The Stan community has a good way of generating fake data from a model. Here’s a reference: https://modernstatisticalworkflow.blogspot.com/2017/04/an-easy-way-to-simulate-fake-data-from.html
Where it seems that in Stan we can specify if the likelihood is actually evaluated. If it isn’t, the posterior predictive draws reduce to prior predictive draws.
This is useful if we want to determine if a model can recover parameters from data generated by itself.
Is it possible to do something similar in pymc3?