Returns the gradient of the Matern kernel.
g = covgrad_matern(xTrain,covf)
g = covgrad_matern computes the derivative of the Matern kernel. If
an input xTrain is present, then it is assumed that the gradient of
the matrix cov_matern(net.xTrain,net.BV) is computed.
The kernel parameters are taken from the global variable net.
With two parameters it is assumed that the covariance matrix of the BV
set and xTrain has been computed and a kernel matrix computation is
spared.
A further reduction in memory requirements is when we ask only a single column of the gradient matrix, specified by a third argument:
g = covgrad_matern(xTrain,covf,iPar)
If iPar==0 then all gradient elements are computed and returned. Often
this operation fails due to memory requirements.
There is no analytical formula to differentiate the Bessel function with respect to its order, thus numerical (finite difference) approximation is employed.
ogpcovgrad, cov_matern, demogp_matern
Copyright (c) Lehel Csató (2001-2004)