All metrics computing a curve should always send back result in the order Xaxis - Yaxis - Everything_Else

[Yann-CV]

🐛 Bug

The metrics giving access to curve are not always returning Xaxis - Yaxis in the same order (considering the common usage of the curve).

To Reproduce

pred = torch.tensor([0, 0.1, 0.8, 0.4])
target = torch.tensor([0, 1, 1, 0])

roc = ROC(task="binary")
fpr, tpr, thresholds = roc(pred, target) # in ROC curve Xaxis = fpr -------- Yaxis = tpr

pr_curve = PrecisionRecallCurve(task="binary")
precision, recall, thresholds = pr_curve(pred, target) # in precision recall curve Xaxis = recall -------- Yaxis = precision

Expected behavior

pred = torch.tensor([0, 0.1, 0.8, 0.4])
target = torch.tensor([0, 1, 1, 0])

roc = ROC(task="binary")
fpr, tpr, thresholds = roc(pred, target) # in ROC curve Xaxis = fpr -------- Yaxis = tpr

pr_curve = PrecisionRecallCurve(task="binary")
recall, precision, thresholds = pr_curve(pred, target) # in precision recall curve Xaxis = recall -------- Yaxis = precision

Environment

TorchMetrics 1.2

Additional context

We are trying to be metric agnostic during automated plot creation.
So far we added a constant object:

METRIC_AXIS = {
     ROC: {"x": 0, "y": 1},
     PrecisionRecallCurve: {"x": 1, "y": 0},
}

[SkafteNicki]

The goal here has been to be consistent with the implementations in sklearn:
https://scikit-learn.org/stable/modules/generated/sklearn.metrics.precision_recall_curve.html
https://scikit-learn.org/stable/modules/generated/sklearn.metrics.precision_recall_curve.html
which has this ordering.

[Yann-CV]

@SkafteNicki see #2207 (comment) (thanks again for the answer)