@palasimi/ipa-cluster/dist/ipa-cluster/dbscan.js

Cluster words with similar IPA transcriptions together

"use strict";
// SPDX-License-Identifier: GPL-3.0-or-later
// Copyright (c) 2023 Levi Gruspe
Object.defineProperty(exports, "__esModule", { value: true });
exports.suggestEpsilon = exports.dbscan = void 0;
// Implements a simplified DBSCAN (clustering algorithm).
const mnemonist_1 = require("mnemonist");
// Simplified DBSCAN implementation with minPoints = 2.
// Returns array of arrays of clusters (as indices).
function dbscan(dataset, epsilon, distance) {
    // When minPoints = 2, DBSCAN clusters are like connected components in a graph.
    const sets = new mnemonist_1.StaticDisjointSet(dataset.length);
    for (let i = 1; i < dataset.length; i++) {
        for (let j = 0; j < i; j++) {
            const a = dataset[j];
            const b = dataset[i];
            if (distance(a, b) <= epsilon) {
                sets.union(i, j);
            }
        }
    }
    return sets.compile();
}
exports.dbscan = dbscan;
// Suggests an epsilon value to use with dbscan according to the "elbow"
// method.
function suggestEpsilon(dataset, distance) {
    // A `VPTree` can be constructed in O(nlogn).
    const tree = mnemonist_1.VPTree.from(dataset, distance);
    // Compute distances of 2nd nearest neighbors.
    // This takes around O(nlogn) (O(logn) for each data point).
    const distances = [];
    for (const data of dataset) {
        const neighbors = tree.nearestNeighbors(2, data);
        distances.push(neighbors[neighbors.length - 1].distance);
    }
    distances.sort((a, b) => a - b);
    // Find the elbow/point with sharpest slope.
    let maxSlope = 0;
    let index = 0;
    for (let i = 1; i < distances.length; i++) {
        const slope = distances[i] - distances[i - 1];
        if (slope > maxSlope) {
            maxSlope = slope;
            index = i;
        }
    }
    // Suggest the elbow.
    return distances[index] || 0;
}
exports.suggestEpsilon = suggestEpsilon;
//# sourceMappingURL=dbscan.js.map