How to model continuous data?

IMO this is best to fit with a Bayesian State Space model, which unfortunately PyMC3 does not have.
Another option is to use a Gaussian process, in a way, the continuous you wanted to capture is encoded in the covariance matrix/kernel.