Fix model_performance() with type = 'roc' (for binary classificat…

…ion problem) (#19)

* change `times` vector in `model_performance()` roc

* fix `model_performance()` roc calculation

* Update NEWS.md

Co-authored-by: Mikołaj Spytek <55801784+mikolajsp@users.noreply.github.com>

NEWS.md

@@ -4,10 +4,12 @@
 * added references to used methods ([#5](https://github.com/ModelOriented/survex/issues/5))
 * changed the package used to draw complex plots from `gridExtra` to `patchwork` ([#7](https://github.com/ModelOriented/survex/pull/7))
 * fixed subtitles in plots ([#11](https://github.com/ModelOriented/survex/issues/11))
+* fixed calculating of roc curves for classification problems
 * added wrapper function for measures provided by `mlr3proba` ([#10](https://github.com/ModelOriented/survex/issues/10))
 * created vignette showing how to use `mlr3proba` with `survex`
 * fixed incompatibility with new ggplot2 version 3.4 
 
+
 # survex 0.1.1
 * The `survex` package is now public
 * `model_parts`, `model_profile`, `predict_parts`, `predict_profile` explanations implemented

R/model_performance.R

@@ -54,7 +54,7 @@
 #'
 #' plot(rsf_ranger_model_performance, cph_model_performance, rsf_src_model_performance)
 #'
-#' cph_model_performance_roc <- model_performance(cph_exp, type = "roc", times = c(100, 500, 1200))
+#' cph_model_performance_roc <- model_performance(cph_exp, type = "roc", times = c(100, 250, 500))
 #' plot(cph_model_performance_roc)
 #'
 #' @rdname model_performance.surv_explainer

R/surv_model_performance.R

@@ -31,12 +31,19 @@ surv_model_performance <- function(explainer, ..., times = NULL, type = "metrics
     else {
         if (is.null(times)) stop("Times cannot be NULL for type `roc`")
         rocs <- lapply(times, function(time) {
-            labels <- 1 - explainer$y[, 2]
-            scores <- explainer$predict_survival_function(explainer$model, newdata, time)
-            labels <- labels[order(scores, decreasing = TRUE)]
-            cbind(time = time, data.frame(TPR = cumsum(labels) / sum(labels), FPR = cumsum(!labels) / sum(!labels), labels))
+            censored_earlier_mask <- (explainer$y[, 1] < time & explainer$y[, 2] == 0)
+            event_later_mask <- explainer$y[, 1] > time
+            newdata_t <- newdata[!censored_earlier_mask, ]
+            labels <- explainer$y[,2]
+            labels[event_later_mask] <- 0
+            labels <- labels[!censored_earlier_mask]
+            scores <- explainer$predict_survival_function(explainer$model, newdata_t, time)
+            labels <- labels[order(scores, decreasing = FALSE)]
+            cbind(time = time, data.frame(TPR = cumsum(labels) / sum(labels),
+                                          FPR = cumsum(!labels) / sum(!labels), labels))
         })
 
+
         rocs_df <- do.call(rbind, rocs)
 
         class(rocs_df) <- c("surv_model_performance_rocs", class(rocs_df))