User guide

The Sequential Monte Carlo (SMC) loop in sequince is assembled from composable steps. Each step receives the current particle state and runtime context, then returns an updated state, context, and optional output. At the end of an iteration, callbacks can inspect the accumulated result and update the context. This proceeds until a user-specified termination criterion is satisfied, which returns a result object for further processing and results visualisation.

Start with the core concepts to understand the objects passed through the SMC loop. The running-inference guides cover pre-built samplers, configuration, and interpreting outputs. The remaining sections explain sampling choices and more specialised execution and extension points.

• Core concepts
  • Array backends
  • Particle state and results
  • Step context
  • Random number generation
• Running inference
  • Configuring inference
  • Understanding outputs
• Sampling options
  • Resampling configuration
  • Mutation configuration
• Advanced workflows
  • Standalone MCMC
  • JAX and functional execution
  • Checkpointing
• Extending Sequince
  • Custom steps
  • Custom callbacks