Hierarchical Tree (Dendrogram)

Hierarchical clustering builds a recursive tree structure called a Dendrogram. The length of the branches represents the distance at which groups merge, and the leaves represent individual data points.

Unlike partitioning algorithms like K-Means, which just group points into K buckets, hierarchical methods give you a full multiscale representation of cluster relationships.

DatasetIrisWinePenguins

Linkage StrategySingle LinkAverage LinkComplete Link

Hierarchical Clustering builds a tree representation (Dendrogram). The branch lengths represent the distance at which groups merge. Use the controls to see how different linkage criteria alter the clustering tree topology.

How it works

1. Initialization: Every point starts as its own cluster (leaf node).
2. Linkage Criterion: The algorithm searches for the two closest clusters in the dataset. Distance between clusters can be measured using different Linkage Strategies:
   • Single Link: The distance between the two closest elements of the clusters. Tends to "chain" points together.
   • Complete Link: The distance between the two farthest elements of the clusters. Tends to form compact, spherical clusters.
   • Average Link: The average distance between all pairs of elements. A balanced approach.
3. Merging: The two closest clusters are merged into a new node. The distance where they merge is recorded as the branch length.
4. Iterate: Step 2 and 3 repeat until all points belong to a single root cluster.

How-to (Code)

To get the full tree structure, access the root property of the HierarchicalClustering instance.

import * as druid from "@saehrimnir/druidjs";

const data = [
  /* ... multi-dimensional data ... */
];

// 1. Initialize with your preferred linkage
const hc = new druid.HierarchicalClustering(data, {
  linkage: "complete",
});

// 2. Fetch the recursive tree structure
const rootNode = hc.root;

// Example parsing logic for tree traversal
function traverse(node) {
  if (node.isLeaf) {
    console.log(`Leaf point index: ${node.index}`);
  } else {
    console.log(`Merge distance: ${node.dist}`);
    traverse(node.left);
    traverse(node.right);
  }
}

traverse(rootNode);