LoginSign Up
← Back to all sparks
O

offsetreg

INFRA · APIS
Velocity0.0

An Extension of Tidymodels Supporting Offset Terms

github.com/mattheaphy/offsetreg ↗

The parsnip extension for exposure models grew from one algorithm to three

tidymodelsparsnipr-packageactuarialoffset-modelsmachine-learning
◆Current state
offsetreg supplies parsnip model specifications for regressions with offsets, the form actuarial and epidemiological work needs when modelling rates over exposure. It launched with a single specification, poisson_reg_offset(), backed by glm and glmnet engines. Version 1.1.0 added two more model types - boost_tree_offset() for boosted trees via xgboost and decision_tree_exposure() for weighted decision trees via rpart - and 1.2.0 has been consolidation: argument passthrough to the underlying glm and glmnet calls, an xgboost 3.0 minimum with API accommodation, and cli-formatted messages.
◆Where it's heading
The package is following the tidymodels extension playbook: establish one model type, then add types rather than engines, and keep pace with parsnip's own releases. Each version has also tightened the guardrails around correct use - a vignette on when offsetreg should and should not be used, check_args() methods on the specifications - which suggests the maintainer is fielding misapplication rather than feature requests.
◆Prediction
Expect additional engines under the existing model types, or a fourth specification, before any change to the offset handling itself, since the package's structure invites extension at the engine layer.

◆Recent moves

  1. 4mo ago

    Argument passthrough to glm and glmnet, xgboost 3.0 minimum

    Lets additional named arguments reach stats::glm() and glmnet::glmnet() through the offset wrappers, which removes a common reason to drop out of the parsnip interface. The xgboost 3.0 requirement and its API accommodation are the maintenance cost of having taken on a boosted-tree engine in the previous release.

    View source ↗
  2. 1y ago

    Boosted trees and weighted decision trees join the offset model family

    ⚡ SPARK

    The release note bundles a routine parsnip 1.3.0 compatibility bump with the 1.1.0 notes, where the real content sits. Going from one model specification to three is what turned offsetreg from a glm wrapper into an extension package with a family of methods behind a common interface.

    View source ↗
  3. 2y ago

    Initial release with Poisson regression over offsets

    The founding release: one model specification, poisson_reg_offset(), wrapping stats::glm() and glmnet::glmnet() as two engines. A narrow starting point that the tree specifications a year later would widen considerably.

    View source ↗