#6: addressing shift in PPV by one row (#7)

Address #6. 

Since it was chosen by design to assess `score > x` for each `x in
score`, I don't think we can naturally end up with a PPV at prevalence.

Other option would be to inflate the resulting dataset manually for this
boundary value, but than we would need to artificially add one more
score and one more percentile to match the number of rows in the
resulting data frame.

Author: slamao

R/PV.R

@@ -32,7 +32,7 @@ nonParametricPV <- function(outcome, score) {
   thresh.predictions <- lapply(score, function(x) as.numeric(score > x))
 
   ppv <- vapply(
-    thresh.predictions[1:(length(score) - 1)],
+    thresh.predictions,
     function(x) {
       yardstick::ppv_vec(
         truth = factor(outcome, levels = c("1", "0")),
@@ -44,7 +44,7 @@ nonParametricPV <- function(outcome, score) {
   )
 
   npv <- vapply(
-    thresh.predictions[1:(length(score) - 1)],
+    thresh.predictions,
     function(x) {
       yardstick::npv_vec(
         truth = factor(outcome, levels = c("1", "0")),
@@ -58,8 +58,8 @@ nonParametricPV <- function(outcome, score) {
   threshold.data <- data.frame(
     score = score,
     percentile = ecdf(score)(score),
-    PPV = c(prev, ppv),
-    NPV = c(npv, 1 - prev)
+    PPV = ppv,
+    NPV = npv
   ) %>%
     mutate(MNPV = 1 - .data$NPV) %>%
     tidyr::fill(