Spaces:
Running
Running
First commit
Browse files- background.png +0 -0
- bird.js +77 -0
- ga.js +37 -0
- index.html +17 -22
- nn.js +85 -0
- pipe.js +40 -0
- screenshots/nn.png +0 -0
- screenshots/sc1.png +0 -0
- screenshots/sc2.png +0 -0
- sketch.js +101 -0
background.png
ADDED
bird.js
ADDED
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
// Neuro-Evolution Flappy Bird with TensorFlow.js
|
2 |
+
// http://thecodingtrain.com
|
3 |
+
// https://youtu.be/cdUNkwXx-I4
|
4 |
+
|
5 |
+
class Bird {
|
6 |
+
constructor(brain) {
|
7 |
+
this.y = height / 2;
|
8 |
+
this.x = 64;
|
9 |
+
|
10 |
+
this.gravity = 0.8;
|
11 |
+
this.lift = -12;
|
12 |
+
this.velocity = 0;
|
13 |
+
|
14 |
+
this.score = 0;
|
15 |
+
this.fitness = 0;
|
16 |
+
if (brain) {
|
17 |
+
this.brain = brain.copy();
|
18 |
+
} else {
|
19 |
+
this.brain = new NeuralNetwork(5, 8, 2);
|
20 |
+
}
|
21 |
+
}
|
22 |
+
|
23 |
+
dispose() {
|
24 |
+
this.brain.dispose();
|
25 |
+
}
|
26 |
+
|
27 |
+
show() {
|
28 |
+
stroke(255);
|
29 |
+
fill(251, 236, 93);
|
30 |
+
ellipse(this.x, this.y, 32, 32);
|
31 |
+
}
|
32 |
+
|
33 |
+
up() {
|
34 |
+
this.velocity += this.lift;
|
35 |
+
}
|
36 |
+
|
37 |
+
mutate() {
|
38 |
+
this.brain.mutate(0.1);
|
39 |
+
}
|
40 |
+
|
41 |
+
think(pipes) {
|
42 |
+
// Find the closest pipe
|
43 |
+
let closest = null;
|
44 |
+
let closestD = Infinity;
|
45 |
+
for (let i = 0; i < pipes.length; i++) {
|
46 |
+
let d = pipes[i].x + pipes[i].w - this.x;
|
47 |
+
if (d < closestD && d > 0) {
|
48 |
+
closest = pipes[i];
|
49 |
+
closestD = d;
|
50 |
+
}
|
51 |
+
}
|
52 |
+
|
53 |
+
let inputs = [];
|
54 |
+
inputs[0] = this.y / height;
|
55 |
+
inputs[1] = closest.top / height;
|
56 |
+
inputs[2] = closest.bottom / height;
|
57 |
+
inputs[3] = closest.x / width;
|
58 |
+
inputs[4] = this.velocity / 10;
|
59 |
+
let output = this.brain.predict(inputs);
|
60 |
+
//if (output[0] > output[1] && this.velocity >= 0) {
|
61 |
+
if (output[0] > output[1]) {
|
62 |
+
this.up();
|
63 |
+
}
|
64 |
+
}
|
65 |
+
|
66 |
+
offScreen() {
|
67 |
+
return this.y > height || this.y < 0;
|
68 |
+
}
|
69 |
+
|
70 |
+
update() {
|
71 |
+
this.score++;
|
72 |
+
|
73 |
+
this.velocity += this.gravity;
|
74 |
+
//this.velocity *= 0.9;
|
75 |
+
this.y += this.velocity;
|
76 |
+
}
|
77 |
+
}
|
ga.js
ADDED
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
// Neuro-Evolution Flappy Bird
|
2 |
+
|
3 |
+
function nextGeneration() {
|
4 |
+
console.log("next generation");
|
5 |
+
calculateFitness();
|
6 |
+
for (let i = 0; i < TOTAL; i++) {
|
7 |
+
birds[i] = pickOne();
|
8 |
+
}
|
9 |
+
for (let i = 0; i < TOTAL; i++) {
|
10 |
+
savedBirds[i].dispose();
|
11 |
