LoginSign Up
← Back to all sparks
R

rstanarm

ANALYTICS
Velocity0.0

Bayesian applied regression modeling in R with standard formula syntax via Stan

github.com/stan-dev/rstanarm ↗

rstanarm is community-maintained now, tracking Stan and lme4 rather than adding models.

bayesianstanregression-modelsdependency-migrationcommunity-maintenance
◆Current state
2.32.2 is entirely infrastructure and dependency work: formula machinery migrated from lme4 to reformulas, `r_eff` no longer computed for loo by default, the Stan R packages repo replaced by R-Universe, rstantools adopted to fix build and export errors, and CRAN NOTE cleanups — contributed largely by four first-time contributors. 2.32.1 and 2.26.1 follow the same pattern, tracking rstan syntax and adding `posterior::as_draws()` support. The last release with substantive modelling content is 2.21.1, which changed how default priors are determined and flipped `autoscale` to FALSE outside default priors.
◆Where it's heading
The package has moved from feature development into ecosystem maintenance, and the contributor list shows why it survives: outside developers keep it compiling against a moving Stan, lme4 and CRAN. The `as_draws()` support and the reformulas migration both point the same way — rstanarm increasingly consumes shared infrastructure (posterior, reformulas, rstantools) instead of carrying its own.
◆Prediction
Expect the next release to track another upstream change — rstan, reformulas or CRAN policy — rather than add model families. The pre-fit model surface looks settled.

◆Recent moves

  1. 10mo ago

    rstanarm 2.32.2 migrates formula machinery to reformulas

    Migrates formula handling from lme4 to reformulas, stops computing `r_eff` for loo by default, and adopts rstantools to fix build errors — mostly from first-time contributors. Real plumbing work, no new modelling capability.

    View source ↗
  2. 2y ago

    rstanarm 2.32.1 fixes unit_vector error, enables LTO

    Avoids a 1-D unit_vector error, updates `.set_nreps`, and enables LTO during compilation. Small fixes between larger Stan-tracking releases.

    View source ↗
  3. 2y ago

    rstanarm 2.26.1 adopts new rstan syntax and as_draws()

    Updates to current rstan syntax including array syntax, and implements `posterior::as_draws()` for rstanarm objects. The as_draws support is what plugs rstanarm into the shared draws tooling the wider Stan ecosystem has standardised on.

    View source ↗
  4. 4y ago

    rstanarm 2.21.3 fixes loo() and adds stan_jm offsets

    A long list of `loo()`, `stan_aov()`, `stan_glmer()` and prediction fixes, plus an offset term for `stan_jm()` longitudinal submodels and K-fold effective parameter counts. The last release in the window with modelling-level additions.

    View source ↗
  5. 6y ago

    rstanarm 2.21.1 changes default prior behaviour

    Changes how default intercept priors are located and scaled and flips `autoscale` to FALSE outside default priors, aligning the package with Regression and Other Stories. Existing code keeps running but fitted results can shift.

    View source ↗