Experimenting with policy gradient methods in Jax