Data Science Config

# CLAUDE.md — Data Science Config

## Notebook Conventions
- One notebook per experiment, named: YYYY-MM-DD-experiment-name.ipynb
- First cell: imports and config (no magic numbers in code cells)
- Last cell: summary of results and next steps
- Clear all outputs before committing

## Experiment Tracking
- Log every run: parameters, metrics, artifacts
- Use MLflow or Weights & Biases for tracking
- Never overwrite previous experiment results
- Tag experiments: exploratory, validation, production

## Reproducibility Requirements
- Pin all package versions in requirements.txt
- Set random seeds: numpy, torch, sklearn
- Document data source, version, and access date
- Include data preprocessing steps in pipeline (not notebook)