hdbscanjs

Hierarchical DBSCAN Clustering in JavaScript -- In Progress

A JavaScript-based and simplified version of hdbscan.

Installing

npm install hdbscanjs

Example

import Clustering from 'hdbscanjs';

const dataset = [
  {data: [0,0], opt: 0},
  ....
];

// two distance measure functions are supported:
// 1) euclidean
// 2) geoDist (take inputs as lonlat points)
const distFunc = Clustering.distFunc.geoDist;

const cluster = new Clustering(dataset, distFunc);
const treeNode = cluster.getTree();

const filterFunc = val => ...;
const bbox = {minX:.., maxX:.., minY:.., maxY:..};
const filteredNodes = treeNode.filter(filterFunc, bbox);

The returned treeNode object contains the following attributes:

• left: a pointer to the left child.
• right: a pointer to the right child.
• data: a list of points in the current cluster
• index: a list of indices corresponding to the points in the current cluster
• opt: a user-defined object that is aggregated (combined as a list using concat) during the clustering process
• dist: the distance between the two child clusters (null if the current node is a leaf)
• edge: the closest pair of points from the two child clusters: [[p1x, p1y], [p2x, p2y]] (null if the current node is a leaf)
• bbox: the bounding box of the current cluster ({minX:.., maxX:.., minY:.., maxY:..})

The treeNode object contains a filter function that performs a top-down recursive filtering operation. If true, the test terminates and the current node is returned. Otherwise, the child nodes are tested. The return value of the filter function is a flag list of treeNode. The filter function is useful for trimming the cluster nodes based on certain conditions (e.g., current viewport).

The filter function takes an optional parameter called bbox, which defines a bounding box. If not null, only the nodes that intersect with the bbox will be returned.

License

This project is licensed under the MIT License.