Complete comparison of three dimensionality reduction methods on 43,019 LAMOST DR5 stellar spectra: PCA (linear), UMAP (topological) and t-SNE (probabilistic). Hyperparameters, results, Procrustes stability and HDBSCAN clusters.
AstroSpectro's most important metrics and scientific findings.
Systematic comparison of PCA, UMAP, t-SNE, and autoencoder applied to AstroSpectro.
AstroSpectro results in two separate studies: unsupervised dimensionality reduction (PCA, UMAP, t-SNE, HDBSCAN) on the PHY-3500 dataset (G1, 43,019 spectra), and the reference XGBoost classifier with its SHAP attribution on the 350k-spectrum generation (G2, run v6, which includes Gaia/photometric features).
UMAP topological projection and unsupervised HDBSCAN clustering of the stellar spectral feature space.
Results of the reference XGBoost classifier (run v6, 350k-spectrum generation G2, five types A–M): balanced accuracy, macro F1 and ROC-AUC from runs.json, per-class performance, SHAP attribution — with the Gaia/photometric features it contains — and its qualitative link with the PHY-3500 UMAP study (G1).