oilmm_mll

gpjax.models.oilmm_mll(model, data)

Log marginal likelihood for the OILMM.

Implements Prop. 9 from Bruinsma et al. (2020):

log p(Y) = correction_terms + sum_i log N((TY)_i | 0, K_i + noise_i I_n)

The correction terms prevent the projection from collapsing and account for data in the (p - m) dimensions orthogonal to the mixing matrix.

Like the other objectives, this is a one-line view of the conditioned process: the evidence is owned by OILMMPosterior and computed once when the model is conditioned.

Parameters:

• model (OILMMModel) – OILMMModel with parameters to evaluate.

• data (Dataset) – Training data with X [N, D] and y [N, P].

Returns:

Scalar log marginal likelihood.

Return type:

ScalarFloat