Hands-On Neuroevolution with Python

The game step function

The game step function executes a single game step using the provided actions. It is implemented as follows:

    def step(self, action, indices=None, name=None):        if indices is None:            indices = np.arange(self.batch_size)        with tf.variable_scope(name, default_name='AtariStep'):            rew, done = gym_tensorflow_module.environment_step(                               self.instances, indices, action)            return rew, done

This function applies the game action that's received from the controller ANN to the current game environment. Please note that this function can execute a single game step simultaneously in multiple game instances. The self.batch_size parameter or the length of the indices input tensor determines the number of game instances we'll have. The function ...