Behrens' Bayesian Learner Model : How to share parameters between steps?

maximum recursion depth exceeded while calling a Python object

So I think the for loop way may not a good way to archive my model.

The for loop is definitely sub-optimal on the PyMC3 side (too many variables to keep track of), but the error you are reporting seems to come from a Python issue/error in your code, not from a PyMC3 limitation.

Nevermind, found something like this here python - "Bracket nesting level exceeded maximum" when working with large dataset in PyMC3 - Stack Overflow

I had no idea about this sort of limitation on theano