[pdp] Map feature columns to human-friendly names

[bdewilde]

changes

• adds tomli as optional project dependency (for PY3.10 only)
• adds "assets" directory, and within it, adds a "features table" toml file mapping PDP column names to human-friendly feature names (and, eventually, fuller descriptions)
• adds a utils func to load+parse features tables from disk
• adds feature values to inference topn feature display function (ported from https://github.com/datakind/student-success-intervention/pull/120) and also optional feature name mapping by way of a features table

context

PDP's internal feature column naming is a bit... technical. Advisors need more friendly naming to help them interpret model outputs correctly.

questions

• Does this seem like a reasonable implementation?
• How do y'all feel about TOML? Personally I love it, and much prefer it over YAML. :D

[vishpillai123]

I think we may want to update the unit test for inference as well to reflect the new output format. Once we do that, I think we're good to merge.

[vishpillai123]

Having human friendly names in our output feature file is a great idea. Definitely onboard with this. I'm also open to .toml file, I don't necessarily have a preference either way.

[vishpillai123]

Should we also update the unit test as well under tests/modeling/test_inference.py?

Other than that, this looks good to me and similar to the changes in the private repo.

[bdewilde]

Yeah, good call. I ported those tests over as-is from the private repo; will modernize and extend here.