ml-regression/src/__tests__/kernel-ridge-regression.test.js

Regression algorithms

import { Matrix } from "ml-matrix";
import { describe, it, expect } from "vitest";

import { KernelRidgeRegression } from "..";

let nSamples = 10;
let nVars = 2;

let Xs = Matrix.random(nSamples, nVars);
Xs.sub(0.5);
let Ys = Matrix.zeros(nSamples, 1);
for (let i = 0; i < nSamples; i++) {
  Ys.set(
    i,
    0,
    Xs.get(i, 0) * Xs.get(i, 0) +
      2 * Xs.get(i, 0) * Xs.get(i, 1) +
      Xs.get(i, 1) * Xs.get(i, 1),
  );
}

describe("Kernel ridge regression", () => {
  it("constant outputs", () => {
    let model = new KernelRidgeRegression(
      [
        [0, 0],
        [1, 1],
      ],
      [[0], [0]],
    );
    expect(
      model.predict([
        [1, 1],
        [2, 5],
        [4, 7],
      ]),
    ).toStrictEqual([[0], [0], [0]]);
  });
  it("Polynomial kernel should overfit the pattern", () => {
    let model = new KernelRidgeRegression(Xs, Ys, {
      kernelType: "polynomial",
      lambda: 0.0001,
      kernelOptions: { degree: 2, constant: 1 },
    });
    let Y = model.predict(Xs.to2DArray());

    for (let i = 0; i < Y.length; i++) {
      expect(Y[i][0]).toBeCloseTo(Ys.get(i, 0), 5e-3);
    }
  });
  it("Gaussian kernel should overfit the pattern", () => {
    let model = new KernelRidgeRegression(Xs, Ys, {
      kernelType: "gaussian",
      lambda: 0.0001,
      kernelOptions: { sigma: 0.1 },
      computeQuality: true,
    });
    let Y = model.predict(Xs.to2DArray());
    for (let i = 0; i < Y.length; i++) {
      expect(Y[i][0]).toBeCloseTo(Ys.get(i, 0), 5e-3);
    }
  });
  it("Load and export model", () => {
    let regression = new KernelRidgeRegression(true, {
      name: "kernelRidgeRegression",
      alpha: 1,
      inputs: 1,
      kernelType: "gaussian",
      kernelOptions: {},
    });
    expect(regression.alpha).toBe(1);
    expect(regression.kernelType).toBe("gaussian");

    let model = regression.toJSON();
    expect(model.name).toBe("kernelRidgeRegression");
    expect(model.alpha).toBe(1);
    expect(model.kernelType).toBe("gaussian");
  });
});