OneAstronomy/sspp-dr9-stellar-spectra

SDSS SSPP DR9 Stellar Spectra Dataset

Dataset Summary

This dataset contains 233,013 stellar spectra from the Sloan Digital Sky Survey (SDSS) Stellar Parameter Pipeline (SSPP) DR9, cross-matched with SEGUE1/SEGUE2 spectroscopic observations.

Each sample contains a flux-calibrated 1D spectrum (3,878 wavelength bins, ~3,820–9,200 Å) plus stellar atmospheric parameters measured by the SSPP.

Data Source

Field	Value
Survey	SDSS SEGUE (DR9)
Pipeline	SSPP (Stellar Parameter Pipeline)
Reference	Lee et al. (2008), Allende Prieto et al. (2008)
Original file	sspp_matched_spectra.h5
ETL date	2026-05-22

Dataset Structure

Splits

Split	Samples
train	186,409
validation	23,302
test	23,302
total	233,013

Features

Column	Type	Description
spectrum_flux_norm	Sequence[float32] (3878)	Continuum-normalised flux (÷ per-spectrum median)
spectrum_flux_raw	Sequence[float32] (3878)	Original calibrated flux
spectrum_ivar	Sequence[float32] (3878)	Inverse variance (noise model)
spectrum_mask	Sequence[bool] (3878)	Bad-pixel mask (True = invalid)
TEFF_ADOP	float32	Effective temperature Teff (K)
LOGG_ADOP	float32	Surface gravity log g (dex)
FEH_ADOP	float32	Metallicity [Fe/H] (dex)
AFE	float32	Alpha-element abundance [α/Fe]
RV_ADOP	float32	Radial velocity (km/s)
SNR	float32	Spectral signal-to-noise ratio
ra / dec	float32	Sky coordinates (degrees)
healpix	float32	HEALPix pixel index
object_id	string	Unique object identifier
SPECOBJID	string	SDSS spectroscopic object ID
...	...	See full schema in dataset_info.json

The common wavelength grid is stored in wavelength_grid.json.

Label Statistics (training set)

Parameter	Min	Max	Mean	Std
TEFF_ADOP (K)	4003	9593	5664.7	802.5
LOGG_ADOP (dex)	0.18	5.00	3.94	0.62
FEH_ADOP (dex)	-4.35	0.74	-0.98	0.64
SNR	10.0	131.4	36.6	18.7

Quality Filtering

The ETL pipeline applied the following filters:

1. ZWARNING ∈ {0, 16} – valid redshift measurement
2. TEFF_ADOP, LOGG_ADOP, FEH_ADOP are finite – all three core parameters must be valid
3. SNR ≥ 10 – minimum signal-to-noise ratio
4. Not entirely masked – at least one valid spectral pixel

Loading the Dataset

from datasets import load_dataset

ds = load_dataset("parquet", data_dir="/mnt/si0009256k6u/ckdata/aiready/sspp/hf_dataset")

# Access train split
train = ds["train"]
print(train[0]["TEFF_ADOP"])   # ~5500 K

Or load individual splits:

import pandas as pd
df = pd.read_parquet("/mnt/si0009256k6u/ckdata/aiready/sspp/hf_dataset/train/data.parquet")

Citation

@article{lee2008sspp,
  title={The SEGUE Stellar Parameter Pipeline. I. Description and Comparison of Individual Methods},
  author={Lee, Young Sun and others},
  journal={The Astronomical Journal},
  year={2008}
}