bayesmith.dispatch.proposal_sampling

Source-derived signatures and contracts. See the reference index for coverage and conventions.

Module contract

NumPyro chains for ordered proposal sweeps, with an optional NUTS remainder.

The small HMCGibbs adapter isolates NumPyro's state contract here. Proposal
counters are state fields, so diagnostics and both RNG streams survive every
checkpoint along with the inner potential and adaptation state.

ProposalState ProposalGibbsState ProposalKernel ProposalGibbs run_proposal_mcmc

ProposalState

Defined in bayesmith.dispatch.proposal_sampling, line 38.

class ProposalState(NamedTuple)
z: dict[str, Any]
rng_key: jax.Array
proposal_diagnostics: ProposalDiagnostics
diverging: jax.Array

No additional source docstring. Read the signature and owning module contract above.

ProposalGibbsState

Defined in bayesmith.dispatch.proposal_sampling, line 45.

class ProposalGibbsState(NamedTuple)
z: dict[str, Any]
hmc_state: Any
rng_key: jax.Array
proposal_diagnostics: ProposalDiagnostics

No additional source docstring. Read the signature and owning module contract above.

ProposalKernel

Defined in bayesmith.dispatch.proposal_sampling, line 52.

ProposalKernel(compiled)

A whole-model proposal chain; correction follows each block's policy.

ProposalKernel.init

init(self, rng_key, num_warmup, init_params, model_args, model_kwargs)

ProposalKernel.sample

sample(self, state, model_args, model_kwargs)

sample_field = 'z'
default_fields = ('z', 'proposal_diagnostics', 'diverging')

ProposalGibbs

Defined in bayesmith.dispatch.proposal_sampling, line 90.

ProposalGibbs(inner_kernel, compiled)

HMCGibbs with actual MH counters and support-aware coordinate refresh.

ProposalGibbs.init

init(self, rng_key, num_warmup, init_params, model_args, model_kwargs)

ProposalGibbs.sample

sample(self, state, model_args, model_kwargs)

default_fields = ('z', 'proposal_diagnostics', 'hmc_state.diverging')

run_proposal_mcmc

Defined in bayesmith.dispatch.proposal_sampling, line 314.

run_proposal_mcmc(graph, compiled, key, *, num_warmup, num_samples, num_chains, chain_method, nuts_options, progress_bar, control=None)

Execute explicit proposal blocks and return a real chain Posterior.