simple-evo's Introduction

ML-chase

Chickens hunted by Foxes, powered by Machine Learning

How does it work

Made with ML5.js and p5.js, this is a basic Neural Net evolved by picking the two best specimens of each group, crossover and mutation.

Very Basic functionality,

Neural Net Inputs:
x position
y position
distance from nearest chicken
Neural Net Outputs:
directions(left, right, up, down)
diagnol directions
added penalties for clump, collision randomness

The main parts of the code are:

The Neural net, the actual brain, a classifier type

 constructor(x, y) {
  this.brain = ml5.neuralNetwork({
 inputs: ["xpos", "ypos", "clumpDist"],
      outputs: [
        "left",
        "right",
        "up",
        "down",
        "diagUpLeft",
        "diagUpRight",
        "diagDownLeft",
        "diagDownRight",
      ],
      task: "classification",
      noTraining: true,
    });
    this.brain.mutate(1);
    ...
   }

The actual move choice classification

  classification = this.brain.classify({
       xpos: this.x,
       ypos: this.y,
       clumpDist: this.clumpDist,
     });

The Generation Choice

function generate() {
  // chickens
  chickens = [];
  gen++;
  dead_chickens.sort((a, b) => {
    return b.score - a.score;
  });
  let a_chick = dead_chickens[0]; // best chicken
  let b_chick = dead_chickens[1]; // second best chicken
  let c_chick = a_chick.brain.crossover(b_chick.brain);

  for (let i = 0; i < 40; i++) {
    c_chick.mutate(0.05);
    let temp_chick = new Chicken(
      randInt(0, window.innerWidth * 0.8),
      randInt(0, 600)
    );
    temp_chick.brain = c_chick;
    chickens.push(temp_chick);
  }
  // foxes
  gen += 0;
  foxes.sort((a, b) => {
    return b.score - a.score;
  });
  let a_foxes = foxes[0]; // best chicken
  let b_foxes = foxes[1]; // second best chicken
  let c_foxes = a_foxes.brain.crossover(b_foxes.brain);

  foxes = [];
  for (let i = 0; i < 20; i++) {
    c_foxes.mutate(0.05);
    let temp_foxes = new Fox(
      randInt(0, window.innerWidth * 0.8),
      randInt(0, 600)
    );
    temp_foxes.brain = c_foxes;
    foxes.push(temp_foxes);
  }
}

Recommend Projects