Apppublic

hugging-science/mmu-crossmatch

sourceHugging Facemitupdated 3mo agoView on Hugging Face

0likes

App README

MultimodalUniverse catalog crossmatch explorer

A Gradio app for exploring which MultimodalUniverse catalogs overlap on the sky, and crossmatching them — for real — on a single HEALPix pixel.

How it works

1.Catalog 1 — pick any MMU catalog.
2.Catalog 2 — the dropdown is trimmed to only the catalogs that actually share sky coverage with catalog 1.
3.HEALPix pixel — trimmed to the pixels both catalogs cover. Selecting one draws it on an instant all-sky coverage map.
4.Run crossmatch — downloads exactly that HEALPix partition of each catalog from HuggingFace and runs a real lsdb spatial crossmatch (like the showcase notebook), then plots the matched source pairs.
5.Show code — prints the exact lsdb snippet for the current selection so you can reproduce it.

The overlap lookup comes from the stored catalog_healpix.csv. MMU v1 partitions every catalog on the same HEALPix order 4 / nside 16 (NESTED) scheme (3072 pixels, 0–3071), so two catalogs overlap exactly when they share a healpix value. The HuggingFace datasets are named UniverseTBD/mmu_{catalog}_{subcatalog}.

Files

File	Purpose
`app.py`	The Gradio app (cascading dropdowns, coverage map, real crossmatch, code).
`catalog_healpix.csv`	`catalog,subcatalog,healpix,size_bytes` lookup table.
`build_lookup.py`	Rebuilds the CSV by crawling the MMU v1 data tree.
`requirements.txt`	Runtime dependencies.

Regenerating the lookup

bash

python build_lookup.py              # all catalogs
python build_lookup.py gaia chandra # just two

Run locally

bash

pip install -r requirements.txt
python app.py

Note: the crossmatch downloads the selected HEALPix partition from HuggingFace at run time, so the first match for a pixel takes a little while (≈1 min for a large catalog like Gaia); the instant coverage map needs no download.