LoginSign Up
← Back to all sparks
P

posterior

ANALYTICS
Velocity0.0

Tools for working with posterior draws and computing convergence diagnostics

github.com/stan-dev/posterior ↗

posterior keeps deepening two things: the rvar type and Pareto-based diagnostics.

bayesianrvarpareto-diagnosticsr-infrastructuredraws-formats
◆Current state
The releases in this window advance on two fronts. The rvar random-variable type gained factor and ordered subtypes (1.4.0), rvar-indexed slicing and `rvar_ifelse()` (1.5.0), base `%*%` matrix multiplication and indexed variable names (1.6.0). Separately, Pareto diagnostics have grown from `pareto_smooth()` options and individual `pareto_khat()`-family functions (1.6.0) through `pit()` for draws and rvars (1.6.1) to exported generalized-Pareto functions and `pareto_pit` (1.7.0). 1.7.1 is a paperwork release for a JOSS submission.
◆Where it's heading
posterior is positioning itself as shared infrastructure rather than an end-user package: 1.7.0 explicitly exports generalized-Pareto machinery 'for use in other packages', and the JOSS paper is a citation vehicle for the same audience. The rvar work points the same way — a random-variable type other Bayesian packages can build on. Cadence is steady but unhurried, roughly one feature release a year.
◆Prediction
More diagnostic functions are likely to be exported for downstream reuse, following the pattern 1.7.0 established with the generalized-Pareto helpers.

◆Recent moves

  1. 2mo ago

    posterior 1.7.1 released for JOSS paper

    A minor release cut to accommodate the JOSS paper. No functional change, but it marks the package formalising its role as citable infrastructure.

    View source ↗
  2. 3mo ago

    posterior 1.7.0 exports generalized-Pareto functions

    Exports generalized-Pareto functions explicitly for other packages to use and adds Pareto PIT via `pareto_pit`. The clearest statement in the window that posterior is building for downstream consumers.

    View source ↗
  3. 10mo ago

    posterior 1.6.1 adds pit() for draws and rvars

    Implements `pit()` for draws and rvar objects, enabling LOO-PIT via weights, and converts lists of matrices into `draws_array`. Extends the diagnostic surface that 1.7.0 then exports.

    View source ↗
  4. 1y ago

    posterior 1.6.0 adds Pareto diagnostics and ESS-based thinning

    The densest release in the window: individual Pareto diagnostic functions, ESS-based automatic thinning, base `%*%` for rvars, indexed variable names, and `extract_variable_array()`. Both of the package's long-running threads advance at once.

    View source ↗
  5. 2y ago

    posterior 1.5.0 adds nested-Rhat and rvar indexing

    Adds nested-Rhat and lets rvars be indexed by other rvars, plus `rvar_ifelse()`. Makes rvar behave more like a native R type rather than a container you unpack.

    View source ↗
  6. 3y ago

    posterior 1.4.0 adds factor and ordered rvar subtypes

    Introduces `rvar_factor()` and `rvar_ordered()` so discrete posteriors are first-class, with `modal_category()`, `entropy()` and `dissent()` to summarise them. The rvar type stops being numeric-only here.

    View source ↗