R/logConROC.R

In logcondens: Estimate a Log-Concave Probability Density from Iid Observations

logConROC <- function(cases, controls, grid, smooth = TRUE) {
  m <- length(controls)
  n <- length(cases)

  res0 <- logConDens(controls, smoothed = smooth, print = FALSE)
  res1 <- logConDens(cases, smoothed = smooth, print = FALSE)

  qs0 <- quantilesLogConDens(1 - grid, res0)
  F1 <- evaluateLogConDens(qs0[, "quantile"], res1, which = c(3, 5), gam = NULL, print = FALSE)

  ## ROC curve
  rocLogCon <- 1 - F1[, "CDF"]

  ## manually adjust
  rocLogCon[grid == 0] <- 0
  rocLogCon[grid == 1] <- 1

  ## compute AUC
  auc <- integrate(f = ROCx, lower = 0, upper = 1, res0 = res0, res1 = res1, smooth = FALSE)$value

  ## same program for smoothed version
  rocLogConSmooth <- NA
  aucSmooth <- NA
  if (identical(smooth, TRUE)) {
    rocLogConSmooth <- 1 - F1[, "smooth.CDF"]
    rocLogConSmooth[grid == 0] <- 0
    rocLogConSmooth[grid == 1] <- 1
    aucSmooth <- integrate(f = ROCx, lower = 0, upper = 1, res0 = res0, res1 = res1, smooth = TRUE)$value
  }

  return(list("m" = m, "n" = n, "fROC" = rocLogCon, "fROC.smooth" = rocLogConSmooth, "auc" = auc, "aucSmooth" = aucSmooth, "res0" = res0, "res1" = res1))
}