Cross Validation

CV <- function (df, n) {
  # Cualquier seed debe funcionar, se probo con el del ejemplo de clase para comparar
  # set.seed(161)
  
  index <- 1:nrow(df)
  shuffle_index <- sample(index)
  shuffle_df <- df[shuffle_index,]
  
  delta <- nrow(df)/n
  
  result <- data.frame()
  
  for (i in 1:n) {
    if (i == 1) {
      test <- shuffle_df[1:delta,]
      train <- shuffle_df[-(1:delta),]
    } else {
      test <- shuffle_df[((i-1)*delta):(i*delta),]
      train <- shuffle_df[-(((i-1)*delta):(i*delta)),]
    }
    
    fit <- lm(data = train)
    ptest <- fit$coefficients[1]
    
    for (j in 2:ncol(df)) {
      ptest <- ptest + fit$coefficients[j]*test[j]
    }
    
    RMSE<-sqrt( sum( (test[1] - as.vector(ptest))^2 )/nrow(test) )
    result <- rbind(result, c(RMSE))
  }
  
  colnames(result) <- c("RMS")
  return(result)
}

faith <- data.frame(faithful$waiting, faithful$eruptions)
CV(faith, 5)

CV(faith, 10)