predict_gp: Get predictions for a model fit with optimize_gp.

In nategarton13/sparseRGPs: Fit Sparse Gaussian Processes

Description

This function directly takes the output from optimize_gp along with prediction locations and estimates the posterior distribution of the latent function conditional on the optimized covariance parameters and knots.

Usage

predict_gp(mod, x_pred, mu_pred = NA, full_cov, vi = FALSE)

Arguments

mod	A list with the same elements and names as that generated by the optimize_gp function. A variable storing the output from the optimize_gp function can be passed directly to this argument.
x_pred	The matrix of prediction locations. These can also be observed data locations.
mu_pred	A vector: the marginal mean of the latent GP at the prediction locations.
full_cov	Logical value indicating whether to return the full predictive covariance matrix if TRUE, or to return just the marginal variances if FALSE.
vi	Logical value indicating whether predictions should be generated from the variational approximation or not. This should be true only if you selected to use variational inference in optimize_gp.

Value

List with the following components:

pred: a list containing pred$pred_mean which is the vector of the predicted means of the latent function, and pred$pred_var which are the predicted variances of the latent function.

sparse: logical value indicating whether predictions were made based on a sparse model

family: cahracter string indicating the distribution of the data conditional on the latent function

x_pred: the matrix of prediction locations

inverse_link: The function that maps the latent function to the conditional mean of the data distribution. In the case of binary data this is the logistic function. In the case of Poisson data this is m*exp(f(x)).

nategarton13/sparseRGPs documentation built on May 27, 2020, 9:46 a.m.