Data & Datasets
This page is the entry point to the data used by AstroSpectro: where the spectra come from, how they are enriched, and which derived tables the analyses run on. Two data generations coexist in the project and are never mixed in a single result: PHY-3500 (G1), used by every dimensionality-reduction page, and the 350k-spectrum generation (G2), used by the reference classifier.
Data generations
| G1 — PHY-3500 | G2 — 350k | |
|---|---|---|
| Feature file | features_20260305T174836Z.csv | features_20260522T153123Z.csv |
| Spectra | N = 43,019 (STAR 42,956 · GALAXY 56 · QSO 7) | 349,998 spectra processed |
| Features | p = 183 spectroscopic descriptors | 207 candidate columns → 99 selected by the classifier |
| Used by | PCA, UMAP, t-SNE, HDBSCAN, autoencoder (PHY-3500 article) | Reference XGBoost classifier (ST_350k_xgb_AFGKM_v6), its SHAP analysis and the scientific validation |
| Label on the site | PHY-3500 · G1 · 43,019 spectra | Run name of the classifier |
The classifier was never evaluated on the PHY-3500 dataset. A classification metric must not be attached to the G1 spectrum count, and a dimensionality-reduction result must not be presented as a property of the G2 data. The reference classifier also contains Gaia/photometric columns — see the Data Leakage Audit.
From archive to feature table
| Data product | Content | Page |
|---|---|---|
| LAMOST DR5 spectra | Low-resolution optical spectra (FITS), flux and inverse variance on a log-linear wavelength grid | LAMOST DR5 |
| Gaia DR3 cross-match | Astrometry, photometry and GSP-Phot parameters added to the catalogue | Gaia DR3 |
| Master catalogue | One row of header metadata per spectrum, optionally enriched with Gaia | Master Catalog |
| Feature table | One row per spectrum, feature_* and match_* columns plus catalogue columns | Feature dataset · Feature catalogue |
| Class labels | Harvard type taken from the LAMOST subclass | Class Taxonomy |
| Quality filters | SNR and Gaia RUWE criteria | Quality Filters |
| Versions | Successive feature tables and their sizes | Dataset Versions |
Class setup
The G1 dataset keeps the LAMOST object classes (STAR, GALAXY, QSO). The reference classifier works on stellar types only — A, F, G, K, M — and excludes B, C, D, S, W, s; see Classification.
What is not documented yet
The exported fact pack does not contain the Gaia cross-match statistics (match rate, effective radius) or the number of spectra removed at each quality-filter step; these figures are therefore not published.