als-statistics/tests/golden/tools/dbscan.js

Modular JS statistics toolkit for Node.js and the browser: descriptive stats, correlations (Pearson/Spearman/Kendall), t-tests & ANOVA (Student/Welch), reliability (Cronbach’s alpha), regression (linear/logistic), clustering (DBSCAN/HDBSCAN), and table/co

// Предполагаемый интерфейс: new DBSCAN(points, eps, minPts).run()
// Возвращает массив меток кластеров длины points.length, где -1 = шум
import DBSCAN from '../../../lib/analyze/dbscan/dbscan.js';
import { makeRng } from '../_rng.js';

function blobs(R, cfg) {
  const { centers, nPer, sigma } = cfg;
  const pts = {};
  centers.forEach(([cx, cy],i) => {
    const xs = R.normal(cx, sigma, nPer);
    const ys = R.normal(cy, sigma, nPer);
    for (let i=0;i<nPer;i++) pts[`v${i}`] = [xs[i], ys[i]];
  });
  return pts;
}

export const dbscanTool = {
  id: 'dbscan',
  cases: 5,
  gen(seed, i) {
    const R = makeRng(`${seed}/db/${i}`);
    const cfg = {
      centers: [[0,0],[6,0],[0,6]],
      nPer: 40 + 5*(i%3),
      sigma: 0.6 + 0.1*(i%2)
    };
    const pts = blobs(R, cfg);
    // eps выбираем заметно меньше межкластерного расстояния
    const eps = 1.2, minPts = 5;
    return { pts, eps, minPts };
  },
  compute({pts, eps, minPts}) {
    const {labels} = new DBSCAN(pts,{eps, minPts});
    // метрики, устойчивые к переименованию кластеров:
    const k = new Set(labels.filter(l => l>=0)).size;
    const sizes = Object.values(labels.reduce((m,l)=>{
      if (l>=0) m[l]=(m[l]||0)+1; return m;
    },{})).sort((a,b)=>b-a);
    const noise = labels.filter(l=>l<0).length / labels.length;
    return { k, sizes, noise };
  },
  compare(exp, got, approx) {
    if (got.k !== exp.k) throw new Error(`clusters k ${got.k} != ${exp.k}`);
    if (got.sizes.length !== exp.sizes.length) throw new Error('sizes length mismatch');
    for (let i=0;i<exp.sizes.length;i++) if (got.sizes[i]!==exp.sizes[i]) throw new Error(`size[${i}] mismatch`);
    approx(got.noise, exp.noise);
  },
};