Computes separate DKGE bases on the CPCA design-aligned or residual parts of the compressed covariance. Helpful for decomposing the latent space into interpretable sub-structures.
Usage
dkge_fit_cpca(
fit,
blocks = NULL,
T = NULL,
part = c("design", "resid", "both"),
rank = NULL,
ridge = 0
)Arguments
- fit
A `dkge` object from [dkge_fit()] or [dkge()].
- blocks
Optional integer vector of effect indices defining the subspace (used when `T` is not supplied).
- T
Optional explicit qxq0 basis matrix; overrides `blocks` when given.
- part
Which portion to return: "design", "resid", or "both".
- rank
Target rank for the returned bases (defaults to `ncol(fit$U)`).
- ridge
Optional ridge term applied before the eigen decompositions.