Report empirical things

[rebeccajohnson88]

Must do:

• [x] Input TRLA not WHD and WHD not TRLA accuracy results
• [x] Find feature importances for those
• [x] add within-covid over time variation- could be maybe- gen graphs; add to report along with refreshed others
• [ ] basic text comparisons- could be maybe

Maybe do:

• [ ] Go back to rawer data sources and find link between attorney agent name and attorney firm (probably trla raw-er data; think this was excluded pretty early from h2a disclosures data because not collected consistently across years)
• [ ] Predict WHD violations conditional on investigation
• [ ] Heatmap stuff

[rebeccajohnson88]

Notes on next steps for TRLA/WHD comparisons:

• [ ] Heatmap: use whd_violations_wTRLA_catchmentonly.csv which has (1) case_number and (2) ACS features. Dots = 4 categories; shading = ACS perc white or other race/eth
• [ ] Text: similarly that file has case_numbers; preprocess text, topic model, and correlate with focal categories

Likely drop: violations

Writing: clean up links (maybe new repo); etc