LoginSign Up
← Back to all sparks
F

filtro

ANALYTICS
Velocity0.0

Tidy tools to apply filter-based supervised feature selection methods

github.com/tidymodels/filtro ↗

filtro moves to S7 and multiplies its feature-scoring methods in a single release.

feature selectiontidymodelss7filter methodsr package
◆Current state
filtro supplies feature-selection filter scores for the tidymodels stack. Version 0.2.0 adds five scoring methods — correlation, random forest importance, information gain, ROC AUC and cross tabulation — and moves the package from S3 to S7. It also gains a ranking layer: show_best_score_* and rank_best_score_* helpers for a single score, plus desirability-function helpers for optimising across several scores at once.
◆Where it's heading
The package is being built out on two axes at once — the catalogue of scores, and the machinery for choosing between them. The desirability functions are the more telling half, since they assume users will filter on several criteria rather than one. Adopting S7 while still pre-1.0 suggests the object model is being settled before the API is frozen.
◆Prediction
Expect more scoring methods on the same S7 interface and a 1.0 release once the score and ranking APIs stop moving; the ranking helpers' naming is the most likely thing to change first.

◆Recent moves

  1. 11mo ago

    Five new filter scores and the move to S7

    ⚡ SPARK

    The release that gives filtro both a real catalogue of scores and the machinery to rank them, on a new S7 foundation. The desirability helpers, which optimise across several scores at once, are the part that changes what the package is for.

    View source ↗
  2. 11mo ago

    Development build: score transformation handling

    A pre-release development snapshot squashing work on score transformation functions, published hours before 0.2.0. It records an in-progress design rather than a shipped change.

    View source ↗