PAMKConfig
VariantPartitioning Around Medoids with estimation of the number of clusters.
Properties
krangeinteger[]≥ 1 itemsunique itemsCandidate numbers of clusters to compare. Neither criterion can score a single cluster, so including 1 selects it only when a Duda-Hart test finds no evidence for more.
criterionstringCriterion used to choose the number of clusters: "asw" = average silhouette width, "multiasw" = average silhouette width computed on random subsets, "ch" = Calinski-Harabasz index.
one of
"asw""multiasw""ch"use_pambooleanFit each candidate with PAM. Disabling it uses CLARA instead, which subsamples and is intended for large datasets.
scalingbooleanScale each variable by its root mean square, after centering, before clustering.
alphanumber≥ 0≤ 1Significance level of the Duda-Hart test, which decides between one cluster and more. Only used when 1 is among the candidates.
n_subsetsinteger | null≥ 1Number of random subsets the average silhouette width is computed on. Applies only when criterion is "multiasw".
Relationships
Used by