validate: Predictive statistics for small validation data

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

Description

This function takes inferential results from Infer and computes predictive statistics on a validation data set. These include RMS, PMCC, CR1 and CR2 as defined in Sahu and Mardia (2005), and the Mahalanobis distance Mahalanobis, the eigenvalue errors DE and the pivoted Cholesky errors DPC as described in Bastos and O'Hagan (2009).

Usage

validate(Results, G, sim_obs = F, ...)

## S4 method for signature 'list,Graph_2nodes'
validate(Results, G, sim_obs = F)

## S4 method for signature 'matrix,Graph_2nodes'
validate(Results, G, sim_obs = F)

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.
sim_obs	if set to T, the observations are ignored and ‘ideal’ simulated observations are used instead. This option can be used to effect a ‘Turing test’, where the real-data case and the ideal case can be compared side-by-side.

Value

A data frame with the predictive statistics as described above.

References

Sahu, S. K., & Mardia, K. V. (2005). A Bayesian kriged Kalman model for short-term forecasting of air pollution levels. Journal of the Royal Statistical Society: Series C (Applied Statistics), 54(1), 223-244. Bastos, L. S. and O'Hagan, A. (2008). Diagnostics for Gaussian process emulators. Technometrics 51, 425-438.

Examples

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 split into a training/validation set
split_data <- split_validation(icesat_obs,sample=500,common=0, t==2)
icesat_validation <- split_data$O_val
icesat_training <- split_data$O_pruned

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

## Now we validate the results with icesat
L1 <- link(SURF,icesat_validation)
e <- link_list(list(L1))
v <- block_list(list(G1=SURF,O1=icesat_validation))
G <- Graph(e=e,v=v)
G_reduced <- compress(G)
val_results <- validate(Results,G_reduced,sim_obs=F)

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