+
}
|
12 |
+
savedBirds = [];
|
13 |
+
}
|
14 |
+
|
15 |
+
function pickOne() {
|
16 |
+
let index = 0;
|
17 |
+
let r = random(1);
|
18 |
+
while (r > 0) {
|
19 |
+
r = r - savedBirds[index].fitness;
|
20 |
+
index++;
|
21 |
+
}
|
22 |
+
index--;
|
23 |
+
let bird = savedBirds[index];
|
24 |
+
let child = new Bird(bird.brain);
|
25 |
+
child.mutate();
|
26 |
+
return child;
|
27 |
+
}
|
28 |
+
|
29 |
+
function calculateFitness() {
|
30 |
+
let sum = 0;
|
31 |
+
for (let bird of savedBirds) {
|
32 |
+
sum += bird.score;
|
33 |
+
}
|
34 |
+
for (let bird of savedBirds) {
|
35 |
+
bird.fitness = bird.score / sum;
|
36 |
+
}
|
37 |
+
}
|
index.html
CHANGED
@@ -1,24 +1,19 @@
|
|
1 |
<!DOCTYPE html>
|
2 |
-
<html>
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
>Spaces documentation</a
|
20 |
-
>.
|
21 |
-
</p>
|
22 |
-
</div>
|
23 |
-
</body>
|
24 |
</html>
|
|
|
1 |
<!DOCTYPE html>
|
2 |
+
<html lang="en">
|
3 |
+
<head>
|
4 |
+
<meta charset="UTF-8" />
|
5 |
+
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
6 |
+
<meta http-equiv="X-UA-Compatible" content="ie=edge" />
|
7 |
+
<script src="https://cdnjs.cloudflare.com/ajax/libs/p5.js/0.8.0/p5.min.js"></script>
|
8 |
+
<script src="https://cdnjs.cloudflare.com/ajax/libs/p5.js/0.8.0/addons/p5.dom.min.js"></script>
|
9 |
+
<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@1.0.4/dist/tf.min.js"></script>
|
10 |
+
<title>NeuroEvolution with tf.js</title>
|
11 |
+
</head>
|
12 |
+
<body>
|
13 |
+
<script src="nn.js"></script>
|
14 |
+
<script src="bird.js"></script>
|
15 |
+
<script src="pipe.js"></script>
|
16 |
+
<script src="ga.js"></script>
|
17 |
+
<script src="sketch.js"></script>
|
18 |
+
</body>
|
|
|
|
|
|
|
|
|
|
|
19 |
</html>
|
nn.js
ADDED
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
// Neuro-Evolution Flappy Bird with TensorFlow.js
|
2 |
+
|
3 |
+
class NeuralNetwork {
|
4 |
+
constructor(a, b, c, d) {
|
5 |
+
if (a instanceof tf.Sequential) {
|
6 |
+
this.model = a;
|
7 |
+
this.input_nodes = b;
|
8 |
+
this.hidden_nodes = c;
|
9 |
+
this.output_nodes = d;
|
10 |
+
} else {
|
11 |
+
this.input_nodes = a;
|
12 |
+
this.hidden_nodes = b;
|
13 |
+
this.output_nodes = c;
|
14 |
+
this.model = this.createModel();
|
15 |
+
}
|
16 |
+
}
|
17 |
+
|
18 |
+
copy() {
|
19 |
+
return tf.tidy(() => {
|
20 |
+
const modelCopy = this.createModel();
|
21 |
+
const weights = this.model.getWeights();
|
22 |
+
const weightCopies = [];
|
23 |
+
for (let i = 0; i < weights.length; i++) {
|
24 |
+
weightCopies[i] = weights[i].clone();
|
25 |
+
}
|
26 |
+
modelCopy.setWeights(weightCopies);
|
27 |
+
return new NeuralNetwork(
|
28 |
+
modelCopy,
|
29 |
+
