R/trainNeuralNet.R

In ghouljd/apnnClassifier: Multiclass classifier based on neural networks.

#' Neural network training of the pnn package for classification
#'
#' @param train_set Training set. (Required)
#' @param test_set Testing set. (Required)
#' @param category_column Category column (Not required. Default value: 1)
#' @param sigma Optimum value for the activation function of the neural network. (Not required)
#'
#' @return pnn \code{list} trained with the classified testing set and network performance statistics.
#'
#' @examples
#' library(apnnClassifier)
#' data(trainData, testData)
#' testData <- as.matrix(testData)
#' # Basic usage.
#' pnn <- trainNeuralNet(train_set = trainData, test_set = testData)
#' # If you know the approximate optimal value and the sorter column is not in the first position of the set.
#' pnn <- trainNeuralNet(train_set = trainData, test_set = testData, category_column =  *sorter column*, sigma = *sigma value*)
#' View(pnn)
#' @export
# It is responsible for executing the prediction of the probabilistic neural network.

trainNeuralNet <- function(train_set, test_set, category_column = 1, sigma) {
  if(missing(train_set))
    stop("The training set shouldn't be empty or null.")
  else if(missing(test_set))
    stop("The testing set shouldn't be empty or null.")
  else if(typeof(train_set) != "list")
    stop("The training set isn't valid.")
  else if (!requireNamespace("pnn", quietly = TRUE))
    stop("Package \"pnn\" needed for this function to work. Please install it.", call. = FALSE)

  print("Begin learning process.")
  pnn = pnn::learn(train_set, category.column = category_column)

  if(missing(sigma)){
    print("Finding optimized minimum value.")
    if (!requireNamespace("rgenoud", quietly = TRUE))
      stop("Package \"rgenoud\" needed for this function to work. Please install it.", call. = FALSE)
    pnn = pnn::smooth(pnn)
  }
  else
    pnn = pnn::smooth(pnn, sigma)
  pnn = pnn::perf(pnn)
  print("End learning process.")
  print(paste("Success rate: ", pnn$success_rate * 100))
  print("Begin testing set evaluation.")
  size <- length(test_set[,1])
  output <- data.frame(category = numeric(size), propability = numeric(size))
  for (index in 1:size) {
    predict <- pnn::guess(pnn, test_set[index,])
    output$category[index] <- ifelse(!is.na(predict), predict$category, predict)
    output$propability[index] <- ifelse(!is.na(predict), max(predict$probabilities), 0)
  }

  pnn$output = output
  print("End process.")
  return(pnn)
}