pod.compute_kriging_group_loo_means

compute_kriging_group_loo_means(gpr, X_2d, y, groups)

Leave-one-group-out Kriging means, for data that contains repeated points.

Each group (e.g. all copies of one original observation in a bootstrap resample) is removed together, so a point is never predicted from its own duplicate. With I the indices of a group and B as in compute_kriging_loo_residuals, the prediction is mu_{-I} = -(B_II)^-1 B_{I,rest} y_rest, the block form used by [UQLab] uq_Kriging_calc_KFold. It reduces to [M26] Eq. 10 when every group has one member.

Parameters

Name	Type	Description	Default
gpr	GaussianProcessRegressor	A fitted Kriging model.	required
X_2d	np.ndarray	Training inputs (N_samples, N_features).	required
y	np.ndarray	Training responses (N_samples,).	required
groups	np.ndarray	Integer group label per sample.	required

Returns

Name	Type	Description
	np.ndarray	np.ndarray: The leave-one-group-out mean for every sample, in response units.