anyupset

Interactive UpSet plots as an anywidget, for marimo and Jupyter.

An UpSet plot shows how items are distributed across combinations of sets. The point of this one is seeing the actual items: hover any bar and its members appear in the side panel; click to pin the selection and pull it back into Python.

[!WARNING] Early days. This is a 0.1 release. The API may still change, and the widget has seen little use beyond its own test suite and demos — pin an exact version if you build on it, and please report anything that breaks.

Written with AI assistance. Nearly all of this project — the widget, the tests, the demos and the CI — was written by Claude in a pair-programming session, then reviewed and directed by a human. It is tested, but it has not had the scrutiny of a codebase grown by hand over time. Read the source before you depend on it.

Install

uv add anyupset

Use

from anyupset import UpSet

upset = UpSet({
    "drama":   {"Titanic", "Casablanca", "Whiplash"},
    "romance": {"Titanic", "Casablanca", "Notting Hill"},
    "comedy":  {"Notting Hill"},
})
upset

Columns are exclusive intersections: each item belongs to exactly one column, the one matching the full set of sets it is in.

Data often arrives the other way round — one row per item, listing what it belongs to. from_memberships flips it for you, and builds the same plot:

upset = UpSet.from_memberships({
    "Titanic":      ["drama", "romance"],
    "Casablanca":   ["drama", "romance"],
    "Whiplash":     ["drama"],
    "Notting Hill": ["romance", "comedy"],
})

Reading the selection

upset.selected_sets      # ['drama', 'romance']
upset.selected_members   # ['Casablanca', 'Titanic'] — original Python objects
upset.selected_values    # the same, reduced to JSON-safe scalars

In marimo, wrap it so downstream cells re-run on interaction:

upset = mo.ui.anywidget(UpSet(...))
upset.value["selected_sets"]     # ['drama', 'romance']
upset.value["selected_values"]   # ['Casablanca', 'Titanic']

.value carries only synced traits. selected_members is deliberately not synced — it holds the original objects, uncapped, and never crosses the wire — but the wrapper proxies attribute access, so upset.selected_members still works. Use .value for reactivity, the attribute when you need fidelity.

Examples

Both demos build the same plot from the same 156-film dataset.

# marimo — reactive: clicking a bar re-runs the cells below it
uv run --group dev marimo edit examples/marimo_demo.py

# Jupyter — re-run the reading cell yourself after clicking
uv run --group jupyter jupyter lab examples/jupyter_demo.ipynb