LoginSign Up
← Back to all sparks
I

insight

ANALYTICS
Velocity0.0

Uniform access to information from any statistical model object in R

github.com/easystats/insight ↗

insight quietly widens the set of model objects the easystats ecosystem can read

model-introspectioneasystatsbayesianperformancer-stats
◆Current state
insight is the extraction layer under the easystats packages — it answers what a fitted model's parameters, data, variance and priors are, for whatever object it is handed. The 1.4.x and 1.5.x releases read as a steady widening of that support list: tidymodels workflows, cmdstanr fits, rstpm2 survival models, lavaan variance-covariance, mice imputations.
◆Where it's heading
Two things move together here. The support list grows toward objects produced outside the easystats world, and performance work targets the helpers that everything else calls — compact_list(), is_empty_object(), find_parameters() on mgcv models. New functions appear occasionally (get_simulated(), vcovFPC()) but the center of gravity is coverage, not capability.
◆Prediction
Expect further model classes to be added as downstream easystats packages need them, and continued alignment with R-devel behavior changes like the weighted-residuals revision.

◆Recent moves

  1. 1mo ago

    compact_list() performance and lavaan variance-covariance support

    Speeds up a helper that many other easystats functions call, and adds explicit get_varcov() support for lavaan objects alongside three brms and rstanarm fixes.

    View source ↗
  2. 2mo ago

    cmdstanr support and finite-population-corrected variance

    CmdStanFit models become readable by find_parameters() and friends — the same backend expansion bayestestR builds on — and vcovFPC() adds finite population correction. format_table() also learns to print tail and bulk ESS.

    View source ↗
  3. 4mo ago

    get_simulated() added; rstpm2 survival models supported

    A new function for simulated response values comparable to posterior predictions, plus support for flexible parametric survival models and mgcv performance work on random-effects fits.

    View source ↗
  4. 6mo ago

    Weighted residuals revised to match R 4.6.0

    A breaking change adopted from upstream R rather than chosen here: get_residuals(weighted = TRUE) now follows R 4.6.0's weighted working residuals for glm objects.

    View source ↗
  5. 6mo ago

    tidymodels workflow objects become readable

    find_predictors(), find_variables() and get_data() start working on tidymodels workflows — a bridge between two R modeling ecosystems that mostly ignore each other. Also speeds up a widely-called internal helper.

    View source ↗
  6. 8mo ago

    lme4 convergence and fixest data extraction fixes

    Two compatibility fixes driven by upstream package changes. No new capability.

    View source ↗