var canvas = document.getElementById("myCanvas"); var ctx = canvas.getContext("2d"); const gravity = 21; const delta = 0.016; const opening = 100; const minOpen = 25; const divisions = 10; const nColums = divisions / 2 - 1; const offset = 300; const tubesVelocity = 1.5; const minOpenSosia = 50; const populationSize = 200; var score = 0; var visualScore = 0; var generation = 0; var bestScore = 0; canvas.width = window.innerWidth -50; canvas.height = 300; class Matrix{ constructor(rows, columns){ this.rows = rows; this.columns = columns; this.data = []; for (let row = 0; row < this.rows; row++){ this.data.push([]); } //this.randomize(); } randomize(){ for (let row = 0; row < this.rows; row++){ for (let column = 0; column < this.columns; column++){ this.data[row][column] = Math.random() * 2 - 1; } } } static multiply(a, b){ if(a.columns != b.rows) return; let prodotto = new Matrix(a.rows, b.columns); for (let i = 0; i < prodotto.rows; i++){ for (let y = 0; y < prodotto.columns; y++){ let somma = 0; for (let z = 0; z < b.rows; z++){ somma += a.data[i][z] * b.data[z][y]; } prodotto.data[i][y] = somma; } } return prodotto; } static convertArray(array){ let risultato = new Matrix(array.length, 1); for (let row = 0; row < risultato.rows; row++){ risultato.data[row][0] = array[row]; } return risultato; } map(funzione){ for (let row = 0; row < this.rows; row++){ for (let column = 0; column < this.columns; column++){ this.data[row][column] = funzione(this.data[row][column]); } } } add(n){ for (let row = 0; row < this.rows; row++){ for (let column = 0; column < this.columns; column++){ this.data[row][column] += n.data[row][column]; } } } mutate(rate){ for (let row = 0; row < this.rows; row++){ for (let column = 0; column < this.columns; column++){ if(Math.random() < rate){ this.data[row][column] = (Math.random() * 2) - 1; } } } } stampa(){ console.log(this); console.table(this.data); } } function sigmoid(n){ return 1 / (1 + Math.exp(-n)); } class NeuralNetwork{ //TODO: I BIAS constructor(numeroNodiInput, numeroNodiHidden, numeroNodiOutput){ this.numeroNodiInput = numeroNodiInput; this.numeroNodiHidden = numeroNodiHidden; this.numeroNodiOutput = numeroNodiOutput; this.pesiInputHidden = new Matrix(numeroNodiHidden, numeroNodiInput); this.pesiInputHidden.randomize(); this.weightBiasIH = new Matrix(numeroNodiHidden, 1); this.weightBiasIH.randomize(); this.pesiHiddenOutput = new Matrix(numeroNodiOutput, numeroNodiHidden); this.pesiHiddenOutput.randomize(); this.weightBiasHO = new Matrix(numeroNodiOutput, 1); this.weightBiasHO.randomize(); } //feed forward predict(inputs){ inputs = Matrix.convertArray(inputs); let nodiHidden = Matrix.multiply(this.pesiInputHidden, inputs); nodiHidden.add(this.weightBiasIH); nodiHidden.map(sigmoid); let nodiOutput = Matrix.multiply(this.pesiHiddenOutput, nodiHidden); nodiOutput.add(this.weightBiasHO); nodiOutput.map(sigmoid); return nodiOutput; } mutate(rate){ this.pesiInputHidden.mutate(rate); this.pesiHiddenOutput.mutate(rate); } } class Tube{ constructor(x, y, width, height){ this.x = x; this.y = y; this.width = width / 1.5; this.height = height; } update(){ this.x -= tubesVelocity; if(this.x < - canvas.width / divisions){ this.x = canvas.width + (canvas.width / divisions) * (divisions / 2); this.height = Math.floor(Math.random() * (canvas.height - opening - minOpen - minOpenSosia)) + minOpen; } } render(){ ctx.fillStyle = "green"; ctx.fillRect(this.x, this.y, this.width, this.height); ctx.fillRect(this.x, this.height + opening, this.width, canvas.height); ctx.lineWidth = 3; ctx.strokeStyle = "black"; ctx.strokeRect(this.x, this.y, this.width, this.height); ctx.strokeRect(this.x, this.height + opening, this.width, canvas.height); } } class Tubes{ constructor(){ this.init(); } init(){ this.tubes = []; for(let i = 0; i < nColums; i++){ this.tubes.push(new Tube(canvas.width / divisions * (i * (divisions / 2 - 1)) + offset, 0, canvas.width / divisions, Math.floor(Math.random() * (canvas.height - opening - minOpen - minOpenSosia)) + minOpen)); } } update(){ for(let i = 0; i < this.tubes.length; i++){ this.tubes[i].update(); } } render(){ for(let i = 0; i < this.tubes.length; i++){ this.tubes[i].render(); } } } class Bird{ constructor(brain = undefined){ this.fitness = 0; this.x = canvas.width / 5; this.y = canvas.height / 2; this.radius = 7; this.isJumping = false; this.velocity = 0; this.nearestTube = 0; this.isDead = false; this.score = 0; this.gravity = gravity; if(brain){ this.brain = brain; }else{ this.brain = new NeuralNetwork(4, 8, 2); } } update(tubes){ if(this.isDead) return; //al massimo togli se rallenta// let tube = tubes.tubes[this.findFirstTube(tubes)]; let output = this.brain.predict([tube.x / canvas.width, tube.height / canvas.height, (tube.height + opening) / canvas.height, this.y / canvas.height]); if(output.data[0][0] > 0.5){ this.velocity -= 11; } this.gravity += output.data[1][0]; this.velocity += this.gravity * delta; //clamp value of velocity if(this.velocity > 5){ this.velocity = 5; }else if(this.velocity < -8){ this.velocity = -8; } this.y += this.velocity; //collisioni for(let i = 0; i < tubes.tubes.length; i++){ let tube = tubes.tubes[i]; if((this.x + this.radius > tube.x && this.x - this.radius < tube.x + tube.width) && !