[hydra-configs-torch] Configs for Linear / Conv Modules

[romesco]

Generate these module confs, brainstorm robust testing method.

[omry]

I think this is probably too challenging for an initial example. We can have the network defined in code (as it is now) and instantiate it with instantiate.

If you want to break new ground with allowing full control of the network initially, you can probably try it out first "on the side". We should probably not add config classes to things before we are actually sure these are useful (initially these can be a part of the test app, generated by configen or by hand).

[romesco]

Which config classes are you suggesting might not be useful? I'm thinking many of the nn/modules ones would be, right?

[romesco]

Or did you mean for that comment to be in the context of the MNIST example (which there is a separate issue for tracking =P).

[omry]

optimizer, dataset, model (at a high level). See the PL Hydra PR.