Top PyPI Packages is a website that creates a monthly dump of the 5,000 most-downloaded packages from the Python Package Index (PyPI). It provides a human-readable list and a machine-readable JSON file for programmatic use.
The generated data is important for the Python community: it has been cited by many academic papers, covering research on topics such as software supply-chain attacks, genetic algorithms, neural type hints and technical debt.
Websites also use the data for analysing things like community adoption of Python 3 and of wheels package files, and for automated dependency updates.
Some are listed here.
Last but not least, CPython core developers often use the data for community analysis, for example to check how widespread certain language features are when considering deprecations and removals, or API changes of Python itself.
It runs on the cheapest DigitalOcean Droplet to query a Google BigQuery dataset from the Python Software Foundation, and builds the JSON file from that. The data is automatically committed from the droplet back to the GitHub repository, and GitHub Actions is used to automatically tag and create a release, which then creates a Digital Object Identifier (DOI) at Zenodo to help make it citable for researchers.
A "Droplet" is the name DigitalOcean uses for virtual machines, which is basically a Linux server. I'm using the cheapest $5/month running Ubuntu 20.04 (LTS) plus $1/month for automated backups. (In July 2022, I'll switch to the new $4/month Droplet.)
On the first of the month, it runs a cron job:
13 11 1 * * ( eval "$(ssh-agent -s)"; ssh-add ~/.ssh/id_rsa-top-pypi-packages; /home/botuser/github/top-pypi-packages/top-pypi-packages.sh ) > /tmp/top-pypi-packages.log 2>&1
/home/botuser/.local/bin/pypinfo --json --indent 0 --limit 5000 --days 30 "" project > top-pypi-packages-30-days.json
PyPI streams data about downloads to Google BigQuery which are accessible as a public dataset. Google provides a free amount of queries per month, and I've been adjusting the amount of data fetched to stay within the free quota (e.g. changing from the top 5k packages to 4k; it used to get data for top packages over the past 30 days and 365 days, but now only for 30 days; and bumping back up to 5k packages).
pypinfo is a handy command-line interface (CLI) to access this BigQuery data and dump it to a JSON file.
Another handy CLI tool called jq minifies the JSON data:
jq -c . < top-pypi-packages-30-days.json > top-pypi-packages-30-days.min.json
This then creates a Digital Object Identifier (DOI) at Zenodo to help make it citable for researchers.
The website, on GitHub Pages, reads in the generated JSON file and shows the top 100 (or 1,000 or 5,000) packages in human-readable form. It's based on Python Wheels, which in nice circular fashion, uses the JSON data from this project.
Thanks to PyPI and Google BigQuery for the data; pypinfo and jq for the tools; Python Wheels for making their code open source; and DigitalOcean for sponsoring this project's hosting. Visit https://do.co/oss-sponsorship to see if your project is eligible.