kernels.harness.SamplerSettings

One escalation stage’s MCMC budget.

Usage

Source

kernels.harness.SamplerSettings(
    chains=4,
    iter_warmup=1000,
    iter_sampling=2500,
    parallel_chains=1,
    target_accept=None,
    max_treedepth=None
)

None means “the entry’s own default” and is not passed at all, so a single default-constructed stage reproduces a plain gallery.fit(). fit_kwargs filters against the entry’s signature, so the same stage list can drive a study that mixes MCMC and likelihood-based entries (the latter simply ignore all of this).

Parameter Attributes

chains: int = 4
iter_warmup: int = 1000
iter_sampling: int = 2500
parallel_chains: int = 1
target_accept: float | None = None
max_treedepth: int | None = None

Methods

Name Description
fit_kwargs() The subset of these settings that fit_fn actually accepts.

fit_kwargs()

The subset of these settings that fit_fn actually accepts.

Usage

Source

fit_kwargs(fit_fn)