Sampling options#
Resampling and mutation control how a particle population is carried between SMC targets. Resampling draws from the weighted population and assigns equal weights to the selected particles. Mutation applies MCMC transitions that explore the current target while preserving the particle weights.
The pre-built samplers expose these choices through their resampling and
mutation arguments. For a custom pipeline, configure the corresponding
steps directly. See Pre-built samplers for the
available sampler designs.
Resampling configuration explains systematic, stratified, and multinomial resampling, when to trigger resampling, and how to inspect ancestry and ESS diagnostics.
Mutation configuration covers mutation algorithms, transition counts, proposal kernels, particle transforms, and mutation diagnostics.
Choose options compatible with your array backend and execution mode; see Array backends and JAX and functional execution. If you use genealogy-based evidence uncertainty estimates, check their resampling assumptions in Evidence and uncertainty before choosing a scheme.