Compare quantum, RBF, and linear kernels on the same feature set

Computes three Gram matrices on the same PCA-reduced embeddings and reports their pairwise Frobenius distance and eigenvalue divergence. Useful for understanding whether the quantum kernel is materially different from the classical RBF at the same feature space (if not, the quantum lift is cosmetic).

Usage

qf_kernel_compare(X_q, reps = 2L, rbf_sigma = NULL)

Arguments

X_q

PCA-reduced, pi-scaled feature matrix (the X_q element of qf_embed_reduce()).

reps

Integer; ZZFeatureMap repetitions. Default 2L.

rbf_sigma

Numeric; RBF kernel bandwidth. If NULL, uses the median heuristic median( || x_i - x_j || ) over the training set.

Value

Named list with K_quantum, K_rbf, K_linear, and a diagnostics data frame summarising pairwise distances + effective rank.