---
title: 🌸 MONET UMAP
emoji: 🗺️
colorFrom: pink
colorTo: green
sdk: static
pinned: false
license: apache-2.0
short_description: Interactive UMAP of the MONET dataset
---

# MONET UMAP

Static HF Space: interactive UMAP scatterplot of the [`jasperai/monet`](https://huggingface.co/datasets/jasperai/monet) dataset (~112M points), built with [Vite](https://vitejs.dev/) + [TypeScript](https://www.typescriptlang.org/) and rendered with [`deck.gl`](https://deck.gl/) (`OrthographicView` + WebGL `ScatterplotLayer`).

- `points.parquet` and the spatially-sharded thumbnail shards live in [`jasperai/monet-umap-data`](https://huggingface.co/datasets/jasperai/monet-umap-data) and are read directly in the browser via [`hyparquet`](https://github.com/hyparam/hyparquet) + [`hyparquet-compressors`](https://github.com/hyparam/hyparquet-compressors) — full-file read for the points, HTTP byte-range reads for hover thumbnails.
- Hover previews paint instantly from pre-baked low-res WebP atlas pages, then upgrade to full-res thumbs once the parquet range read resolves.
- No backend: pure `static` HF Space SDK.

## Local dev

```bash
npm install
npm run dev   # http://localhost:5173
```

## Build & deploy

Deploys to [`jasperai/monet-umap`](https://huggingface.co/spaces/jasperai/monet-umap)
(`static` SDK). The Space ships only `index.html` + the JS/CSS bundle (~775 kB);
all data artifacts (`points.parquet`, `tiles/`, `atlases/`, `preview/`, `idx/`)
are fetched at runtime from the HF bucket that `pipeline.py upload_dataset` syncs
to. The bucket base URL is baked in at build time from `.env.production`
(overridable by the gitignored `.env.production.local`).

Run from the repo root (needs `HF_TOKEN` with write access to the Space):

```bash
cd space

# 1. Strip the dev-only symlink (space/public/data → ./work). If left in place,
#    `vite build` deep-copies the whole work dir into dist/.
rm -f public/data

# 2. Build the static bundle and ship the HF Space frontmatter (this README).
npm run build                 # tsc -b && vite build → space/dist/
cp README.md dist/README.md

# 3. Push dist/ to the Space (creates it on first run; static SDK).
cd ..
python -m data_prep.upload_to_hub --space --folder space/dist --repo-id jasperai/monet-umap
```

One-liner equivalent:

```bash
(cd space && rm -f public/data && npm run build && cp README.md dist/README.md) \
  && python -m data_prep.upload_to_hub --space --folder space/dist --repo-id jasperai/monet-umap
```

Notes:

- The `upload_to_hub --space` command `upload_folder`s `dist/` with
  `delete_patterns=["*"]`, so the Space always mirrors the local build. Pass
  `--private` (honored only at create time) or `--squash-history` if needed.
- To retarget the live Space at a different bucket prefix (e.g. `umap-1m`)
  without rebuilding, upload a `runtime-config.json` override — see the root
  `README.md` ("Retarget the deployed Space") for the full env/runtime
  resolution order.