Add Avalon.jl MLP Model

[trevor-yu-087]

No description provided.

[trevor-yu-087]

Closes #5

[trevor-yu-087]

Notebook ready for review:
Known issues: Learning rate is 1e-3 instead of 1e-4. I think I might change the spec to 1e-3, it's more reasonable for getting better loss on this smaller model.

Avalon.jl doesn't work in its current state so relevant primitives are copied over from repo.

[trevor-yu-087]

Script finished, ready for review

[UrbanPistek]

Not really familar with Julia but in general my main comments are around just structure and grouping things together to be a bit more clear what everything is doing.

For example the function Linear, Net struct and then the creation of the net object - I assume these are all for initializing the model structure, I think simply grouping them together, adding a bit more comments or even making a class for the model itself just makes it clear what these are doing, the others are pretty self explainitory.

The other example would be creating a train function instead of just having it all in main().

Nothing major but these small structure changes I think would help with understading what each components is doing especially if you are unfamilar woth Julia.

I'll leave it up to our discrestion whether you want to make the changes here or just keep it in mind going forward.

[UrbanPistek]

Another side note, again I am not familar with the Julia ecosystem but what about enviroments and dependancies? At least there should be something sepcifying dependancies and the version used here.