Compute empirical ROC vertices with whole tied-score blocks

Higher scores always indicate the named positive class. Consuming a tied block at once yields trapezoidal AUC equivalent to half credit for ties. The direction is never selected from the observed performance.

Usage

roc_vertices(x)

Arguments

x

List: Normalized labels and probabilities from roc_probabilities().

Value

Data frame of class, FPR, TPR, AUC, and omitted-pair counts.

Examples

roc_vertices(roc_probabilities(factor(c("no", "yes")), c(0.1, 0.9)))

#>   class fpr tpr auc omitted
#> 1   yes   0   0   1       0
#> 2   yes   0   1   1       0
#> 3   yes   1   1   1       0