Remove KMeans model dependency from some models (#820)

Author: ishii-norimi

lib/model/gmeans.js

@@ -1,6 +1,69 @@
 import Matrix from '../util/matrix.js'
 
-import { KMeans } from './kmeans.js'
+class KMeans {
+	constructor(x, k) {
+		this._x = x
+		this._k = k
+
+		const n = this._x.length
+		const idx = []
+		for (let i = 0; i < this._k; i++) {
+			idx.push(Math.floor(Math.random() * (n - i)))
+		}
+		for (let i = idx.length - 1; i >= 0; i--) {
+			for (let j = idx.length - 1; j > i; j--) {
+				if (idx[i] <= idx[j]) {
+					idx[j]++
+				}
+			}
+		}
+		this._c = idx.map(v => this._x[v])
+
+		this._d = (a, b) => Math.sqrt(a.reduce((s, v, i) => s + (v - b[i]) ** 2, 0))
+	}
+
+	get centroids() {
+		return this._c
+	}
+
+	fit() {
+		const p = this.predict()
+
+		const c = this._c.map(p => Array.from(p, () => 0))
+		const count = Array(this._k).fill(0)
+		const n = this._x.length
+		for (let i = 0; i < n; i++) {
+			for (let j = 0; j < this._x[i].length; j++) {
+				c[p[i]][j] += this._x[i][j]
+			}
+			count[p[i]]++
+		}
+		let d = 0
+		for (let k = 0; k < this._k; k++) {
+			const mc = c[k].map(v => v / count[k])
+			d += this._c[k].reduce((s, v, j) => s + (v - mc[j]) ** 2, 0)
+			this._c[k] = c[k].map(v => v / count[k])
+		}
+		return d
+	}
+
+	predict() {
+		const p = []
+		const n = this._x.length
+		for (let i = 0; i < n; i++) {
+			let min_d = Infinity
+			p[i] = -1
+			for (let k = 0; k < this._k; k++) {
+				const d = this._d(this._x[i], this._c[k])
+				if (d < min_d) {
+					min_d = d
+					p[i] = k
+				}
+			}
+		}
+		return p
+	}
+}
 
 const cvTable = [
 	[0.514, 0.578, 0.683, 0.779, 0.926],
@@ -123,16 +186,13 @@ export default class GMeans {
 	}
 
 	_split_cluster(datas, k = 2) {
-		const kmeans = new KMeans()
-		for (let i = 0; i < k; i++) {
-			kmeans.add(datas)
-		}
-		while (kmeans.fit(datas) > 0);
+		const kmeans = new KMeans(datas, k)
+		while (kmeans.fit() > 0);
 		return this._create_clusters(kmeans, datas)
 	}
 
 	_create_clusters(model, datas) {
-		const k = model.size
+		const k = model.centroids.length
 		const p = model.predict(datas)
 		const ds = []
 		for (let i = 0; i < k; ds[i++] = []);

lib/model/lbg.js

@@ -1,7 +1,5 @@
 import Matrix from '../util/matrix.js'
 
-import { KMeans } from './kmeans.js'
-
 /**
  * Linde-Buzo-Gray algorithm
  */
@@ -65,9 +63,28 @@ export default class LBG {
 			new_centroids.push(cp, cn)
 		}
 
-		const kmeans = new KMeans()
-		kmeans._centroids = new_centroids
-		while (kmeans.fit(datas) > 0) this._centroids = kmeans.centroids
+		this._centroids = new_centroids
+		let d = 0
+		do {
+			const p = this.predict(datas)
+
+			const size = this._centroids.length
+			const c = this._centroids.map(p => Array.from(p, () => 0))
+			const count = Array(size).fill(0)
+			const n = datas.length
+			for (let i = 0; i < n; i++) {
+				for (let j = 0; j < datas[i].length; j++) {
+					c[p[i]][j] += datas[i][j]
+				}
+				count[p[i]]++
+			}
+			d = 0
+			for (let k = 0; k < size; k++) {
+				const mc = c[k].map(v => v / count[k])
+				d += this._centroids[k].reduce((s, v, j) => s + (v - mc[j]) ** 2, 0)
+				this._centroids[k] = c[k].map(v => v / count[k])
+			}
+		} while (d > 0)
 	}
 
 	/**

lib/model/xmeans.js

@@ -1,6 +1,69 @@
 import Matrix from '../util/matrix.js'
 
-import { KMeans } from './kmeans.js'
+class KMeans {
+	constructor(x, k) {
+		this._x = x
+		this._k = k
+
+		const n = this._x.length
+		const idx = []
+		for (let i = 0; i < this._k; i++) {
+			idx.push(Math.floor(Math.random() * (n - i)))
+		}
+		for (let i = idx.length - 1; i >= 0; i--) {
+			for (let j = idx.length - 1; j > i; j--) {
+				if (idx[i] <= idx[j]) {
+					idx[j]++
+				}
+			}
+		}
+		this._c = idx.map(v => this._x[v])
+
+		this._d = (a, b) => Math.sqrt(a.reduce((s, v, i) => s + (v - b[i]) ** 2, 0))
+	}
+
+	get centroids() {
+		return this._c
+	}
+
+	fit() {
+		const p = this.predict()
+
+		const c = this._c.map(p => Array.from(p, () => 0))
+		const count = Array(this._k).fill(0)
+		const n = this._x.length
+		for (let i = 0; i < n; i++) {
+			for (let j = 0; j < this._x[i].length; j++) {
+				c[p[i]][j] += this._x[i][j]
+			}
+			count[p[i]]++
+		}
+		let d = 0
+		for (let k = 0; k < this._k; k++) {
+			const mc = c[k].map(v => v / count[k])
+			d += this._c[k].reduce((s, v, j) => s + (v - mc[j]) ** 2, 0)
+			this._c[k] = c[k].map(v => v / count[k])
+		}
+		return d
+	}
+
+	predict() {
+		const p = []
+		const n = this._x.length
+		for (let i = 0; i < n; i++) {
+			let min_d = Infinity
+			p[i] = -1
+			for (let k = 0; k < this._k; k++) {
+				const d = this._d(this._x[i], this._c[k])
+				if (d < min_d) {
+					min_d = d
+					p[i] = k
+				}
+			}
+		}
+		return p
+	}
+}
 
 /**
  * x-means
@@ -92,16 +155,13 @@ export default class XMeans {
 	}
 
 	_split_cluster(datas, k = 2) {
-		const kmeans = new KMeans()
-		for (let i = 0; i < k; i++) {
-			kmeans.add(datas)
-		}
-		while (kmeans.fit(datas) > 0);
+		const kmeans = new KMeans(datas, k)
+		while (kmeans.fit() > 0);
 		return this._create_clusters(kmeans, datas)
 	}
 
 	_create_clusters(model, datas) {
-		const k = model.size
+		const k = model.centroids.length
 		const p = model.predict(datas)
 		const ds = []
 		for (let i = 0; i < k; ds[i++] = []);