Model selection using CV

cv_model_selection.R

# snippet: not all code included

# revised loop to do cross-validation
for (m in 1:length(allModels)) {
  allpredicted = rep(NA,n)   # storage for honest predictions
  
  for (ii in 1: nfolds) {    # ii is an easier string to search for index
    groupii = (cvgroups == ii)
    trainset = bodyfat[!groupii,]  # all data EXCEPT for group ii
    testset = bodyfat[groupii, ]   # data in group ii
    modelfit = lm(allModels[[m]], data=trainset) # fit to train set
    predicted = predict(modelfit, newdata = testset)   # predict for test set
    allpredicted[groupii] = predicted              # store in ordered locations
  }
  
  y = bodyfat$BodyFatSiri
  CVvalue = mean((y - allpredicted)^2); CVvalue  
  allCVvalues[m] = CVvalue
  
  obs.pred = data.frame(y = y, yhat = allpredicted)
  allplots[[m]] <- gf_point(y ~ yhat, data = obs.pred) %>%
    gf_labs(title = paste("Compare Observed to Predicted for Model", m),
            y = "Observed", x = "Predicted")
  
}

allCVvalues