nipals-pca/test/appTest.js

A NIPALS implementation of PCA that handles missing values

const chai = require('chai');
const {expect} = chai;
chai.use(require('chai-roughly'));

const { Dataset } = require('../src/Dataset');
const { PCA } = require('../src/pca');
var fs = require('fs');
var Papa = require('papaparse');

let inputMatrixPromise =  parseExampleInput(fs.readFileSync('test/data/iris/IRIS.csv', 'utf8'));
let mv_inputMatrixPromise =  parseExampleInput(fs.readFileSync('test/data/iris/IRIS_mv.csv', 'utf8'));
let meansPromise =  parseExampleInput(fs.readFileSync('test/data/iris/irisMeanValues.txt', 'utf8'));
let stdPromise =  parseExampleInput(fs.readFileSync('test/data/iris/irisStdValues.txt', 'utf8'));
let mv_meansPromise =  parseExampleInput(fs.readFileSync('test/data/iris/irisMeanValues_mv.txt', 'utf8'));
let mv_stdPromise =  parseExampleInput(fs.readFileSync('test/data/iris/irisStdValues_mv.txt', 'utf8'));
let scoresPromise =  parseExampleInput(fs.readFileSync('test/data/iris/irisT.txt', 'utf8'));
let loadingsPromise =  parseExampleInput(fs.readFileSync('test/data/iris/irisP.txt', 'utf8'));
let mv_scoresPromise =  parseExampleInput(fs.readFileSync('test/data/iris/IRIS_mv_T.txt', 'utf8'));
let mv_loadingsPromise =  parseExampleInput(fs.readFileSync('test/data/iris/IRIS_mv_P.txt', 'utf8'));

describe('Test Dataset creation and normalization', () => {

	let jsonMatrix;
	let means,stdevs;

	let dataset;

	before(async ()=>{
		jsonMatrix = await inputMatrixPromise;
		//remove last column
		jsonMatrix.map(r => r.splice(-1, 1));

		dataset = new Dataset(jsonMatrix);

		means = await meansPromise;
		means = means.map(([value])=>parseFloat(value));

		stdevs = await stdPromise;
		stdevs = stdevs.map(([value])=>1/parseFloat(value))
	});

	it('Check centering', () => {
		expect(means).to.roughly(0.001).deep.equal(dataset.means);
	})

	it('Check UV scaling', () => {
		expect(stdevs).to.roughly(0.001).deep.equal(dataset.stdevs);
	})
})

describe('Test PCA', () => {

	let pca;

	before(async ()=>{
		let components = 3;

		jsonMatrix = await parseExampleInput(fs.readFileSync('test/data/iris/IRIS.csv', 'utf8'));
		//remove last column
		jsonMatrix.map(r => r.splice(-1, 1));

		dataset = new Dataset(jsonMatrix);

		scores = await scoresPromise;
		scores.splice(0,1);
		scores = scores.map(row=>row.slice(1,components+1).map(v=>parseFloat(v)));

		loadings = await loadingsPromise;
		loadings.splice(0,1);
		loadings = loadings.map(row=>row.slice(1,components+1).map(v=>parseFloat(v)));

		pca = new PCA(dataset,components);
		pca.fit();
	});

	let tolerance = 0.001;

	for (let a = 0; a < 3; a++) {
		it(`PCA loadings and scores are correct for component ${a+1}`, () => {
			expect(loadings.map(row=>Math.abs(row[a]))).to.roughly(tolerance).deep.equal(pca.P.getColumn(a).map(v=>Math.abs(v)));
			expect(scores.map(row=>Math.abs(row[a]))).to.roughly(tolerance).deep.equal(pca.T.getColumn(a).map(v=>Math.abs(v)));
		})
	}
})

describe('Test MV Dataset creation and normalization', () => {

	let jsonMatrix;
	let means,stdevs;

	let dataset;

	before(async ()=>{
		jsonMatrix = await mv_inputMatrixPromise;
		//remove last column
		jsonMatrix.map(r => r.splice(0, 1));

		dataset = new Dataset(jsonMatrix);

		means = await mv_meansPromise;
		means = means.map(([value])=>parseFloat(value));

		stdevs = await mv_stdPromise;
		stdevs = stdevs.map(([value])=>1/parseFloat(value))
	});

	it('Check centering', () => {
		expect(means).to.roughly(0.001).deep.equal(dataset.means);
	})

	it('Check UV scaling', () => {
		expect(stdevs).to.roughly(0.001).deep.equal(dataset.stdevs);
	})
})

describe('Test PCA with missing values', () => {

	let pca;

	before(async ()=>{
		let components = 3;

		jsonMatrix = await parseExampleInput(fs.readFileSync('test/data/iris/IRIS_mv.csv', 'utf8'));
		//remove last column
		jsonMatrix.map(r => r.splice(0, 1));

		dataset = new Dataset(jsonMatrix);

		scores = await mv_scoresPromise;
		scores.splice(0,1);
		scores = scores.map(row=>row.slice(1,components+1).map(v=>parseFloat(v)));

		loadings = await mv_loadingsPromise;
		loadings.splice(0,1);
		loadings = loadings.map(row=>row.slice(1,components+1).map(v=>parseFloat(v)));

		pca = new PCA(dataset,components);
		pca.fit();
	});

	let tolerance = 0.001;

	for (let a = 0; a < 3; a++) {
		it(`PCA loadings and scores are correct for component ${a+1}`, () => {
			expect(loadings.map(row=>Math.abs(row[a]))).to.roughly(tolerance).deep.equal(pca.P.getColumn(a).map(v=>Math.abs(v)));
			expect(scores.map(row=>Math.abs(row[a]))).to.roughly(tolerance).deep.equal(pca.T.getColumn(a).map(v=>Math.abs(v)));
		})
	}
})

function parseExampleInput(file) {

	return new Promise((resolve, reject) => {
		Papa.parse(file, {
			header: false,
			skipEmptyLines: true,
			complete(results, file) {

				let data = results.data;

				resolve(data)
			},
			error(err, file) {
				reject(err)
			}
		})
	})

}