point_confusion_matrix when both ground truth and predicted anomalies lists are empty.

[Linh-nk]

• Orion version: 0.5.1
• Python version: 3.8.16
• Operating System: Darwin Kernel Version 21.6.0

Description

When both ground truth and predicted anomalies lists are empty, point_confusion_matrix function gave ValueError: min() arg is an empty sequence

What I Did

Paste the command(s) you ran and the output.
If there was a crash, please include the traceback here.

[sarahmish]

we should add a condition to handle this situation.

expected behavior would be to return an empty confusion matrix, i.e. None.

[DMZ22]

Opened #672 for this, following your spec of returning an empty confusion matrix.

Two notes from digging in: the min()/max() in _point_partition run before the start/end overrides, so the crash also happens when a range is supplied (point_confusion_matrix([], [], start=0, end=9) raises today). The PR reorders that, which means a known range is enough to build the partition and yields [10, 0, 0, 0] — all true negatives — rather than an empty matrix.

The empty matrix is returned as (None, 0, 0, 0) so it keeps the 4-tuple the metrics unpack, matching the (None, fp, fn, tp) shape contextual.py already uses when true negatives can't be determined. Precision/recall/f1 then fall through to nan via the existing zero-division handling, and accuracy raises the existing missing-true-negatives error.