this.input_nodes,
|
30 |
+
this.hidden_nodes,
|
31 |
+
this.output_nodes
|
32 |
+
);
|
33 |
+
});
|
34 |
+
}
|
35 |
+
|
36 |
+
mutate(rate) {
|
37 |
+
tf.tidy(() => {
|
38 |
+
const weights = this.model.getWeights();
|
39 |
+
const mutatedWeights = [];
|
40 |
+
for (let i = 0; i < weights.length; i++) {
|
41 |
+
let tensor = weights[i];
|
42 |
+
let shape = weights[i].shape;
|
43 |
+
let values = tensor.dataSync().slice();
|
44 |
+
for (let j = 0; j < values.length; j++) {
|
45 |
+
if (random(1) < rate) {
|
46 |
+
let w = values[j];
|
47 |
+
values[j] = w + randomGaussian();
|
48 |
+
}
|
49 |
+
}
|
50 |
+
let newTensor = tf.tensor(values, shape);
|
51 |
+
mutatedWeights[i] = newTensor;
|
52 |
+
}
|
53 |
+
this.model.setWeights(mutatedWeights);
|
54 |
+
});
|
55 |
+
}
|
56 |
+
|
57 |
+
dispose() {
|
58 |
+
this.model.dispose();
|
59 |
+
}
|
60 |
+
|
61 |
+
predict(inputs) {
|
62 |
+
return tf.tidy(() => {
|
63 |
+
const xs = tf.tensor2d([inputs]);
|
64 |
+
const ys = this.model.predict(xs);
|
65 |
+
const outputs = ys.dataSync();
|
66 |
+
return outputs;
|
67 |
+
});
|
68 |
+
}
|
69 |
+
|
70 |
+
createModel() {
|
71 |
+
const model = tf.sequential();
|
72 |
+
const hidden = tf.layers.dense({
|
73 |
+
units: this.hidden_nodes,
|
74 |
+
inputShape: [this.input_nodes],
|
75 |
+
activation: "sigmoid"
|
76 |
+
});
|
77 |
+
model.add(hidden);
|
78 |
+
const output = tf.layers.dense({
|
79 |
+
units: this.output_nodes,
|
80 |
+
activation: "softmax"
|
81 |
+
});
|
82 |
+
model.add(output);
|
83 |
+
return model;
|
84 |
+
}
|
85 |
+
}
|
pipe.js
ADDED
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
// Neuro-Evolution Flappy Bird with TensorFlow.js
|
2 |
+
|
3 |
+
class Pipe {
|
4 |
+
constructor() {
|
5 |
+
this.spacing = 125;
|
6 |
+
this.top = random(height / 6, (3 / 4) * height);
|
7 |
+
this.bottom = height - (this.top + this.spacing);
|
8 |
+
this.x = width;
|
9 |
+
this.w = 80;
|
10 |
+
this.speed = 6;
|
11 |
+
}
|
12 |
+
|
13 |
+
hits(bird) {
|
14 |
+
if (bird.y < this.top || bird.y > height - this.bottom) {
|
15 |
+
if (bird.x > this.x && bird.x < this.x + this.w) {
|
16 |
+
return true;
|
17 |
+
}
|
18 |
+
}
|
19 |
+
return false;
|
20 |
+
}
|
21 |
+
|
22 |
+
show() {
|
23 |
+
fill(75, 127, 83);
|
24 |
+
rectMode(CORNER);
|
25 |
+
rect(this.x, 0, this.w, this.top);
|
26 |
+
rect(this.x, height - this.bottom, this.w, this.bottom);
|
27 |
+
}
|
28 |
+
|
29 |
+
update() {
|
30 |
+
this.x -= this.speed;
|
31 |
+
}
|
32 |
+
|
33 |
+
offscreen() {
|
34 |
+
if (this.x < -this.w) {
|
35 |
+
return true;
|
36 |
+
} else {
|
37 |
+
return false;
|
38 |
+
}
|
39 |
+
}
|
40 |
+
}
|
screenshots/nn.png
ADDED
screenshots/sc1.png
ADDED
