Puts hyperparameters back into the Sparse OGP.
ogpunpak(hp)
ogpunpak(hp) takes the GLOBAL Gaussian process structure net,
the GLOBAL structure ep of TAP/EP parameters together with a
hyperparameter vector hp, and returns a Gaussian Process identical to
the input model, except that the covariance bias bias, the input
weight vector parametrised by net.kpar(1:net.nin) and the covariance
function specific parameters in kpar have all been set to the
corresponding elements of hp. The function updates the inverse kernel
matrix.
The TAP/EP parameters are computed only if we ask for it -- no computation time is wasted.
ogp, ogppak, ogpevidgrad, ogptrain
Copyright (c) Lehel Csató (2001-2004)