(this.y - this.radius > tube.height && this.y + this.radius < tube.height + opening) || (this.y + this.radius > canvas.height) || (this.y - this.radius < 0)){ //this.y = canvas.height / 2; this.nearestTube = 0; this.isDead = true; this.score = score; } } } findFirstTube(tubes){ if(this.x - this.radius > tubes.tubes[this.nearestTube].x + tubes.tubes[this.nearestTube].width){ this.nearestTube++; visualScore++; } this.nearestTube %= tubes.tubes.length; return this.nearestTube; } render(){ if(this.isDead) return; ctx.beginPath(); ctx.lineWidth = 1; ctx.fillStyle = "rgba(255, 0, 0, 0.5)"; ctx.strokeStyle = "black"; ctx.arc(this.x, this.y, this.radius, 0, 2 * Math.PI); ctx.stroke(); ctx.fill(); ctx.closePath(); } calculateFitness(totalScore){ this.fitness = this.score / totalScore; return this.score / totalScore; } mutate(rate){ this.brain.mutate(rate); } } function random(min, max){ return (Math.random() * (max - min)) + min; } class Population{ constructor(size){ this.size = size; this.mutationRate = 0.01; this.birds = []; for(let i = 0; i < this.size; i++){ this.birds.push(new Bird()); } } update(tubes){ for(let i = 0; i < this.birds.length; i++){ this.birds[i].update(tubes); } if(this.isAllDead()){ if(visualScore > bestScore) bestScore = visualScore; let totalFitness = 0; let totalScore = this.calculateTotalScore(); //calculate fitness for(let i = 0; i < this.birds.length; i++){ totalFitness += this.birds[i].calculateFitness(totalScore); } let newBirds = []; for(let i = 0; i < this.size; i++){ let mother = this.pickOne(this.birds); let father = this.pickOne(this.birds); let child = this.crossover(mother.brain, father.brain); //slider.value = 0; child.mutate(this.mutationRate); newBirds.push(child); } this.birds = newBirds; tubes.init(); score = 0; visualScore = 0; generation++; generationCounter.innerHTML = "Generation: " + generation; bestScoreCounter.innerHTML = "Best Score : " + bestScore; }else{ score += 1; } } render(){ for(let i = 0; i < this.birds.length; i++){ this.birds[i].render(); } } isAllDead(){ let counter = 0; for(let i = 0; i < this.birds.length; i++){ if(this.birds[i].isDead){ counter++; } } if(counter === this.birds.length) return true; return false; } crossover(motherBrain, fatherBrain){ let childBrain = new NeuralNetwork(4, 8, 2); for(let i = 0; i < motherBrain.pesiInputHidden.rows; i++){ for(let j = 0; j < motherBrain.pesiInputHidden.columns; j++){ if(random(0, 100) > 50){ childBrain.pesiInputHidden.data[i][j] = fatherBrain.pesiInputHidden.data[i][j]; }else{ childBrain.pesiInputHidden.data[i][j] = motherBrain.pesiInputHidden.data[i][j]; } } } for(let i = 0; i < motherBrain.pesiHiddenOutput.rows; i++){ for(let j = 0; j < motherBrain.pesiHiddenOutput.columns; j++){ if(random(0, 100) > 50){ childBrain.pesiHiddenOutput.data[i][j] = fatherBrain.pesiHiddenOutput.data[i][j]; }else{ childBrain.pesiHiddenOutput.data[i][j] = motherBrain.pesiHiddenOutput.data[i][j]; } } } let child = new Bird(childBrain); return child; } calculateTotalScore(){ let score = 0; for(let i = 0; i < this.birds.length; i++){ score += this.birds[i].score; } return score; } pickOne(birds){ let index = 0; let r = Math.random(); while(r > 0){ r = r - birds[index].fitness; index++; } index--; return birds[index]; } } var tubes; var population; var slider = document.getElementById("myRange"); var generationCounter = document.getElementById("generation"); var bestScoreCounter = document.getElementById("bestScore"); function init(){ tubes = new Tubes(); population = new Population(populationSize); } function update(){ population.update(tubes); tubes.update(); } function render(){ ctx.beginPath(); ctx.fillStyle = "blue"; ctx.fillRect(0, 0, canvas.width, canvas.height); tubes.render(); population.render(); ctx.font = "25px Arial"; ctx.fillStyle = "black"; ctx.fillText("Score: " + visualScore , 10, 25); ctx.closePath(); } function gameLoop(){ for(let i = 0; i < slider.value; i++){ update(); } render(); window.requestAnimationFrame(gameLoop); } init(); window.requestAnimationFrame(gameLoop);

AI Generation Learns Flappy Birds