mmu_sdss_dr17 HATS Catalog Collection

This is the collection of HATS catalogs representing mmu_sdss_dr17.

This dataset is part of the Multimodal Universe, a large-scale collection of multimodal astronomical data. For full details, see the paper: The Multimodal Universe: Enabling Large-Scale Machine Learning with 100TBs of Astronomical Scientific Data.

Access the catalog

We recommend the use of the LSDB Python framework to access HATS catalogs. LSDB can be installed via pip install lsdb or conda install conda-forge::lsdb, see more details in the docs. The following code provides a minimal example of opening this catalog:

import lsdb

# Full sky coverage.
catalog = lsdb.open_catalog("https://huggingface.co/datasets/UniverseTBD/mmu_sdss_dr17")
# One-degree cone.
catalog = lsdb.open_catalog(
    "https://huggingface.co/datasets/UniverseTBD/mmu_sdss_dr17",
    search_filter=lsdb.ConeSearch(ra=132.0, dec=25.0, radius_arcsec=3600.0),
)

Each catalog in this collection is represented as a separate Apache Parquet dataset and can be accessed with a variety of tools, including pandas, pyarrow, dask, Spark, DuckDB.

File structure

This catalog is represented by the following files and directories:

• collection.properties — textual metadata file describing the HATS collection of catalogs
• mmu_sdss_dr17 — main HATS catalog directory
  • dataset/ — Apache Parquet dataset directory for the main catalog
    • ... parquet metadata and data files in sub directories ...
  • hats.properties — textual metadata file describing the main HATS catalog
  • partition_info.csv — CSV file with a list of catalog HEALPix tiles (catalog partitions)
  • skymap.fits — HEALPix skymap FITS file with row-counts per HEALPix tile of fixed order 10
• mmu_sdss_dr17_10arcs/ — default margin catalog to ensure data completeness in cross-matching, the margin threshold is 10.0 arcseconds
  • ... margin catalog files and directories ...

Catalog metadata

Metadata of the main HATS catalog, excluding margins and indexes:

Number of rows	Number of columns	Number of partitions	Size on disk	HATS Builder
3,930,301	31	968	176.3 GiB	hats-import v0.7.1, hats v0.7.1

Catalog columns

The main HATS catalog contains the following columns:

