sequince

GitHub | PyPI | Documentation | Paper | DOI

Sequince is a modular Sequential Monte Carlo (SMC) library for Bayesian inference, particle filtering, and related sequential methods.

I am the lead developer of sequince.

Features

  • Array API support for NumPy, JAX, PyTorch, and CuPy backends.
  • Composable SMC pipelines built from reusable weighting, resampling, mutation, and control steps.
  • Prebuilt pipelines with make_standard_bayesian_smc and other recipes.
  • Run analysis via SMCResult, including posterior resampling and diagnostics helpers.
  • Bayesian evidence estimation alongside posterior sampling.
  • Checkpointing and resumable sampling runs.

See the documentation to get started or read the paper describing the motivation and design philosophy behind the code.

  • sequince-density: Density estimators and fitted proposal distributions
  • sequince-bilby: Interface for using sequince samplers via bilby
  • orng: Random number generation across array backends
  • acai: Backend-independent control flow and iteration