, forw_backw_log_prob, forw_backw_log_prob, results_kwargs
def proposal(self, current):
"""
Returns a sample from the proposal centred at current, and its log-probability
"""
if self.schedule is None and (self.mu is None or self.C is None):
raise ValueError("Kameleon has not seen data yet." \
"Either call set_batch_covariance() or set update schedule")
L_R = self._construct_proposal_covariance(current)
proposal = sample_gaussian(N=1, mu=current, Sigma=L_R, is_cholesky=True)[0]
proposal_log_prob = log_gaussian_pdf(proposal, current, L_R, is_cholesky=True)
# probability of proposing current when would be sitting at proposal
L_R_inv = self._construct_proposal_covariance(proposal)
proopsal_log_prob_inv = log_gaussian_pdf(current, proposal, L_R_inv, is_cholesky=True)
return proposal, proposal_log_prob, proopsal_log_prob_inv
def proposal(self, current, current_log_pdf, **kwargs):
if current_log_pdf is None:
current_log_pdf = self.target_log_pdf(current)
proposal = sample_gaussian(N=1,
mu=current,
Sigma=self.L_C,
is_cholesky=True,
cov_scaling=self.step_size)[0]
forw_backw_log_prob = self.proposal_log_pdf(current,
proposal[np.newaxis, :])[0]
proposal_log_pdf = self.target_log_pdf(proposal)
results_kwargs = {}
# probability of proposing current when would be sitting at proposal is symmetric
return proposal, proposal_log_pdf, current_log_pdf, forw_backw_log_prob, forw_backw_log_prob, results_kwargs
def proposal(self, current, current_log_pdf, **kwargs):
if self.Z is None and self.schedule is None:
raise ValueError("%s has not seen data yet. Call set_batch()" % self.__class__.__name__)
if current_log_pdf is None:
current_log_pdf = self.target_log_pdf(current)
L_R = self._construct_proposal_covariance(current)
proposal = sample_gaussian(N=1, mu=current, Sigma=L_R, is_cholesky=True)[0]
proposal_log_prob = log_gaussian_pdf(proposal, current, L_R, is_cholesky=True)
proposal_log_pdf = self.target_log_pdf(proposal)
# probability of proposing y when would be sitting at proposal
L_R_inv = self._construct_proposal_covariance(proposal)
proopsal_log_prob_inv = log_gaussian_pdf(current, proposal, L_R_inv, is_cholesky=True)
log_acc_prob = proposal_log_pdf - current_log_pdf + proopsal_log_prob_inv - proposal_log_prob
log_acc_prob = np.min([0, log_acc_prob])
results_kwargs = {}
return proposal, proposal_log_pdf, current_log_pdf, proposal_log_prob, proopsal_log_prob_inv, results_kwargs
# proposals centred at those points
Ys = np.array([[-20, 9.7], [-10, 0], [0,-3], [10, 0], [20, 9.7]])
for j in range(len(colors)):
# pick point at random, embed, gradient
y = Ys[j]
phi_y = feature_map_single(y, omega, u)
grad_phi_y = feature_map_grad_single(y, omega, u)
# draw a number of proposals at the current point
n_proposals = 100
X_star = np.zeros((n_proposals, D))
# generate proposal samples in feature space
for i in range(n_proposals):
plt.plot(y[0], y[1], '*', markersize=15, color=colors[j])
# construct covariance, adding exploration noise
R = eta**2 * np.dot(grad_phi_y, np.dot(C, grad_phi_y.T))
L_R = np.linalg.cholesky(R)
# sample proposal
x_star = sample_gaussian(N=1, mu=y, Sigma=L_R, is_cholesky=True)
X_star[i] = x_star
plt.plot(X_star[:, 0], X_star[:, 1], 'x', color=colors[j])
plt.show()
