component scales (singular values)

Return the per-component scale of a fitted projection. For the SVD-based methods in this package (svd_wrapper, pca, bi_projector), this is the vector of singular values of the pre-processed data matrix, i.e. the square roots of the eigenvalues of \(X^\top X\). It is not the sample standard deviation of the score columns: those equal sdev(x) / sqrt(n - 1) (the convention used by stats::prcomp). The two differ only by the constant factor sqrt(n - 1), so variance-explained ratios (sdev^2 / sum(sdev^2)) are identical either way.

Usage

sdev(x)

Arguments

x

the model fit

Value

a numeric vector of singular values (one per component)

Note

The name sdev is retained for backward compatibility and prcomp-familiarity; the value stored is the raw singular value. A future major version may switch to the prcomp scaling (d / sqrt(n - 1)); do not rely on the absolute magnitude across versions – use ratios, or rescale explicitly.