Advanced workflows#
These guides cover running mutation algorithms independently, compiling SMC iterations with JAX, and continuing a run from a saved checkpoint. Each builds on the configuration and result interfaces introduced in Running inference.
Standalone MCMC shows how to use mutation steps for posterior sampling with a standalone MCMC runner, then extract chains and diagnostics.
JAX and functional execution explains functional execution with explicit random keys, how to compile an iteration, and the requirements on steps, targets, and runtime state.
Checkpointing covers checkpoint writers, saved state, and resuming execution, including restoring the random-number-generator state.
For JAX workflows, first configure your array backend and review Random number generation. To select mutation and resampling algorithms, see Sampling options.