R/CV.R

In mlegp: Maximum Likelihood Estimates of Gaussian Processes

`CV` <-
function(gp, predictObserved = TRUE, verbose = FALSE) {

  Var = solve(gp$invVarMatrix)

  Z.pred = matrix(0, ncol=1, nrow = gp$numObs)
  pred.error = matrix(1, ncol=1, nrow = gp$numObs)
  complete = matrix(FALSE, ncol=1, nrow = gp$numObs)

  i = 1
  for (i in 1:gp$numObs) {
	if (verbose) {
        s = paste("cross validate at index # ", i)
     	cat(s)
     	cat("\n")
	}

	if (complete[i]) next

        index = matrix(TRUE, gp$numObs)
	for (j in 1:gp$numObs) {
		index[j] = !all(gp$X[i,] == gp$X[j,])
	}
	
        newX = gp$X[index,]
        if (!is.matrix(newX)) newX = as.matrix(newX)

        p = predictNewYCV(gp, newX, gp$Z[index],gp$mu[index],gp$X[i,], Var[index,index])
     
        n = 0
        if(predictObserved) {
      	n = gp$nugget
        if (length(gp$nugget) > 1) {
	      n = gp$nugget[i]
        }
      }

      v2 = calcPredictionErrorCV(gp,newX, gp$X[i,], Var[index,index],n) 
      Z.pred[!index] = p
      pred.error[!index] = v2
		
      complete = complete | !index

  }

  pred.error[pred.error < 0] = 0

  return (cbind(Z.pred, pred.error))
}