Sampler#

jeanspy.sampler.Sampler

See methods and properties for individual lookup pages, or the alphabetical API dictionary to search all classes. The full class contract and existing member anchors are retained below.

class jeanspy.sampler.Sampler(model, p0_generator, nwalkers=None, prefix='', reset=False, pool=None, wbic=False, **kwargs)[source]#

Bases: object

wrapper class for emcee.EnsembleSampler

Notes

Inputs and units. model supplies posterior/blobs and parameter names; p0_generator provides starting positions; nwalkers is an ensemble size; prefix sets output filename; reset=True deliberately resets the requested store; pool controls parallel evaluation. run_mcmc receives the run length/options in its real signature.

Returns and shape. HDF5 chain and diagnostic blobs; get_chain returns (draw,walker,parameter), or flattened draws. get_log_prob/get_blobs use corresponding draw/walker axes. get_dataframe returns a pandas table.

Validity. Workers share a stateful model only through the supported process/storage setup. Use independent ensembles for R-hat; interacting walkers are not independent chains.

Errors. Incompatible persisted analysis identity raises before continuing. Missing/malformed state, invalid parameter conversion or storage failures raise.

Backend. Python/emcee host sampler around NumPy/SciPy CPU likelihoods.

Differentiation. No physical-parameter automatic differentiation on this API.

Examples. examples/docs_inference.py; complete tutorials for production diagnostics.

check_parameter_conversion(p0_generator=None)[source]#: check the conversion of parameters. This is useful to check if the parameters are properly converted from p0 to params. It will print the p0 and params and their comparison.

set_wrapper_function()[source]#

Install the current log-probability callable in this worker process.

Returns None. The module-level wrapper lets multiprocessing workers evaluate their initialized model without repeatedly serializing it. This is host-side worker initialization, not a numerical solver.

reset_pool(pool)[source]#: reset the pool.

property filename#: Return the HDF5 backend’s filename.

burn_in(nsteps, p0_generator, **kwargs)[source]#

Advance warmup and continue from its final ensemble.

Warmup draws remain in the backend. Exclude them explicitly with get_chain(discard=…) when analyzing production draws. The chain already samples the posterior and must not be weighted by the posterior density a second time.

run_mcmc(iterations, loops, reset=False, p0_generator=None, enable_convergence_check=True, **kwargs)[source]#

run the sampler. p0 is generated by p0_generator. save and monitor the chain and lnprob using backends. blobs_dtype is obtained by model.

iterations: number of iterations for each loop loops: number of loops reset: discard the stored chain and initialize a new run p0_generator: override the constructor’s initial-state generator

Parameters:: p0_generator (Callable | None)

get_blobs(flat=False, thin=1, discard=0)[source]#: get blobs from the backend.

get_chain(flat=False, thin=1, discard=0)[source]#: get chain from the backend.

get_log_prob(flat=False, thin=1, discard=0)[source]#: get log_prob from the backend.

get_last_sample()[source]#: get the last sample from the backend.

get_dataframe(thin=1, discard=0, with_lnprob=True)[source]#: get the dataframe from the backend.