Enum RandomEmbeddingKernel

Which of REMBO's two kernels (Wang et al. 2016, §3.2) the inner Bayesian optimizer uses.

public enum RandomEmbeddingKernel

Fields

LowDimensional = 0: The low-dimensional kernel: the inner sampler models the objective as a function of the latent point y directly (a GP in d dimensions). Cheap and the paper's choice for very high extrinsic dimension, but it can waste effort on the large part of the box Y that projects onto the boundary of the real box (two far-apart y's can map to nearly the same point).
HighDimensional = 1: The high-dimensional kernel: the GP measures similarity in the projected real parameter space p_X(A·y), so y's that map to the same clipped point are treated as identical — removing the boundary over-exploration and making a single embedding far more reliable. It fits a GP in the real (embedded) parameter space and optimizes expected improvement over the reachable subspace, so it is best suited to moderate extrinsic dimension, and it also embeds categorical parameters (the §3.2 categorical/Hamming kernel).