zero code

zero.py

    environment = environments.GoEnvironment(board_size=19)
    player_mcts = mcts.MCTS(
        environment,
        networks.NeuralNetwork(board_size=environment.getStateSize(), action_size=environment.getActionSize()),
    )

    training_samples = collections.deque(maxlen=opt.training_samples_buffer_size)
    
    for iteration_number in range(opt.num_iters):
        for eps in range(opt.number_episodes):
            player_mcts.reset()
            new_experiences = play_against_yourself(environment, player_mcts, nr_games=1)
            training_samples.extend(new_experiences)

        opponent_mcts = mcts.MCTS(environment, copy.deepcopy(player_neural_network))
        
        player_neural_network.train(training_samples)

        player_wins, player_losses, draws = arena.fight(environment, player_mcts, opponent_mcts)

        win_ratio = float(player_wins) / (player_wins + player_losses)

        if win_ratio < opt.update_threshold:
            player_neural_network = opponent_network