Header-only AUROC/ROC-AUC in C++11

1. Include include/AUROC.hpp

2. Call AUROC(label, score, n) and get the result

C++11 and standard libraries

• T1, T2: Type of array elements, should be a numerical type

• label: Array of ground truth labels, 0 is negative, 1 is positive

• score: Array of predicted scores, can be any real finite number

• n: Number of elements in the array, I assume it’s correct

• Return: AUROC/ROC-AUC score, range [0.0, 1.0]; or NaN if inputs are problematic

1. How fast is it?

   On my desktop PC, ~1.35s for ~8M records.
   The only sort takes ~80% of the running time.

2. How accurate is it?

   From tests and my daily usage, the difference with sklearn.metrics.roc_auc_score is mostly less than 1e-15.

3. Where are the test code and data?

   I didn’t upload them to GitHub.

4. Why not using plain C?

   qsort array index might be very tedious.
   Otherwise, C11 is required for the additional argument passed to the comparison function.