off-by-one in MultiClassJaccard ignore_index handling

[martinmeinke]

🐛 Bug

I believe the bugfix #1386 introduced aother bug when computing MulticlassJaccardIndex.

This bug appears when num_classes == ignore_index.

To Reproduce

Code sample

import torch
from torchmetrics.classification import MulticlassJaccardIndex
target = torch.tensor([2, 1, 0, 0])
preds = torch.tensor([2, 1, 0, 1])
metric = MulticlassJaccardIndex(num_classes=3, ignore_index=3)
metric(preds, target)
print(metric.confmat)
print(metric.compute())

results in:

IndexError: index 3 is out of bounds for dimension 0 with size 3

Expected behavior

setting ignore_index == num_classes should work

Environment

pip show torchmetrics
Name: torchmetrics
Version: 0.11.4

Fix: https://github.com/Lightning-AI/torchmetrics/pull/1860/files

[github-actions]

Hi! thanks for your contribution!, great first issue!