Architecture flowchart

[swo]

• [x] Add architecture diagram to documents
• [ ] Clarify if this reflects current reality or hoped-for reality
• [ ] Figure out if we should name different components (eg with linmod.whatever classes)

Resolves #52

cc @afmagee42 @thanasibakis

[afmagee42]

We're missing a link from "clean data" to "evaluator."

[swo]

We're missing a link from "clean data" to "evaluator."

good catch, I added eval data as a separate stream, separate from the training data

[afmagee42]

Clarify if this reflects current reality or hoped-for reality

This is close to reality.

• FWIW I don't think we currently have any "common pre-processing."
• The "evaluator" is currently a few lines of code in main.py, which is fine, but the convergence diagnostic stuff is also a few lines in main.py so it feels a bit off balance?
• The links to the visualization aren't quite right. We don't visualize the scores, we take the forked clean eval data (good call on separating that out) and combine it directly with the model outputs to make the eval viz.

Figure out if we should name different components (eg with linmod.whatever classes)

Probably best not to at this point, since the boxes make conceptual sense but don't correspond neatly in the same ways to the same sorts of scripts/classes/functions.

[swo]

I'll move these to an issue, so we can at least get this merged and get started on it