R/my_knn_cv.R

In samcho11/STAT302package: What the Package Does (One Line, Title Case)

#' k-Nearest Neighbors Cross-Validation
#'
#' This function splits the data into k groups for evaluating test set and
#' training sets to predict the best value referring the neighbors.
#'
#' @param train Input data frame.
#' @param cl True class value of your training data.
#' @param k_nn Integer representing the number of neighbors.
#' @param k_cv Integer representing the number of folds.
#'
#' @keywords prediction
#'
#' @return A list consists of a vector \code{class} comprised of prediction made
#'         by \code{knn()} function from class package, and misclassification
#'         rate \code{cv_err}, a value between 0 and 1 representing the
#'         proportion of observations that were classified incorrectly.
#'
#' @importFrom stats model.frame model.matrix model.response na.omit predict pt sd
#' @importFrom dplyr filter select pull
#' @importFrom class knn
#'
#' @examples
#' penguins <- STAT302package::my_penguins
#' penguins2 <- na.omit(penguins)
#' my_cl <- penguins2 %>% dplyr::pull(species)
#' result_nn1 <- my_knn_cv(penguins2[, 3:6], my_cl, 1, 5)
#'
#'
#' @export
my_knn_cv <- function(train, cl, k_nn, k_cv) {
  # Split data in k_cv parts, randomly
  k <- k_cv
  fold <- sample(rep(1:k, length = nrow(train)))
  split_data <- data.frame("x" = train, "split" = fold)
  split_class <- data.frame("y" = cl, "split" = fold)

  # Omit NA values
  #split_data <- na.omit(split_data)

  # Empty dataset for recording
  cv_err <- rep(NA, k_cv)

  # Iterate through k_cv times
  for (i in 1:k) {
    # Assign training and test variable
    data_train <- split_data %>% dplyr::filter(split != i)
    data_train <- data_train %>% dplyr::select(-split)

    data_test <- split_data %>% dplyr::filter(split == i)
    data_test <- data_test %>% dplyr::select(-split)

    class_train <- split_class %>% dplyr::filter(split != i)
    class_train <- class_train %>% dplyr::select(-split)

    class_test <- split_class %>% dplyr::filter(split == i)
    class_test <- class_test %>% dplyr::select(-split)

    result <- knn(data_train, data_test, class_train[, 1], k_nn)
    cv_err[i] <- mean(result != class_test[, 1])
  }
  full_knn <- knn(train, train, cl, k_nn)
  output_cv <- mean(cv_err)
  my_output <- list("class" = full_knn, "cv_err" = output_cv)
  return(my_output)
}