Refactor codebase to use AnnData (Fixes #56)

[jvivian]

Motivation

• Make DFM code (and dashboard eventually) agnostic to our covid19 project data
• H5AD is a superior format for cases where you need to track both observational and variable metadata (along with some numeric data matrix $X$).
• Despite all of the code already working, I decided "why not rewrite all of it while on vacation?" like some sort of idiot.

Modules

covid19.py

Moved all project-related code to this module (e.g. get_df()). The principle function used from this module is now get_project_h5ad.

io.py

Module for handling interoperability between CSVs/DataFrames and AnnData/H5AD

processing.py

Module that prepares all data to run in DFM (such as differencing variables and normalization) via the DataProcessor class.

dfm.py

Module that handles running the model via a ModelRunner. A complete example of running the model across all 50 states:

from covid19_drdfm.covid19 import get_project_h5ad
from covid19_drdfm.dfm import ModelRunner

ad = get_project_h5ad()
model = ModelRunner(ad, batch='State')
model.run() # Data will output to `./output` by default

[codecov]

Codecov Report

Attention: Patch coverage is 92.12598% with 20 lines in your changes are missing coverage. Please review.

Project coverage is 93.1%. Comparing base (dd5a202) to head (8d7c3c5).

Files	Patch %	Lines
covid19_drdfm/processing.py	84.5%	6 Missing and 7 partials ⚠️
covid19_drdfm/dfm.py	93.4%	2 Missing and 2 partials ⚠️
covid19_drdfm/cli.py	88.8%	3 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##            main     #74     +/-   ##
=======================================
- Coverage   96.3%   93.1%   -3.2%     
=======================================
  Files          5       7      +2     
  Lines        193     308    +115     
  Branches      35      50     +15     
=======================================
+ Hits         186     287    +101     
- Misses         5      11      +6     
- Partials       2      10      +8     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.