screenshots/sc2.png
ADDED
sketch.js
ADDED
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
// Daniel Shiffman & Manuel Romero
|
2 |
+
// Neuro-Evolution Flappy Bird with TensorFlow.js
|
3 |
+
// http://thecodingtrain.com
|
4 |
+
// https://youtu.be/cdUNkwXx-I4
|
5 |
+
|
6 |
+
const TOTAL = 250;
|
7 |
+
let birds = [];
|
8 |
+
let savedBirds = [];
|
9 |
+
let pipes = [];
|
10 |
+
let counter = 0;
|
11 |
+
|
12 |
+
let bg;
|
13 |
+
let slider;
|
14 |
+
let displayGeneration;
|
15 |
+
let displaySpeed;
|
16 |
+
let generationNumber = 1;
|
17 |
+
|
18 |
+
function keyPressed() {
|
19 |
+
if (key === "S" || key === "s") {
|
20 |
+
let bird = birds[0];
|
21 |
+
saveJSON(bird.brain, "bird.json");
|
22 |
+
}
|
23 |
+
}
|
24 |
+
|
25 |
+
function setup() {
|
26 |
+
bg = loadImage("background.png");
|
27 |
+
createCanvas(640, 480);
|
28 |
+
displayGeneration = createP("Generation");
|
29 |
+
displaySpeed = createP("Speed");
|
30 |
+
slider = createSlider(1, 10, 1);
|
31 |
+
for (let i = 0; i < TOTAL; i++) {
|
32 |
+
birds[i] = new Bird();
|
33 |
+
}
|
34 |
+
|
35 |
+
tf.setBackend('cpu');
|
36 |
+
}
|
37 |
+
|
38 |
+
function draw() {
|
39 |
+
for (let n = 0; n < slider.value(); n++) {
|
40 |
+
if (counter % 75 == 0) {
|
41 |
+
pipes.push(new Pipe());
|
42 |
+
}
|
43 |
+
counter++;
|
44 |
+
|
45 |
+
for (let i = pipes.length - 1; i >= 0; i--) {
|
46 |
+
pipes[i].update();
|
47 |
+
|
48 |
+
for (let j = birds.length - 1; j >= 0; j--) {
|
49 |
+
if (pipes[i].hits(birds[j])) {
|
50 |
+
savedBirds.push(birds.splice(j, 1)[0]);
|
51 |
+
}
|
52 |
+
}
|
53 |
+
|
54 |
+
if (pipes[i].offscreen()) {
|
55 |
+
pipes.splice(i, 1);
|
56 |
+
}
|
57 |
+
}
|
58 |
+
|
59 |
+
for (let i = birds.length - 1; i >= 0; i--) {
|
60 |
+
if (birds[i].offScreen()) {
|
61 |
+
savedBirds.push(birds.splice(i, 1)[0]);
|
62 |
+
}
|
63 |
+
}
|
64 |
+
|
65 |
+
for (let bird of birds) {
|
66 |
+
bird.think(pipes);
|
67 |
+
bird.update();
|
68 |
+
}
|
69 |
+
|
70 |
+
if (birds.length === 0) {
|
71 |
+
counter = 0;
|
72 |
+
generationNumber++;
|
73 |
+
nextGeneration();
|
74 |
+
pipes = [];
|
75 |
+
}
|
76 |
+
}
|
77 |
+
|
78 |
+
// All the drawing stuff
|
79 |
+
background(bg);
|
80 |
+
|
81 |
+
displayGeneration.html(
|
82 |
+
`Generation Number: <strong>${generationNumber}</strong>`
|
83 |
+
);
|
84 |
+
|
85 |
+
displaySpeed.html(`Speed:`);
|
86 |
+
|
87 |
+
for (let bird of birds) {
|
88 |
+
bird.show();
|
89 |
+
}
|
90 |
+
|
91 |
+
for (let pipe of pipes) {
|
92 |
+
pipe.show();
|
93 |
+
}
|
94 |
+
}
|
95 |
+
|
96 |
+
// function keyPressed() {
|
97 |
+
// if (key == ' ') {
|
98 |
+
// bird.up();
|
99 |
+
// //console.log("SPACE");
|
100 |
+
// }
|
101 |
+
// }
|