Module pca 

Principal Component Analysis (PCA)

Implements dimensionality reduction for node feature matrices.

Two solvers are available:

• Randomized SVD (default): Halko-Martinsson-Tropp algorithm. O(n·d·k), numerically stable, automatic orthogonality. Industry standard (scikit-learn, cuML, Spark MLlib).
• Power Iteration (legacy): Extract one eigenvector at a time from the covariance matrix, then deflate and repeat. With Gram-Schmidt re-orthogonalization for stability.

Structs

PcaConfig

PCA configuration

PcaResult

PCA result containing components and explained variance

Enums

PcaSolver

Solver strategy for PCA.

Functions

pca

Run PCA on a feature matrix.