I have a question regarding inference on out of sample. Let’s say I have got a trace using training data and I have some test data with U variables known. Using these U variables, I want to have predictions of \lambda. For this, I was thinking of using predictive posterior sampling, but that won’t work since X hidden states need to be estimated again for each new sample. What would be the best way to have an estimate of \lambda for a new sample U? How would sampling work in this case? Probably I can post this as a separate question.