YAML-run pipeline, part 1 (modeling/eval)

[thanasibakis]

Major changes

linmod.models

Numpyro model functions are now contained in a class with corresponding model-specific data preprocessing and sample postprocessing. This builds on Scott's work in #14; it was just easier to reimplement here than handle the rebasing on main.

These model classes have three methods:

• A constructor, which accepts a dataframe obeying our standard model input format described in the repository's root README. The model's required predictors are extracted here and stored in the model object.
• A numpyro model method, which can be given directly to whichever numpyro sampler we like
• A postprocessing method, which accepts a numpyro MCMC object and returns a dataframe obeying our standard model output format described in the repository's root README

exploration/demo/ --> present_day_forecasting/

The demo has become pretty sophisticated over time and is close enough to our target goal of tooling to move to the root of the repo.

Because of the refactoring done here for linmod.models, I think we're able to merge the fitting scripts for all models, as well as evaluation and visualization scripts, into one (not so large) main.py; feel free to disagree with me here, though!

This main.py naturally has some behaviors we want to configure, so it accepts a path to a YAML file as a commandline argument. I haven't put any default behavior handling here, so all keys must be in this YAML.

See the present_day_forecasting/README.md for how to run it.

[afmagee42]

One assumption that we've been hard-coding and need to make configurable is that we want to model all lineages. (I will accept this being ruled out of scope of this PR, but for the July data I'm playing with, the hierarchical model shows pretty clearly that most of the lineages have negligible proportions.)