https://towardsdatascience.com/deepmind-just-released-haiku-and-rlax-for-neural-networks-and-reinforcement-learning-a6468f9352cc