Class EuclideanSphericalHamerlyKMeans<V extends NumberVector>

  • Type Parameters:
    V - vector datatype
    All Implemented Interfaces:
    Algorithm, ClusteringAlgorithm<Clustering<KMeansModel>>, KMeans<V,​KMeansModel>

    @Priority(-100)
    @Reference(authors="Erich Schubert, Andreas Lang, Gloria Feher",
               title="Accelerating Spherical k-Means",
               booktitle="Int. Conf. on Similarity Search and Applications, SISAP 2021",
               url="https://doi.org/10.1007/978-3-030-89657-7_17",
               bibkey="DBLP:conf/sisap/SchubertLF21")
    public class EuclideanSphericalHamerlyKMeans<V extends NumberVector>
    extends SphericalKMeans<V>
    A spherical k-Means algorithm based on Hamerly's fast k-means by exploiting the triangle inequality in the corresponding Euclidean space.

    Please prefer SphericalHamerlyKMeans, which uses a tighter bound based on Cosines instead.

    FIXME: currently requires the vectors to be L2 normalized beforehand

    Reference:

    Erich Schubert, Andreas Lang, Gloria Feher
    Accelerating Spherical k-Means
    Int. Conf. on Similarity Search and Applications, SISAP 2021

    Since:
    0.8.0
    Author:
    Alexander Voß, Erich Schubert
    • Field Detail

      • LOG

        private static final Logging LOG
        The logger for this class.
      • varstat

        protected boolean varstat
        Flag whether to compute the final variance statistic.
    • Constructor Detail

      • EuclideanSphericalHamerlyKMeans

        public EuclideanSphericalHamerlyKMeans​(int k,
                                               int maxiter,
                                               KMeansInitialization initializer,
                                               boolean varstat)
        Constructor.
        Parameters:
        k - k parameter
        maxiter - Maxiter parameter
        initializer - Initialization method
        varstat - Compute the variance statistic