Project Instructions

Notebooks

• Keep notebooks focused on one analysis or experiment.
• Clear all outputs before committing. Don’t commit cell outputs to git.
• Move reusable logic into .py modules. Notebooks are for exploration and presentation, not library code.

Data

• Never commit raw data to the repo. Document where to get it and how to set it up.
• Use relative paths for data files. Don’t hardcode absolute paths.
• Document data schemas and assumptions in comments or a data dictionary.

Reproducibility

• Pin all dependencies with exact versions.
• Set random seeds for any stochastic process.
• Document the steps to reproduce results from scratch.

Code Quality

• Type hints on functions. Docstrings on anything non-obvious.
• Use pandas and numpy idiomatically. Avoid row-by-row loops when vectorized operations work.
• Keep data transformations in named functions, not inline chains that span 20 lines.

Git

• Write descriptive commit messages: “add feature importance analysis” not “update notebook.”
• Use .gitignore for data files, model artifacts, and notebook checkpoints.