This is the source code from the Reproducible Data Science with Clojure workshop, presented live in Berlin on March 14, 2025.
Despite sophisticated tooling, data scientists still battle fundamental challenges day to day, like reproducibility, maintainability, and sharing their work. While traditional notebooks offer interactivity and quick feedback, they’re plagued with hidden state dependencies, version control complexity, and production deployment hurdles. Converting notebook-based analyses into production-ready code often requires extensive refactoring, untangling implicit dependencies, debugging hidden state issues, and deciphering sparse documentation. And that’s before tackling today’s reality of excessively large, unstructured data dumps typically lacking any metadata or explanation, making it difficult to find useful data in the first place.
Clojure’s data science ecosystem has been maturing rapidly in recent years. With it’s stable toolkit, immutable data structures, and functional paradigm, Clojure offers a compelling alternative to traditional data science workflows. Imagine knowing exactly which version of your code produced which dataset. Or seamlessly deploying the same code you used in an exploratory analysis to production. And imagine that code also ran in a state of the art literate programming environment, but also simultaneously in your own, familiar IDE.
This hands-on workshop will introduce a new way of thinking about working with data, demonstrating how Clojure’s libraries and tooling solve many pain points in current data science workflows.
This workshop is a high-level overview of Clojure's data processing capabilities, demonstrating an end-to-end process for analysing a public dataset. At the time of writing the tools and libraries demonstrated here represent the most stable and recommended setup for working with data in Clojure.
If you're a workshop participant or want to follow along, there is an installation guide available in the 1_installation_guide.clj notebook, which is also published online here.
For developers of the workshop itself, the notebooks are namespaces in the src/notebooks directory.
To build and deploy the book you will need the quarto cli and babashka installed on your system.
There is a build script to convert the Clojure namespaces into a quarto book in bin/build.clj, which you can run via the configured babashka tasks like so:
bb build
There are also other bb tasks available, which you can see by running bb tasks:
The following tasks are available:
clean Deletes the `book` directory
build Builds the quarto publishable-book
deploy Deploys the book to the quarto pub location specified in `bin/_publish.yml`
preview Preview the already built book, url will be printed once it's ready
release Clean, build, and deploy all at once
You can update the bb tasks to publish your own version of the book using quarto publish, to any of the many available free options it supports. If you want to use quarto-pub with your own account, update the details in bin/_publish.yml.