Machine Learning With Neat: February 9, 2024

I continued working on my AI pong project with the help of the video above.

I first made some minor changes to the run_neat function, then added a function that creates the game to evaluate the AI’s fitness level. The function should look like this:

def eval_genomes(genomes, config):
    width, height = 700, 500
    window = pygame.display.set_mode((width, height))


    for i, (genome_id1, genome1) in enumerate(genomes):
        if i == len(genomes) - 1:
            break
        genome1.fitness = 0
        for genome_id2, genome2 in genomes[i+1:]:
            genome2.fitness = 0 if genome2.fitness ==  None else genome2.fitness
            game = PongGame(window, width, height)
            game.train_ai(genome1, genome2, config)




def run_neat(config):
    #p = neat.Checkpointer.restore_checkpoint('neat-checkpoint-12')
    p = neat.Population(config)
    p.add_reporter(neat.StdOutReporter(True))
    stats = neat.StatisticsReporter
    p.add_reporter(stats)
    p.add_reporter(neat.Checkpointer(1))


    winner = p.run(eval_genomes, 50)

Next, I created the train_ai function, which creates the two neural networks and begins to train them against each other. I also created the calculate_fitness function, but I did not have enough time to write the rest of it. The train_ai function look like this:

    def train_ai(self, genome1, genome2, config):
        net1 = neat.nn.FeedForwardNetwork.create(genome1, config)
        net2 = neat.nn.FeedForwardNetwork.create(genome2, config)


        run = True
        while run:
            for event in pygame.event.get():
                if event.type == pygame.QUIT:
                    quit()


            output1 = net1.activate((self.left_paddle.y, self.ball.y, abs(self.left_paddle.x - self.ball.x)))
            output2 = net2.activate((self.right_paddle.y, self.ball.y, abs(self.right_paddle.x - self.ball.x)))


            game_info = self.game.loop()
            self.game.draw()
            pygame.display.update()


            if game_info.left_score >= 1 or game_info.right_score >= 1:
                self.calculate_fitness(genome1, genome2, game_info)
                break


def calculate_fitness(self, genome1, genome2, game_info):
    pass