LocScalePTM.run_mcmc()

LocScalePTM.run_mcmc()#

LocScalePTM.run_mcmc(seed, warmup, posterior, num_chains=4, fast_warmup=0.5, thinning_posterior=1, thinning_warmup=1, warm_start=True, which=None, strategy='iwls-nuts', cache_path=None, apply_jitter=False, **initialization_kwargs)[source]#

Run MCMC sampling and return sampling results.

Parameters:

  • seed (int) – MCMC scheduling parameters: seed and durations.

  • warmup (int) – MCMC scheduling parameters: seed and durations.

  • posterior (int) – MCMC scheduling parameters: seed and durations.

  • num_chains (int) – Number of parallel chains to run. (default: 4)

  • strategy (Literal['iwls-nuts', 'iwls_fixed', 'iwls_fixed-nuts', 'nuts', 'iwls-iwls_fixed', 'manual']) – Which kernel strategy to use for sampling. See setup_default_mcmc_kernels(). (default: 'iwls-nuts')

  • cache_path (str | Path | None) – If provided, load/save cached sampling results. (default: None)

  • apply_jitter (bool) – Whether to apply initial jitter to chain initialisations. Only has an effect if jittering is specified in the liesel.goose.MCMCSpec for any one variable. Think of this rather as an off-switch than an on-switch. (default: False)

  • warm_start (bool) – If True, the model will be initialized by finding posterior modes via initialize(). (default: True)

  • **initialization_kwargs – Forwarded to initialize() when warm_start is True.

Return type:

SamplingResults

Returns:

SamplingResults – The liesel.goose.SamplingResults sampling results object containing chains and diagnostics.

Contents