package prbnmcn-dagger-stats

0.0.5

• lmh: acceptance statistics for LMH and incremental LMH samplers

• [Smc] uses generative samplers (type [Random.State.t -> 'a]) instead of type ['a Dist.t]

• adapt to latest version of [prbnmcn-stats]

• remove [samplei] function

• smc: make [Smc.Make] functor generative (report by @nilsbecker)

• smc: improve error message (report by @nilsbecker)

0.0.4

• smc: domainslib-based multicore implementation

• bugfix: Smc_inference.run_custom spawned too many particles

• fix implicit deps (@zapashcanon)

0.0.3

API changes

• smc: provide API for programmable resampling (review by @nilsbecker, many thanks!)

• gsl_dist: functorize over GSL signature

  • band-aid for unavailability of gsl package in ocaml 5

  • since explicit dependency on gsl is dropped, relicensed to MIT

code quality/bug fixes

• smc: much improved documentation (report and review by @nilsbecker)

• removed uses of Obj.magic

• bugfix: properly handle terminated particles in smc

  • terminated particles now continue to participate in resampling events

• bugfix: Smc.fork off-by-one corrected

• bugfix: Cps_monad.map_array reversed order of elements

misc

• added gamma distribution to stats_dist

• added kalman filter example

• added polynomial regression example

• added approximate bayesian computation example

• more tests

0.0.2

• Dependency: prbnmcn-stats.0.0.3 -> prbnmcn-stats.0.0.4

• Add beta distribution to Gsl samplers

• Refactor Cps monad

• Add SMC inference

• Simplify handler type, modularize effect definitions away from Cps_monad

• Fix typo: bernouilli -> bernoulli (report by @nilsbecker)

0.0.1

• First release of prbnmcn-dagger, a library for probabilistic programming