sequince.mcmc

Classes

MCMCResult	Result of applying one or more checkpointable MCMC mutation blocks.
MarkovChainMonteCarlo	Checkpointable standalone MCMC built from a Sequince mutation step.

Functions

make_mcmc(log_prob_fn, *[, mutation, n_steps, ...])

Construct standalone MCMC using an existing Sequince mutation backend.

Module Contents

class sequince.mcmc.MCMCResult

Result of applying one or more checkpointable MCMC mutation blocks.

states: collections.abc.Sequence[sequince.state.ParticleState]

context: sequince.steps.base.StepContext

outputs: collections.abc.Sequence[tuple[Any, ...]]

classmethod from_smc_result(result)

Build an MCMC result from a mutation-only sequential run.

Parameters:

result (sequince.result.SMCResult)

Return type:

MCMCResult

encode_for_hdf5(*, strict=False)

Return the logical HDF5 payload for this MCMC result.

Parameters:

strict (bool)

Return type:

dict[str, Any]

save(h5_file_or_group, path='mcmc_result', *, strict=False)

Save this MCMC result below path in an open HDF5 handle.

Parameters:

• h5_file_or_group (Any)

• path (str)

• strict (bool)

Return type:

None

classmethod load(h5_file_or_group, path='mcmc_result')

Load an MCMC result from path in an open HDF5 handle.

Parameters:

• h5_file_or_group (Any)

• path (str)

Return type:

Self

property final_state: sequince.state.ParticleState

Return type:

sequince.state.ParticleState

property chain: sequince.typing.Array

Return mutation blocks concatenated as (steps, chains, dims).

Return type:

sequince.typing.Array

property diagnostics: dict[str, Any]

Return diagnostics from the final mutation output.

Return type:

dict[str, Any]

samples(*, burn_in=0, thin=1, flat=True)

Return the chain after burn-in and thinning.

Parameters:

• burn_in (int)

• thin (int)

• flat (bool)

Return type:

sequince.typing.Array

class sequince.mcmc.MarkovChainMonteCarlo(mutation, *, callbacks=None, checkpoint_writers=None)

Checkpointable standalone MCMC built from a Sequince mutation step.

Parameters:

• mutation (sequince.steps.mutation.MutationStep)

• callbacks (collections.abc.Sequence[Callable[..., None]] | None)

• checkpoint_writers (collections.abc.Sequence[sequince.checkpoint.CheckpointWriter] | None)

mutation

run(initial_state=None, *, rng=None, shared_context=None, metadata=None, outputs=None, runtime=None)

Run one configured MCMC block.

Parameters:

• initial_state (sequince.state.ParticleState | None)

• rng (orng.RandomGenerator | numpy.random.Generator | int | None)

• shared_context (collections.abc.Mapping[str, Any] | None)

• metadata (dict[str, Any] | None)

• outputs (collections.abc.Sequence[tuple[Any, ...]] | None)

• runtime (sequince.smc.SMCRuntime | None)

Return type:

MCMCResult

resume_from(checkpoint, *, rng=None, shared_context=None)

Restore a checkpoint for continuation with run().

Parameters:

• checkpoint (sequince.checkpoint.SMCCheckpoint | collections.abc.Mapping[str, Any])

• rng (orng.RandomGenerator | numpy.random.Generator | int | None)

• shared_context (collections.abc.Mapping[str, Any] | None)

Return type:

sequince.smc.SMCRuntime

sequince.mcmc.make_mcmc(log_prob_fn, *, mutation='minipcn', n_steps=100, mutation_kwargs=None, particle_transform=None, callbacks=None, checkpoint_writers=None)

Construct standalone MCMC using an existing Sequince mutation backend.

Parameters:

• log_prob_fn (Callable[..., sequince.typing.Array | float])

• mutation (str | sequince.steps.mutation.MutationStep)

• n_steps (int)

• mutation_kwargs (dict[str, Any] | None)

• particle_transform (sequince.transforms.ParticleTransform | None)

• callbacks (collections.abc.Sequence[Callable[..., None]] | None)

• checkpoint_writers (collections.abc.Sequence[sequince.checkpoint.CheckpointWriter] | None)

Return type:

MarkovChainMonteCarlo