pred_variance_large: Predictive variance for large data sets

In shazhe/mvst0: Multi-variate ST modelling on globe

Description

This function takes inferential results from Infer and computes the predictive variance from the posterior GMRF at locations for a very large number of data points. This function only yields valid results if the non-zero locations in each row of the incidence matrix constructed using the validation data are a subset of those used to generate the posterior precision matrix, see references.

Usage

pred_variance_large(Results, G)

## S4 method for signature 'list,Graph_2nodes'
pred_variance_large(Results, G)

## S4 method for signature 'matrix,Graph_2nodes'
pred_variance_large(Results, G)

Arguments

Results	a list generated by the function Infer
G	An object of class Graph_2nodes describing the relationship between the observations and the processes.

Value

Object of class Obs.

References

Jonathan C. Rougier, Andrew Zammit-Mangion and Nana Schoen (2014). Computation and visualisation for large-scale Gaussian updates. http://arxiv.org/abs/1406.5005.

Examples

## Not run: 
require(Matrix)
data(icesat)
data(surf_fe)

## First create observation object
icesat_obs <- Obs(df=icesat,
                 abs_lim = 5,
                 avr_method = "median",
                 box_size=100,
                 name="icesat")

## Now create GMRF defined over some FE basis
Mesh <- initFEbasis(p=surf_fe$p,
                    t=surf_fe$t,
                    M=surf_fe$M,
                    K=surf_fe$K)

mu <- matrix(0,nrow(Mesh),1)
Q <- sparseMatrix(i=1:nrow(surf_fe$p), j = 1:nrow(surf_fe$p), x = 1)

my_GMRF <- GMRF(mu = mu, Q = Q,name="SURF",t_axis = 0:6)
SURF <-GMRF_basis(G = my_GMRF, Basis = Mesh)

L1 <- link(SURF,icesat_obs)
e <- link_list(list(L1))
v <- block_list(list(O = icesat_obs, G = SURF))
G <- new("Graph",e=e,v=v)
G_reduced <- compress(G)
Results <- Infer(G_reduced)

Obs_test <- pred_variance_large(Results,G_reduced)

## End(Not run)

shazhe/mvst0 documentation built on May 29, 2019, 9:20 p.m.