beanmachine.ppl.inference.proposer.single_site_random_walk_proposer module
- class beanmachine.ppl.inference.proposer.single_site_random_walk_proposer.SingleSiteRandomWalkProposer(node, step_size: float)
Bases:
beanmachine.ppl.inference.proposer.single_site_ancestral_proposer.SingleSiteAncestralProposer- beta_dist_from_moments(mu, sigma)
Returns a Beta distribution.
- Parameters
mu – mu value
sigma – sigma value
- Returns
returns the Beta distribution given mu and sigma.
- dirichlet_dist_from_moments(mu, sigma)
Returns a Dirichlet distribution. The variances of a Dirichlet distribution are inversely proportional to the norm of the concentration vector. However, variance is only set as a scalar, not as a vector. So the individual variances of the Dirichlet are not tuned, only the magnitude of the entire vector.
- Parameters
mu – mu value
sigma – sigma value
- Returns
returns the Dirichlet distribution given mu and sigma.
- do_adaptation(world, accept_log_prob, *args, **kwargs) None
- gamma_dist_from_moments(expectation, sigma)
Returns a Gamma distribution.
- Parameters
expectation – expectation value
sigma – sigma value
- Returns
returns the Beta distribution given mu and sigma.
- get_proposal_distribution(world: beanmachine.ppl.world.world.World) torch.distributions.distribution.Distribution
Propose a new value for self.node using the prior distribution.