LoginSign Up
← Back to all sparks
G

gratia

ANALYTICS
Velocity0.0

R package gratia by gavinsimpson — release notes from GitHub.

github.com/gavinsimpson/gratia ↗

The tidy front-end for GAMs, now stable enough that upstream ggplot2 sets its release calendar.

gammgcvggplot2statistical-graphicsmodel-diagnosticsr-package
◆Current state
gratia wraps mgcv-fitted generalized additive models in tidy data frames and ggplot2 graphics — smooth_estimates(), fitted_values(), derivatives(), draw() and appraise() cover evaluation, prediction and diagnostics. The API reached its intended shape at 0.9.0, when every generated column was renamed to a dot-prefixed form, and 0.10.0 added conditional_values() for covariate-conditional prediction plots.
◆Where it's heading
The package has moved through a long rewrite cycle and out the other side. Successive releases replaced evaluate_smooth() with smooth_estimates(), rebuilt draw() on top of it, then renamed the entire output vocabulary to avoid colliding with user variables. That work is finished; 0.11.1 is driven almost entirely by ggplot2 4.0.0 compatibility, with new mgcv family support for quantile residuals riding along. Development now tracks upstream breakage rather than internal redesign.
◆Prediction
Expect the next releases to continue absorbing ggplot2 4.x and mgcv changes, with incremental family coverage in quantile_residuals() as the visible new work. The entries give no signal on which mgcv families come next.

◆Recent moves

  1. 11mo ago

    ggplot2 4.0.0 compatibility, plus four more quantile-residual families

    Primarily a compatibility release for ggplot2 4.0.0, which forced a fix to residual histogram binning so the middle bin centres at zero. quantile_residuals() picks up mgcv's scat, nb, betar and tw families, and partial_derivatives() now handles multiple smooths and skips random-effect and univariate terms instead of erroring.

    View source ↗
  2. 1y ago

    conditional_values() replaces vis.gam for conditional prediction plots

    conditional_values() lands as a tidy replacement for mgcv::vis.gam(), plotting predictions conditional on chosen covariates in the marginaleffects idiom. penalty() and basis() also gain reparameterization that exposes the penalty null space directly.

    View source ↗
  3. 2y ago

    parametric_effects() joins the dot-prefix rename; LSS families begin

    Catches parametric_effects() up to the 0.9.0 renaming, relabels random-effect plots by smooth label rather than variable name, and begins location-scale-shape family support in fitted_samples(). Cleanup work finishing what the previous release started.

    View source ↗
  4. 2y ago

    Every generated column gains a dot prefix

    ⚡ SPARK

    The release that fixed gratia's output vocabulary: every generated column across smooth_estimates(), fitted_values(), derivatives() and their siblings gained a dot prefix, and derivatives() switched to storing evaluation points under their real variable names. It is the boundary every downstream script written against gratia has to cross.

    View source ↗
  5. 3y ago

    Real variable names in smooth_samples(), plus dplyr 1.1.0 fixes

    Absorbs dplyr 1.1.0 changes and fixes smooth_samples() to return real variable names instead of .x1, .x2 placeholders — a precursor to the wholesale renaming at 0.9.0. draw.gam() also learns tensor products with a marginal random-effect smooth.

    View source ↗
  6. 4y ago

    M1 example output fixes and confint tibble returns

    A patch release driven by example output differing on M1 Macs, with confint methods returning tibbles and smooth plots relabelled to "Partial effect". Reference-output housekeeping rather than capability.

    View source ↗