Project subject clusters into the DKGE latent space

For each subject, computes \(Z_s = B_s^\top K U\) so that every row represents a subject cluster embedded in the \(r\)-dimensional DKGE latent space. These projections are commonly used for training classifiers or computing Haufe-style encoders.

Usage

dkge_project_clusters_to_latent(fit)

Arguments

fit

Fitted object of class `dkge` containing `Btil`, `K`, and `U`.

Value

A list of length `S` (number of subjects). Element `s` is a `P_s x r` matrix holding the latent representation of subject `s`'s clusters.

Examples

# \donttest{
toy <- dkge_sim_toy(
  factors = list(A = list(L = 2), B = list(L = 3)),
  active_terms = c("A", "B"), S = 3, P = 20, snr = 5
)
fit <- dkge(toy$B_list, toy$X_list, kernel = toy$K, rank = 2)
#> Warning: Argument 'kernel' is deprecated; use 'K' instead.
Z_list <- dkge_project_clusters_to_latent(fit)
dim(Z_list[[1]])  # P_1 x r matrix
#> [1] 20  2
# }