LCMKernel#

class gpjax.kernels.LCMKernel(kernels, coregionalization_matrices)[source]#

Linear Model of Coregionalization kernel.

Generalises ICM by combining Q independent latent kernels, each with its own coregionalization matrix: K = Σ_q B^(q) ⊗ k_q(X, X).

When Q=1 this is equivalent to an ICM and retains Kronecker structure. When Q>1 the sum is materialised to a dense matrix.

Parameters:

kernels (list[AbstractKernel]) – List of Q latent kernels.
coregionalization_matrices (tuple[CoregionalizationMatrix, ...]) – List of Q coregionalization matrices. All must have the same num_outputs.

classmethod from_icm_components(icm_kernels)[source]#

Build an LCM from a list of ICMKernel instances.

Parameters:: icm_kernels (list) – List of ICMKernel objects. Each contributes its base_kernel and coregionalization_matrix as one LCM component.
Returns:: An LCMKernel combining all components.
Return type:: LCMKernel

property latent_kernels: tuple[AbstractKernel, ...]#: Tuple of latent kernels.