Standardized Active Learning Distance

Distance for the greedy-sampling family of active learning methods.

Numeric and integer columns stay numeric, logical and factor columns are one-hot encoded, and every geometry column is standardized to mean 0 and standard deviation 1. With a finite pool, center and scale are estimated from the candidate pool. With xdt = NULL, numeric centers and scales are derived from finite search-space bounds under a uniform distribution, and dummy-column centers and scales are derived from uniform categorical levels.

Missing or dependency-inactive values are mapped to 0 in the standardized geometry.

Creates a standardized active-learning distance.

Format

R6::R6Class object inheriting from ALDistanceGeometry.

Construction

clx_ald("standardize")

See also

Other ALDistance: ALDistance, ALDistanceGeometry, mlr_al_distances, mlr_al_distances_affine, mlr_al_distances_gower