LoginSign Up
← Back to all sparks
M

mlr3filters

ANALYTICS
Velocity0.0

Feature selection filters for the mlr3 machine learning framework

github.com/mlr-org/mlr3filters ↗

mlr3filters grows one feature-selection filter at a time

feature-selectionmlr3machine-learningr-statsfilters
◆Current state
mlr3filters provides feature-filter methods to mlr3. Its releases follow a consistent shape: one or two new filters, broader feature-type support on existing ones, and error-message work. Boruta and a univariate Cox filter arrived in 0.8.0; 0.9.0 extended Boruta to logical, factor and ordered features and moved param_set to an active binding.
◆Where it's heading
This is incremental infrastructure that tracks mlr3's own conventions — cli printing, prototype-based dictionaries, featureless learners as defaults — while slowly widening which data types each filter accepts. Nothing in the recent history suggests a change of scope.
◆Prediction
Expect another filter or two plus continued feature-type broadening, keeping pace with mlr3 core conventions.

◆Recent moves

  1. 3mo ago

    Filter dictionary listing now uses prototypes

    A single fix to how the filter dictionary converts to a data.table. Follows mlr3's shared dictionary conventions.

    View source ↗
  2. 11mo ago

    Boruta handles logical, factor and ordered features

    Widens Boruta's accepted feature types, moves param_set to an active binding and fixes resampling reuse in FilterPerformance. The type broadening is what most users will notice.

    View source ↗
  3. 2y ago

    Boruta and univariate Cox filters added

    Adds two filters, including a survival-specific one contributed from the mlr3proba side. The clearest capability addition in this window.

    View source ↗
  4. 3y ago

    Missing-value tagging and wider CarScore feature support

    Tags filters that handle missing values gracefully and broadens CarScore's supported types. Incremental robustness work.

    View source ↗
  5. 3y ago

    Missing-value checks and featureless learner defaults

    Checks features for missing values to give better errors, removes deprecated functions and switches defaults to the featureless learner. Housekeeping that changes default behaviour.

    View source ↗
  6. 3y ago

    Survival CAR score filter and pipeline documentation

    Adds FilterCarSurvScore, switches importance and performance filters to a featureless default learner, and documents PipeOpFilter usage. Same one-filter-per-release rhythm.

    View source ↗