Hamiltonian Monte Carlo (HMC) for HMM example

[slinderman]

SSM's hidden Markov model (HMM) objects expose a function to compute the marginal likelihood of the data, summing over the discrete latent states. This function can be automatically differentiated with jax.grad. Use Tensorflow Probability's Hamiltonian Monte Carlo (HMC) functionality to perform Bayesian inference over HMM parameters, using the marginal likelihood and a prior on parameter values.