tomusdrw / kdtree

K-d tree implementation for nearest neighbours searching.

JavaScript k-d tree implementation.

Supports:

• building balanced tree from predefined data
• performing search for any given number of nearest neighbours

See examples of usage:

• Nearest points on plane
• Classification (knn method)

See test suite.

You can tune distance metric (default is euclid) and provide bucket size. Example:

var data = [
	[1, 2, 3, {additional: "data"}],
	[4, 5, 6, {additional: "data2"}]
];
var k = data[0].length - 1;
var options = {
	/**
	 * Defines bucket (terminal node) maximal size (default: 10)
	 */
	bucketSize : 10,
	/**
	 * Data dimension (default: data[0].length)
	 */
	k : k,
	/**
	 * Dissimilarity function (default: sqrt)
	 */
	dissim : function(sum) {
		return Math.sqrt(sum);
	},
	/**
	 * Coordinate distance (default: (a-b)^2)
	 */
	coordinateDistance : function(a, b) {
		var d = a - b;
		return d * d;
	}
};
var tree = kdtree.buildTree(data, options);
var noOfNeighboursToFind = 2;
var neighbours = tree.search([1, 2, 3], noOfNeighboursToFind);

neighbours[0][k].additional === "data"; //true
neighbours[1][k].additional === "data2"; //true

Implementation is based on paper: JH Friedman, JL Bentley, "An algorithm for finding best matches in logarithmic expected time.", 1977