Relational K-Means Space Sampler

K-means-style pool sampler driven directly by the pairwise distance matrix.

This sampler clusters the fitted pool with the relational k-means objective. For Euclidean distances it is equivalent to ordinary k-means in an unknown embedding. For non-Euclidean distances it remains a heuristic.

If known_pool is given, those points influence the implicit centroids but are not eligible for selection.

Creates a new relational k-means space sampler.

Format

R6::R6Class object inheriting from SpaceSamplerDistance.

Arguments

distance

(ALDistance)
Distance used to construct the pairwise dissimilarity matrix.

Construction

clx_sps("relational_kmeans")

See also

Other SpaceSampler: SpaceSampler, SpaceSamplerDistance, mlr_space_samplers, mlr_space_samplers_chain, mlr_space_samplers_conditional, mlr_space_samplers_gsx, mlr_space_samplers_kmeans, mlr_space_samplers_kmedoids, mlr_space_samplers_lhs, mlr_space_samplers_sobol, mlr_space_samplers_uniform