Maybe this is a very silly question, but How you should pose (if possible) in PyMC3 a bayesian search problem (like described here). How to update the model with each new interaction? (There is a javascript app that is incorrect, but just to give you an idea)