Version 0.4 -- 2023-09-01

Added

• nested: NS-SMC sampler of Salomone et al (2018)
• datasets: Liver
• distributions:
  • LogNormal
  • Mixture
  • FlatNormal
  • mixMissing (to deal with missing data)
  • VaryingCovNormal (issue #55)
• smoothing: FFBS-MCMC, FFBS-hybrid
• collectors: Paris algorithm (hybrid version)
• smc_samplers:
  • single-run variance estimates
  • Tempering (fixed exponents)
  • AdaptiveTempering has a new argument, max_iter, to put a cap on the number of iterations.

Version 0.3 -- 2021-10-25

Added

• new module: binary_smc
• smc_samplers: waste-free SMC (now default)
• resampling: added killing resampling scheme
• new tutorial notebook: how to define complicated state-space models

Changed

• qmc: now based on scipy.stats.qmc (remove Fortran code dependency)

Added

• new module: datasets (cleaner way to load standard datasets)
• new module: variance_estimators (single-run genealogy based estimators à la Lee and Whiteley)

Changed

• collectors: new implementation (breaks compatibility)
• utils: performance improvements (multi-processing, distinct seeds)

The main point of this release is to point to a version that remains Python 2 compatible (and also works for Python 3, up to version 3.7). Forthcoming versions will likely not support Python 2, in order to better support Python 3.8+ versions. (particles used time.clock() to measure the CPU time of a SMC run, but this feature has been removed from 3.8, and replacements are available only since version 3.3).

Unless you really need Python 2 for some reason (in that case I'd be happy to hear why), do not bother about this release, and simply use the latest version of the master branch.