Examples for Recovering Latent States from Continuous HMM

As a novice in pyro(and probabilistic programming in general), I’m running into issues converting the HMM example to train on continuous data and export the discrete latent path(viterbi). I’ve looked into infer_discrete and similar examples with little success.

That being said, are there any good resources for this? I know #1802 talks about this issue but it doesnt appear to have been resolved.

Hi @swwolf,

I believe infer_discrete is your best option. We do have some examples of infer_discrete, including a quite complex example (sir_hmc.py).

In this afternoon’s Pyro team meeting we discussed adding more prediction code to our examples, so that each example demonstrated both inference and prediction. Maybe instead of a dedicated infer_discrete tutorial, we could address #1802 by simply adding to the HMM tutorial?