Skip to content

Datasets

Rules (also enforced in CI by tools/validate_dataset_cards.py):

  1. Every dataset folder has a DATASET_CARD.md with source, publisher, licence, licence status, date accessed, grades.
  2. Never commit a dataset without a verified licence in its card.
  3. Real data is fetched from the original publisher with python tools/fetch_data.py <name>; raw/ is git-ignored.
  4. The repo ships only small SYNTHETIC practice tables in clean/ (CC0, invented numbers, fictional district names).
  5. Restricted data (DHS, census/NLSS microdata) is never added.
Folder Grades Ships Real-data licence
class-survey-synthetic 8–9 36-row survey CC0 (synthetic)
census-districts 8–12 77-row district table + schools table real: CC BY-SA (ODN preliminary)
kathmandu-air-quality 9–11 731 days of messy PM2.5 real: CC BY-SA (ODN)
climate-monthly 10–12 360 months temp + rain real: CC BY 4.0 (World Bank CCKP)
health-synthetic 11–12 1,500 records real DHS: not redistributable

To add a dataset: copy docs/templates/DATASET_CARD_TEMPLATE.md, fill every field, open a PR. Full licence evidence: docs/alignment/dataset-licence-register.md.

Free real datasets for exercises

Each chapter's Exercises with real data uses openly available datasets downloaded from their publishers (World Bank, Our World in Data, USGS, Open-Meteo, NASA POWER, Open Data Nepal, ICIMOD and others). They are not stored here. The directory, with licences, download links and offline fallbacks, is docs/datasets/free-datasets.md (codes RD-01 … RD-20).