Instantiating Prior class with empirical distribution

[meraldoantonio]

Is there a way to instantiate the Prior class using an empirical distribution, such as by passing a NumPy array?

The goal is to construct a Bayesian class that supports performing iterative Bayesian updates, where the posterior distribution from one inference can be used as the prior for the next (i.e. continuous learning).

[wd60622]

I don't think it is supported at the moment.

The Interpolated class would likely be route to implementing. Maybe in the way it is used here.

Would you want to experiment with it and give an implementation a try?

Related to #782

[wd60622]

Hi @meraldoantonio

We've recently made an update which allows anything that looks like the VariableFactory class to be used in various places. For instance, as MMM parameters

On top of the dims and create_variable, having a to_dict and registered deserialization strategy should make this possible.

Give it a try and let us know how it goes!