erindb/causal_simulations

implement in pyro?

Opened this issue · 2 comments

If we ever want to learn meaning functions for the explanans and explanandum someday, it would be nice to have a pyro version, right? And tobi's physics is written in python, so maybe we should just use the same physics engine? But learning meaning functions is a long way off, I'm not sure about the relative speeds or inference options, and we already have a version in javascript. Allen thinks it would be fun to implement this, which is awesome! Does that seem like a good plan?

Pros:

  • interfaces nicely with python
  • easy to simplify stickiness as linear combination of continuous functions
  • great if we end up with big tensors someday (we don't need them now, though)

Cons:

  • We can't differentiate the physics simulator.
  • There's not really a good enumerate

I can't remember the other things we talked about...

Actually, those cons aren't really cons.

  • if the physics simulator is written in tensor operations, we'd be fine. and it probably could be, if it's not already.
  • pyro has parallel enumeration

Currently talking with Eli about issues with this.