Metrics
20 metrics across 3 evaluation suites.
Clustering — BEN
3 metrics
Dimensionality Reduction — DRE
10 metrics
Spearman distance corr (latent vs UMAP)
DRE.evaluate_dimensionality_reduction
Coranking Q_NX local average over the selected neighbourhood range
DRE Lee & Verleysen 2009
Coranking Q_NX global average
DRE
Argmax of Q_NX(k) — optimal neighbourhood scale
DRE
Weighted composite of Q_local + Q_global
DRE
Spearman dist corr (latent vs tSNE)
DRE
Coranking Q_NX local (tSNE)
DRE
Coranking Q_NX global (tSNE)
DRE
Argmax Q_NX(k) in tSNE
DRE
Weighted composite (tSNE)
DRE
Latent Space Evaluation — LSE
7 metrics
Intrinsic manifold dimensionality score
LSE.SingleCellLatentSpaceEvaluator
PCA eigenvalue decay rate
LSE
Effective latent dim (1/Σ w_i² when Σ w_i=1)
LSE
Covariance eigenvalue anisotropy
LSE
Primary developmental axis strength
LSE
Perturbation robustness score
LSE
Weighted composite (trajectory-aware)
LSE