scattered: a composable library for 3D scatterplots

3D scatterplots probably suck, but they're much more fun than regular scatterplots!

Caution

This project is as Work-In-Progress as it gets.

The basic idea is:

• arrow/dataframe as input (with 'x', 'y', 'z' columns)
• minimal dependencies & small bundle size
• webgpu rendering
• javascript library (npm), jupyter widget (pypi), web page (like quak)

use in python

Install:

uv add scattered

or: pip install scattered

Then use:

import scattered
import numpy as np
import pandas as pd

df = pd.DataFrame({
    "x": np.random.rand(5),
    "y": np.random.rand(5),
    "z": np.random.rand(5),
})

scattered.Widget(df)

use in javascript

Install:

pnpm add scattered

or: npm install scattered or yarn add scattered

Then use:

import * as sctrd from "scattered";

const url = "https://raw.githubusercontent.com/dvdkouril/sample-3d-scatterplot-data/main/penguins.arrow";
const c = sctrd.display(url,
    { // encoding
        x: "x",
        y: "y",
        z: "z",
        color: "category",
    });

let appEl = document.querySelector('#app');
if (c) {
    appEl.appendChild(c);
}

about

This project is developed by David Kouřil (web, bsky).

why

I think we're missing a modern, simple to use library for interactive 3D scatterplots. Visualization libraries are typically centered around 2D plots to make exporting as vector graphics easier.

However, learning is the primary motivation:

1. learn a bit of webgpu
2. explore the composability principles for visualization tools
3. learn about bundling, and how to maintain multi-package project

scattered is very much inspired by quak and also follow the structure of that repo.

related and inspo

From researching 3D scatterplots in the wild.

• https://matplotlib.org/stable/gallery/mplot3d/scatter3d.html
• https://plotly.com/python/3d-scatter-plots/
• https://jupyter-scatter.dev (i don't think there's 3D option)
• https://abdenlab.org/eigen-tour/ (repo: https://github.com/abdenlab/eigen-tour)
• https://projector.tensorflow.org