Name	_healpix_29	spectrum.flux	spectrum.ivar	spectrum.lsf_sigma	spectrum.lambda	spectrum.mask	VDISP	VDISP_ERR	Z	Z_ERR	ra	dec	healpix	ZWARNING	SPECTROFLUX_U	SPECTROFLUX_G	SPECTROFLUX_R	SPECTROFLUX_I	SPECTROFLUX_Z	SPECTROFLUX_IVAR_U	SPECTROFLUX_IVAR_G	SPECTROFLUX_IVAR_R	SPECTROFLUX_IVAR_I	SPECTROFLUX_IVAR_Z	SPECTROSYNFLUX_U	SPECTROSYNFLUX_G	SPECTROSYNFLUX_R	SPECTROSYNFLUX_I	SPECTROSYNFLUX_Z	SPECTROSYNFLUX_IVAR_U	SPECTROSYNFLUX_IVAR_G	SPECTROSYNFLUX_IVAR_R	SPECTROSYNFLUX_IVAR_I	SPECTROSYNFLUX_IVAR_Z	object_id	survey
Data Type	int64	list[float]	list[float]	list[float]	list[float]	list[bool]	float	float	float	float	double	double	int64	bool	float	float	float	float	float	float	float	float	float	float	float	float	float	float	float	float	float	float	float	float	string	string
Nested?	nan	spectrum	spectrum	spectrum	spectrum	spectrum	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan
Value count	3,930,301	17,648,419,380	17,648,419,380	17,648,419,380	17,648,419,380	17,648,419,380	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301	3,930,301
Example row	345406689359877987	[23.33, 23.33, 17.75, 12.16, … (4662 total)]	[0, 0.06879, 0, 0.07073, 0.07516, … (4662 total)]	[0, 1.085, 1.086, 1.086, 1.086, … (4662 total)]	[3554, 3555, 3555, 3556, 3557, … (4662 total)]	[True, False, True, False, False, … (4662 total)]	350.8	99.61	0.5869	0.0003262	131.9	25.11	306	False	-0.1748	0.2874	1.657	4.724	7.257	1.65	11.03	6.335	4.591	2.056	-0.06638	0.2546	1.628	4.673	7.835	9.908	11.31	7.078	5.998	3.395	5832203123784177664	boss
Minimum value	168688995934	-179130608.0	-0.0	-23.887535095214844	-1.0	False	-0.0	-7.0	-0.011446905322372913	-6.0	9.09079089979059e-05	-11.25283	0	False	-1119.229248046875	-940.6095581054688	-865.4228515625	-865.4229125976562	-980.3538818359375	0.005544150713831186	0.005543752107769251	0.005543776787817478	0.005543770734220743	0.0032096304930746555	-38421780.0	-26692180.0	-11347789.0	-8527988.0	-26592832.0	5.743364113186544e-07	7.989542041286768e-07	1.7297466001764406e-06	8.407338896176952e-07	3.4694645023591875e-07	1000925336738752512	boss
Maximum value	3458764448209686099	180670528.0	1627974912.0	3.4599151611328125	10415.9765625	True	850.0	4525.4833984375	7.051929950714111	25020.80078125	359.9999399999999	70.287347	3071	True	28427.263671875	129483.3828125	346752.21875	780947.1875	1373210.875	39.119293212890625	41.30256271362305	24.57221794128418	15.387298583984375	7.61610746383667	48433084.0	33275774.0	13660257.0	109366224.0	359602944.0	41.673828125	42.04224395751953	29.711549758911133	19.10918617248535	10.141582489013672	999975078378629120	sdss

"Nested" indicates whether the column is stored as a nested field inside another "struct" column.

"Value count" may be different from the total number of rows for nested columns: each nested element is counted as a single value.

Crossmatch with another catalog

HATS catalogs can be efficiently crossmatched using LSDB, which leverages the HEALPix partitioning to avoid loading the full datasets into memory:

import lsdb

mmu_sdss_dr17 = lsdb.open_catalog("https://huggingface.co/datasets/UniverseTBD/mmu_sdss_dr17")
other = lsdb.open_catalog("https://huggingface.co/datasets/<org>/<other_catalog>")

crossmatched = mmu_sdss_dr17.crossmatch(other, radius_arcsec=1.0)
print(crossmatched)

See the LSDB documentation for more details on crossmatching and other operations.

Dataset-specific context

Original survey
This dataset is based on the Sloan Digital Sky Survey (SDSS) Data Release 17. It includes optical spectra from the SDSS Legacy and SEGUE programs (survey = sdss), BOSS (survey = boss), and eBOSS (survey = eboss), built with the unmodified Multimodal Universe SDSS builder.

Data modality
The dataset consists of optical spectra covering wavelength ranges from 3800 to 9200 Å (SDSS spectrograph) and 3600 to 10400 Å (BOSS spectrograph). Each spectrum includes flux, wavelength, inverse variance (ivar), line-spread function width, and pixel-level masks. The dataset contains 3,930,301 spectra (sdss 806,176; boss 2,031,730; eboss 1,092,395).

Typical use cases
SDSS spectra have been used in numerous scientific publications. Machine learning applications include building data-driven representations of galaxy spectra and identifying outliers.

Caveats
Spectra have varying lengths; non-positive spectrum.lambda values are padding. Selection: SPECPRIMARY == 1, TARGETTYPE == SCIENCE, PLATEQUALITY == good. object_id is the SDSS SPECOBJID as a plain digit string. The survey column records the SDSS program.

Citation
SDSS data releases are in the public domain. Users should include the appropriate acknowledgements for the SDSS, SEGUE-1, SEGUE-2, BOSS, and eBOSS samples when using this dataset.