R/auc.R

In fmoyaj/aucvar: Calculate the variance of

#' Calculate the area under the ROC curve
#'
#' @param p_pred A vector of the predicted probabilities for the observations
#' in the data set.
#' @param label_true A vector containing the true class labels in the data set.
#'
#' @return The sample estimate of the area under ROC curve, realized based on a two-
#' sample U-statistic estimator that is akin to the Mann-Whitney two-sample U-statistic.
#' It is also the estimated probability that the binary classifier will score a
#' randomly drawn positive sample higher than a randomly drawn negative sample.
#'
#' @export
#'
#' @examples
#' library(aucvar)
#' my_data <- na.omit(breastcancer) # Omit NA values
#' full_model <- glm(Class~`Clump Thickness`+ `Uniformity of Cell Size` +
#' `Uniformity of Cell Shape`+ `Marginal Adhesion` + `Single Epithelial Cell Size`
#' + `Bare Nuclei` + `Bland Chromatin` + `Normal Nucleoli` + `Mitoses`,
#'  family=binomial(link="logit"), data=my_data)
#' prob <- predict(full_model, type="response")
#' labels <- my_data$Class
#' auc(prob, labels)
#'
#' @references
#' \cite{H. B. Mann and D. R. Whitney (1947). On a test of whether one of two random
#' variables is stochastically larger than the other. Annals of Mathematical Statistics 18:
#'  50-60.}
auc <- function(p_pred, label_true){

  # Check arguments
  if((!is.vector(p_pred) || !is.vector(label_true)) && !is.factor(label_true))
  {
    base::stop("p_pred and label_true must be vectors or a factor")
  }

  # label_true must have the same length as p_pred
  pred_length <- base::length(p_pred)
  label_length <- base::length(label_true)
  if(pred_length != label_length){
    base::stop("p_pred and label_true must have the same length")
  }

  # label_true must have two levels
  if (!base::length(base::levels(base::as.factor(label_true)))){
    base::stop("label_true must have two levels")
  }

  pred <-ROCR::prediction(p_pred, label_true)
  perf <- ROCR::performance(pred, measure = "auc", fpr.stop=1) # Stop if FPR is 1
  return(base::as.numeric(perf@y.values))
}