gptk (version 1.0)

gpLogLikeGradients: Compute the gradients for the parameters and X.

Description

computes the gradients of the Gaussian process log likelihood with respect to the model parameters (and optionally, as above with respect to inducing variables and input data) given the target data, input data and inducing variable locations.

Usage

values <- gpLogLikeGradients(model, X=model$X, M, X_u, gX_u.return=FALSE, gX.return=FALSE, g_beta.return=FALSE)

Arguments

model

the model structure for which gradients are computed.

X

the input data locations for which gradients are computed.

M

the scaled and bias removed target data for which the gradients are computed.

X_u

the inducing variable locations for which gradients are computed.

gX_u

flag to return the gradient of the log likelihood with respect to the inducing variables. If inducing variables aren't being used this returns zero.

gX

flag to return the gradient of the log likelihood with respect to the input data locations.

g_beta

flag to return the gradient of the log likelihood with respect to beta.

Value

• gParam
  contains the gradient of the log likelihood with respect to the model parameters (including any gradients with respect to beta).

See Also

gpLogLikelihood.

Examples

## missing