LoginSign Up
← Back to all sparks
B

butcher

ANALYTICS
Velocity0.0

Reduce the memory footprint of fitted R model objects before deployment

github.com/tidymodels/butcher ↗

butcher expands from trimming models to trimming whole tidymodels workflows

tidymodelsmodel-serializationmemoryr-statsmaintenance
◆Current state
butcher strips the parts of fitted R model objects that bloat serialized size without being needed for prediction, and it grows one method family at a time. The 0.3.x line piled up coverage for MASS, klaR, mixOmics, ipred, survival and xgboost objects; 0.4.0 changes the target from individual models to resampling and tuning containers, and hands maintenance to Max Kuhn.
◆Where it's heading
Coverage is following where tidymodels users actually accumulate size — rset, rsplit, tune_results and workflow_set objects are the things that get large in a tuning run, not single fits. The maintainer handoff points the package further into the tidymodels orbit rather than the broad model zoo it started as. Releases are otherwise small and fix-driven.
◆Prediction
Expect further methods for tidymodels container classes and continued upkeep against xgboost and torch-backed engines; the entries show no move toward automatic size reporting or a different API.

◆Recent moves

  1. 8mo ago

    Methods for rset, tune_results and workflow_set; new maintainer

    The clearest directional entry in this window: butcher can now trim resampling and tuning containers, which is where tidymodels runs actually accumulate size. Max Kuhn taking over maintenance ties the package more tightly to the tidymodels release cycle.

    View source ↗
  2. 11mo ago

    kknn methods no longer remove the call

    A single-method correction with no effect outside kknn fits.

    View source ↗
  3. 1y ago

    Component existence check fixed; nestedmodels methods removed

    Internal correctness fix plus removal of a method family added two releases earlier. Nothing changes for typical users.

    View source ↗
  4. 2y ago

    butcher 0.3.4

    A CRAN resubmission for regenerated reference manuals, with no code change.

    View source ↗
  5. 2y ago

    Methods for nestedmodels and mgcv::gam; clearer butchering messages

    Extends coverage to nested models and trims more of mgcv fits, and improves the messaging when butchering cannot help — a recurring source of confusion with survival::coxph.

    View source ↗
  6. 3y ago

    Eight new model families gain butcher methods

    A large coverage sweep across mixOmics, klaR, ipred, MASS, survival, xrf, mda and dbarts. Representative of the pre-0.4.0 strategy: broaden the model zoo rather than move up the stack.

    View source ↗