Glossary
Terms are listed alphabetically. Click a term to see its full definition.
Balanced Accuracy (BA) Classification metric that averages per-class recall, giving equal weight to each class regardless of class frequency. Used throughout AstroSpectro and AstroVision due to class imbalance.
BAO (Baryon Acoustic Oscillations) Characteristic scale imprinted in the large-scale structure of the universe by sound waves in the early universe. Used as a standard ruler in cosmological distance measurements. See ξ Dark Energy.
Ca II H&K lines Calcium ionised doublet at 3933 Å (K) and 3968 Å (H). Strong metallicity proxies and chromospheric activity indicators. Rank #1–5 in AstroSpectro SHAP analysis.
CPL (Chevallier-Polarski-Linder) A dark energy equation-of-state parameterisation: w(a) = w₀ + w_a(1-a). One of three models compared in ξ Dark Energy.
Equivalent Width (EW) Measure of spectral line strength — the width of a rectangular absorption feature with the same integrated flux as the observed line. Used extensively in AstroSpectro feature engineering.
HDBSCAN Hierarchical Density-Based Spatial Clustering of Applications with Noise. Used in AstroSpectro to identify 20 stellar population clusters in UMAP space without labels.
Harvard Spectral Sequence The stellar classification sequence O-B-A-F-G-K-M ordered by decreasing temperature. AstroSpectro classifies A, F, G, K, M types.
LAMOST DR5 The 5th data release of the Large Sky Area Multi-Object fiber Spectroscopic Telescope survey (~9 million spectra). Primary data source for AstroSpectro.
MC Dropout Monte Carlo Dropout — a Bayesian approximation technique for uncertainty quantification in neural networks. Used in AstroVision.
ROC-AUC Area Under the Receiver Operating Characteristic curve. Measures discrimination ability; the AstroSpectro reference run (v6, which includes four Gaia/photometric features — see the Data Leakage Audit) achieves 0.986 macro-average.
SHAP (SHapley Additive exPlanations) A game-theoretic method for explaining ML model predictions by attributing each feature's contribution to the output. Central validation tool in AstroSpectro.
spectro_only=True The canonical feature mode in AstroSpectro that includes only physically interpretable spectroscopic features, excluding positional and instrumental metadata. See spectro_only=True Mode.
T_eff (Effective Temperature) The temperature of a blackbody with the same total luminosity as a star. Primary axis of the Harvard spectral sequence.
UMAP (Uniform Manifold Approximation and Projection) Non-linear dimensionality reduction algorithm. Used in AstroSpectro to project 183 spectral features to 2D, revealing the Harvard sequence without supervision.
ξosc (Oscillating dark energy) An exploratory dark energy parameterisation with an oscillating equation of state. One of three models in ξ Dark Energy (v0.1.0-exploratory, not a claimed detection).
See also: Acronyms