$ show MultimodalUniverse/MultimodalUniverse
The Multimodal Universe is a large-scale astronomical data repository that aggregates ~30 surveys (Legacy Survey, HSC, JWST, Gaia, SDSS, DESI, TESS, etc.) into a unified, HEALPix-partitioned HDF5 format, exposed through Hugging Face datasets. It covers multiple modalities: images (multi-band), spectra (1D), time series (light curves), and tabular catalogs. The repository provides dataset builder scripts (GeneratorBasedBuilder subclasses) per survey, cross-matching utilities, and benchmark training code (PyTorch Lightning) for downstream tasks like photometric redshift estimation and galaxy property regression. While primarily a data research project, it also includes application-like components such as baseline classifiers (Informer for PLAsTiCC) and experimental benchmark pipelines. The data is designed for training foundation models in astronomy, with streaming support and standardized column naming. Notable patterns include HEALPix-based spatial indexing, parallel processing in build scripts, and a consistent Hugging Face dataset interface across all surveys. The codebase is well-structured for its purpose, though many scripts are tightly coupled to specific surveys and external data sources, limiting direct reuse without adaptation.
The LLM identified 3 files worth highlighting for reuse:
Provides cross_match_datasets() and build_master_catalog() functions for spatial joins of astronomical datasets using sky coordinates. Implements cone search and HEALPix-based matching. Self-contained utility that could be extracted into a standalone library for cross-matching any catalog with RA/Dec columns.
Defines MMU and CrossMatchedMMU LightningDataModules that load the unified dataset from Hugging Face datasets with streaming, batching, and cross-matching support. Isolates data loading logic from model code, making it reusable for any downstream task using the Multimodal Universe data. Demonstrates a clean pattern for large-scale astronomical ML data loading.
Implements an abstract base class TESS_Downloader with concrete subclasses (SPOC_Downloader, TGLC_Downloader, QLP_Downloader) for downloading TESS light curves from different pipelines. Encapsulates authentication, file discovery, and retry logic. The abstract class pattern is reusable for downloading time-series data from any survey with multiple data products.