TODO: try ADAM or something?

TODO: try ADAM, play with parameters? (smaller network?)

Is there a convenient way to do this without programming it up for all common use cases? Otherwise, will have to do really ugly and inefficient reshaping

I've tried a couple of things and I can't quite get it to work right.

In test/test_ddpg.jl, I've tried importing Base.convert, and defining convert(::type{A}, Vector{Real}), and importing Base.convert in DRL.rl as well. I've also tried a couple of different loading orders.

It would be great to be able to switch between deep learning backends, so users aren't tied to mxnet. The three that come to mind are mxnet, tensorflow, and knet. This could also simplify model building.

Something like this is probably a big endeavor, but I don't think we need a complete framework like Keras to make it work.

It would be good to start by abstracting away the deep learning building, backprop, etc to something that can serve as a common interface for multiple backends.

The question is, should something like this have its own package?