FlappyBird-NeuroEvolution

Click (here) to play the Demo

Using Machine Learning to play Flappy Bird, use NEAT algorithm from Carrot.js

Incorporating Carrot.js

Initialize NEAT

//Initalize variables for neat 
let populate = 50;
let GAMES = 70;
let elitism = Math.round(0.2 * GAMES);
let rate = 1.5;
let amount = 8.5; 

//Initalize Neat itself 
const neat = new Neat(5, 2, {
  population: populate,
  elitism: elitism,
  mutation_rate: rate,
  mutation_amount: amount,
  equal: false
})

Create Population Method

function populating() {
 neat.population = neat.population.map(function(genome) { 
// grab a random mutation method
const random_mutation_method = methods.mutation.FFW[Math.floor(Math.random() * methods.mutation.FFW.length)]
// mutate the genome
genome.mutate(random_mutation_method)
// return the mutated genome
  return genome
})
  
//populate the mutated genomes into the active birds array 
for (let i = 0; i < neat.population.length; i++) {
  activeBirds.push(new Bird(neat.population[i]));
}
//Counter increment for New Generation
 countGen++;
}

Start a New Generation

// If there is no more birds start the next generation
 if (activeBirds.length == 0) {
 //sort the population by score, highest to lowest
     neat.sort(); 
 //new array to push for new generation
  const newGeneration = [];
 // gets the best of previous generation and inserts them into the next population
     for (let i = 0; i < elitism; i++) {
        newGeneration.push(neat.population[i]);
    }
    // test to see if parent gets returned
    for (let i = 0; i < POP - elitism; i++) {
      newGeneration.push(neat.getOffspring());
    }
     neat.population = newGeneration;
     populating(); 
    }
  }

Thanks to the Coding Train for the structure of the flappy bird game, the original version of the repo: https://github.com/CodingTrain/Toy-Neural-Network-JS

Thanks to xviniette for inspiration of the machine learning flappy bird demo in js, famous FlappyLearning repo: https://github.com/xviniette/FlappyLearning