Initialization methods for BIRCH-based k-means and EM clustering.
Interface Summary Interface Description CFInitWeightInitialization weight function for k-means initialization with BETULA.
Class Summary Class Description AbstractCFKMeansInitializationAbstract base class for CF k-means initializations. AbstractCFKMeansInitialization.ParParameterization class. CFKPlusPlusLeavesK-Means++-like initialization for BETULA k-means, treating the leaf clustering features as a flat list, and called "leaves" in the publication. CFKPlusPlusLeaves.ParParameterization class. CFKPlusPlusTreeInitialize K-means by following tree paths weighted by their variance contribution. CFKPlusPlusTree.ParParameterization class. CFKPlusPlusTrunkTrunk strategy for initializing k-means with BETULA: only the nodes up to a particular level are considered for k-means++ style initialization. CFKPlusPlusTrunk.ParParameterization class. CFRandomlyChosenInitialize K-means by randomly choosing k existing elements as initial cluster centers for Clustering Features. CFRandomlyChosen.ParParameterization class. CFWeightedRandomlyChosenInitialize K-means by randomly choosing k existing elements as initial cluster centers for Clustering Features. CFWeightedRandomlyChosen.ParParameterization class. InterclusterWeightInitialization via n2 * D2²(cf1, cf2), which supposedly is closes to the idea of k-means++ initialization. SquaredEuclideanWeightUse the squared Euclidean distance only for distance measurement. VarianceWeightVariance-based weighting scheme for k-means clustering with BETULA.