Record scripts and experiment results. No need to abstract out the generic elements yet; just recording the scripts experiment-by-experiment seems fine.

Proposed experiment directory structure:

• hatespeech
  • README.md describing purpose, setup, code, and results
  • data
    • README.md on where/how to get the data, or in some cases the data itself can go here
  • notebooks
    • foo.ipynb (may refer to main codebase over time, if we start abstracting things)
  • images
    • visualizations, logos, whatever for the README file.