Fit DKGE from precomputed subject effect kernels — dkge_fit_from

Converts a list of subject-level effect kernels \(K_s \in \mathbb{R}^{q \times q}\) into synthetic GLM inputs that reuse [dkge_fit()] without modifying the core implementation. Each kernel is factorised into a symmetric square root, scaled to keep the pooled design metric unchanged, and paired with an identity design matrix so the resulting DKGE fit matches the supplied kernels.

Usage

dkge_fit_from_kernels(
  K_list,
  effect_ids,
  subject_ids = NULL,
  design_kernel = NULL,
  sqrt_tol = 1e-10,
  ...
)

Arguments

K_list: List of symmetric positive semi-definite matrices sharing the same effect ordering.
effect_ids: Character vector of length \(q\) naming the shared effect (anchor) indices.
subject_ids: Optional character vector naming subjects. Defaults to the names of `K_list` or sequential identifiers.
design_kernel: Optional design kernel passed to [dkge_fit()]. Defaults to the \(q \times q\) identity matrix, which matches the whitened anchor setup.
sqrt_tol: Eigenvalue tolerance used when extracting square roots.
...: Additional arguments forwarded to [dkge_fit()].

Value

A `dkge` object identical to one obtained from [dkge_fit()], with provenance annotated to record the kernel-driven construction.

Examples

# \donttest{
q <- 5
Ks <- replicate(3, {
  X <- matrix(rnorm(q * q), q)
  S <- crossprod(X)
  S / sqrt(sum(diag(S)))
}, simplify = FALSE)
fit <- dkge_fit_from_kernels(Ks, effect_ids = paste0("z", seq_len(q)))
# }