It counts false positives over the full matrix (including the diagonal) but normalises by n*(n-1) - total_positives, and total_positives also includes the diagonal. Hence the FPR of 1.357 above — an impossible value.
This function is exported, documented, and used by the CausationEntropy's own evaluation tests, so any accuracy numbers computed with it are unreliable.
It counts false positives over the full matrix (including the diagonal) but normalises by
n*(n-1) - total_positives, andtotal_positivesalso includes the diagonal. Hence the FPR of 1.357 above — an impossible value.This function is exported, documented, and used by the CausationEntropy's own evaluation tests, so any accuracy numbers computed with it are unreliable.