@ai-on-browser/data-analysis-models/lib/model/segmented.js

Data analysis model package without any dependencies

import Matrix from '../util/matrix.js'

/**
 * Segmented regression
 */
export default class SegmentedRegression {
	// http://funyakofunyao.click/2017/12/16/%E6%8A%98%E3%82%8C%E7%B7%9A%E5%9E%8B%E5%9B%9E%E5%B8%B0%E5%88%86%E6%9E%90%E3%80%82%E3%82%B9%E3%83%97%E3%83%A9%E3%82%A4%E3%83%B3%E5%85%A5%E9%96%80/
	/**
	 * @param {number} [seg] Number of segments
	 */
	constructor(seg = 3) {
		this._b = null
		this._seg = seg
		this._r = 40
	}

	/**
	 * Fit model.
	 * @param {Array<Array<number>>} x Training data
	 * @param {Array<Array<number>>} y Target values
	 */
	fit(x, y) {
		x = x.map(v => v[0])
		y = Matrix.fromArray(y)
		const n = x.length

		let min = Infinity
		let max = -Infinity
		for (let i = 0; i < n; i++) {
			min = Math.min(min, x[i])
			max = Math.max(max, x[i])
		}
		const sepx = []
		for (let i = 0; i < this._r; i++) {
			sepx.push(min + (i * (max - min)) / this._r)
		}
		sepx.push(max)

		const sepidx = Array.from({ length: this._seg - 1 }, (_, i) => i)

		let min_err = Infinity
		let best_w = null
		let best_seps = null
		const xs = Matrix.ones(n, sepidx.length + 2)
		for (let i = 0; i < n; i++) {
			xs.set(i, 1, x[i])
		}
		do {
			for (let i = 0; i < n; i++) {
				const v = x[i]
				for (let k = 0; k < sepidx.length; k++) {
					xs.set(i, k + 2, Math.max(0, v - sepx[sepidx[k]]))
				}
			}
			const xtx = xs.tDot(xs)

			const w = xtx.solve(xs.tDot(y))
			const yh = xs.dot(w)
			yh.sub(y)
			const e = yh.norm()
			if (e < min_err) {
				min_err = e
				best_w = w
				best_seps = sepidx.map(i => sepx[i])
			}
		} while (this._next_idx(sepidx, sepx.length))
		this._w = best_w
		this._seps = best_seps
	}

	_next_idx(a, n) {
		for (let i = a.length - 1; i >= 0; i--) {
			a[i]++
			if (a[i] <= n - a.length + i) {
				for (let k = i + 1; k < a.length; k++) {
					a[k] = a[i] + k - i
				}
				break
			}
		}
		return a[a.length - 1] < n
	}

	/**
	 * Returns predicted values.
	 * @param {Array<Array<number>>} x Sample data
	 * @returns {Array<Array<number>>} Predicted values
	 */
	predict(x) {
		const n = x.length
		const xs = new Matrix(n, this._seps.length + 2)
		for (let i = 0; i < n; i++) {
			const v = x[i][0]
			xs.set(i, 0, 1)
			xs.set(i, 1, v)
			for (let k = 0; k < this._seps.length; k++) {
				xs.set(i, k + 2, Math.max(0, v - this._seps[k]))
			}
		}
		return xs.dot(this._w).toArray()
	}
}