factor out agents?

[KiaraGrouwstra]

Looking at a few RL libs, I noticed some had their agents as the main program, while e.g. the DAT257x notebooks expressed agents as classes, implementing methods to give an action as well as to learn from the environment feedback.

It seemed to me like over here agents were tightly coupled with the main program as well, but was wondering if some Haskell equivalent to the Python class would be nice.
I'm still new-ish to Haskell, but had been thinking in the direction of some type class with state monad covering the state (agent-specific) and act/learn methods.

Mostly just throwing this out there to get a sense of what the considerations on this were over here. :)

[stites]

That's probably the right way forward (and was starting to be the direction of the algorithms repo).

I was a bit afraid of over-abstracting before introducing function approximators, so I think I left this project thinking that what was really needed was more raw, unabstracted examples in the reinforce-zoo. The idea being that if there are enough examples a more natural abstraction should fall out of the woodwork.