gemlib.abc.abc_rejection_sampler

gemlib.abc.abc_rejection_sampler#

gemlib.abc.abc_rejection_sampler(accept_fn, num_samples, model, key)#

Approximate Bayesian Computation rejection sampler

Parameters:

  • accept_fn (Callable[[PyTree], Array]) – Python callable which takes a batch of samples from model and returns a boolean array of shape num_samples to keep. Any tolerance/threshold logic is responsibility of accept_fn.

  • num_samples (int) – total number of samples

  • model (JointDistribution) – a tfd.JointDistribution to draw proposal from

  • key (Array) – a PRNG key

Returns:

A tuple of (posterior, acceptance_rate).

Return type:

tuple[PyTree, float]

Contents