Skip to contents

Gower Active Learning Distance

Source: R/ALDistanceGower.R
mlr_al_distances_gower.Rd

Distance for mixed-type active-learning search spaces.

Numeric and integer columns contribute absolute differences, optionally rescaled according to the scale hyperparameter. Logical and factor columns contribute 0 for equal values and 1 for different values. The final distance is the average across all observed per-column contributions.

Numeric columns with zero scale are ignored. Missing values are excluded from the pairwise average for the affected column.

Creates a new Gower active-learning distance.

Format

R6::R6Class object inheriting from ALDistance.

Construction

clx_ald("gower")

The configuration hyperparameter scale controls numeric scaling:

  • "minmax_auto" divides by the observed pool range when a pool is supplied, and by the finite search-space range when fitted with xdt = NULL.

  • "minmax_space" divides by the search-space range.

  • "minmax_empirical" divides by the observed pool range.

  • "standardize" divides by the observed pool standard deviation.

  • "off" leaves numeric differences unscaled.

See also

Other ALDistance: ALDistance, ALDistanceGeometry, mlr_al_distances, mlr_al_distances_affine, mlr_al_distances_standardize