这是一个关于稀疏高斯过程的matlab源码，可以用于计算测试输入的高斯预测值。- spgp_pred　computes the SPGP predictive distribution for a set of

test inputs. You need to supply a set of　pseudo-inputs　or　basis

vectors　for the approximation, and suitable hyperparameters for the

covariance. You can use any method you like for finding the

pseudo-inputs , with the simplest obviously being a random subset of

the data. It is coded for Gaussian covariance function, but you could

very easily alter this. It is also fine to use for high dimensional

data sets.



spgp_lik　is the SPGP (negative) marginal likelihood and gradients

with respect to pseudo-inputs and hyperparameters. So you can use this

if you wish to try to optimize the positioning of pseudo-inputs and

find good hyperparameters, before using　spgp_pred . I would recommend

initializing the pseudo-inputs on a random subset of the data, and

initializing the hyperparameters sensibly. Its current limitations are

that 1) it is slow and memory intensive for high dimensional data sets

2) it is heavi