R/MSEZ.R

In CorReg: Linear Regression Based on Linear Structure Between Variables

# ' Computes the MSE on the joint distribution of the dataset
# ' @param X the dataset to predict
# ' @param X_appr an optional learning set
# ' @param B the structure tested (if known)
# ' @param Z binary adjacency matrix of the structure (size p)
# ' @param scale boolean defining wether the dataset has to be scaled or not
# ' @export
MSEZ <- function(X = X, X_appr = NULL, B = NULL, Z = Z, scale = TRUE) {
   X = scale(X)
   I1 = which(colSums(Z) == 0)
   X1 = cbind(1, X[, I1])
   X2 = X[, -I1]
   I2 = (1:ncol(Z))[-I1]
   res = sum(apply(X1, 2, var))

   if (is.null(B)) {
      if (is.null(X_appr)) {X_appr = X
      } else {
         X_appr = scale(X_appr)
      }
      B = hatB(Z = Z, X = X_appr)
   }
   X2 = X[, -colSums(Z) != 0]
   res = res + sum(apply(X2 - X1 %*% B[-(I2 + 1), I2], 2, var))
   return(res)
}