marimo-publication

Example of publishing serverless interactive visualizations using marimo

Goal:

• Publish an interactive visualization as a standalone website

Approach:

• Using the open source marimo Python library,
• Enter figures and text using a familar notebook interface,
• Test locally to quickly author a manuscript or report,
• Deploy to a static webpage, where all code is run directly in the reader's browser.

Benefits:

• No need to run costly webservers
• Use the rich Python ecosystem for data science and visualization
• Publications are 100% reproducible, with all code available to the reader

Drawbacks:

• Not appropriate for long-running or costly computational tasks
• Upper limit of ~2GB for in-browser tasks

Deployment

To deploy the visualization defined in app.py to a public website, simply set up GitHub Pages for this repository. Select the option for deploying from Actions. The Action defined in .github/workflows/deploy.yaml will trigger a build and deployment every time a commit is pushed to this repository. After you enable Actions, you will need to push a commit to trigger the Action for the first time.

Development

Set up local environment

1. Install Python
2. Install dependencies (pip install -r requirements.txt) (ideally within a virtual environment)

Edit visualization

Using the local environment, the visualization in app.py can be edited interactively before publishing to the web.

marimo edit app.py

Adding data

Files added to the public/ folder will be included in the deployment. Following the marimo documentation, using the pattern path_to_csv = mo.notebook_location() / "public" / "data.csv" will ensure that the data is sources from the appropriate location during local development as well as the deployed website.

Testing deployment

To run a local test of the conversion to HTML-WASM:

rm -rf test_build
marimo export html-wasm app.py -o test_build --mode run
python -m http.server --directory test_build

And then open localhost:8000 in your browser.