q-exp/test/generaliz.js

Reinforcement learning (Q-Learning) library

"use strict";

/**
 * Generalisation test
 */

var Gen   = require('../generaliz.js');
var State = require('../state.js');

// Test dataset
var S = [
	new State([1,5,3]),
	new State([2,5,4]),
	new State([2,5,7]),
	new State([1,2,3]),
	new State([0,3,2]),
	new State([2,2,1]),
	new State([1,1,1]),
	new State([3,4,5])
]
var R = [
	3,
	3,
	10,
	10,
	3,
	0,
	0,
	10
]

const maxIters = 100;
const alpha = 0.0001;
var ϴ = Gen.fit(S,R,maxIters,alpha,'GD');

// Comparison
console.log('=============== GRADIENT DESCENT =========='.green)
var estimate  = Gen.estimate(ϴ);
S.forEach((s,n) => {
	var actualR = R[n];
	var estR    = estimate(s);

	console.log('#',n,' ',
		'expected: '.green,actualR,' ',
		'get : '.magenta,estR
	);
})