Add support for neural networks about rsrl HOT 2 OPEN

What do you mean by "support for NNs"? Do you mean that current algorithms are matrix or table-based?

from rsrl.

The algorithms, for the most part, are written to be generic and implementation-independent of the function approximator you choose to use. The only real exceptions are those that specifically require linear-in-the-features fa.

So if you take a look at the fa module you'll see a few traits (like VFunction and QFunction) which are interfaces to function approximators used in the algorithms. These abstract away the underlying technique you're using and provide a single interface to updating and evaluation etc.

The issue I have at the moment is just finding the time to implement these for external crates that implement NNs. Further, as this crate is still in alpha, these are open to major changes and it can be tedious to do all the re-implementations.

Cheers,
Tom

from rsrl.