tech.tablesaw.api.ml.clustering.HierarchicalClustering Maven / Gradle / Ivy
package tech.tablesaw.api.ml.clustering;
import smile.clustering.linkage.Linkage;
import tech.tablesaw.api.NumericColumn;
/**
* Agglomerative Hierarchical Clustering. Hierarchical agglomerative clustering
* seeks to build a hierarchy of clusters in a bottom up approach: each
* observation starts in its own cluster, and pairs of clusters are merged as
* one moves up the hierarchy. The results of hierarchical clustering are
* usually presented in a dendrogram.
*
* In general, the merges are determined in a greedy manner. In order to decide
* which clusters should be combined, a measure of dissimilarity between sets
* of observations is required. In most methods of hierarchical clustering,
* this is achieved by using a linkage criteria which specifies the dissimilarity
* of sets as a function of the pairwise distances of observations in the sets. These
* pairwise distances are specified in a dissimilarity matrix ({@link Proximity}) and the
* linkage criteria that is used to determine the set-level dissimilarities is specified
* in a Linkage object ({@link LinkageFactory}).
*
* Hierarchical clustering has the distinct advantage that any valid measure
* of distance can be used. In fact, the observations themselves are not
* required: all that is used is a matrix of distances.
*
*
References
*
* - David Eppstein. Fast hierarchical clustering andv other applications of dynamic closest pairs. SODA 1998.
*
*
* @see HierarchicalClustering.java
*
* @author Haifeng Li (docs) and Chris Baker (code)
*/
public class HierarchicalClustering {
private final smile.clustering.HierarchicalClustering hclust;
/**
* Constructor.
*
* @param linkageType the type of dissimilarity measure you would like
* to use when determining which clusters to merge
* @param columns the columns of data containing the various measures (i.e., variables)
* describing each observations
*/
public HierarchicalClustering(LinkageFactory.type linkageType, NumericColumn... columns) {
Proximity proximity = new Proximity(columns);
Linkage linkage = new LinkageFactory().createLinkage(proximity.getMatrix(), linkageType);
this.hclust = new smile.clustering.HierarchicalClustering(linkage);
}
public int[][] getTree() {
return hclust.getTree();
}
public double[] getHeight() {
return hclust.getHeight();
}
/**
* Cluster a set of observations into a given number of cluster.
*
* @param k the number of groups/clusters that the observations should be
* broken into
*
* @return the cluster memberships of the observations
*/
public int[] partition(int k) {
// given number of desired clusters, returns clusters
return hclust.partition(k);
}
/**
* Cluster a set of observations based on a given height to cut a dendrogram
* tree at.
*
* @param h the height at which the tree is cut
*
* @return the cluster memberships of the observations
*/
public int[] partition(double h) {
// given height to cut tree at, returns clusters
return hclust.partition(h);
}
}