Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix weighted AUROC on gpu #606

Merged
merged 5 commits into from
Nov 8, 2021
Merged

Fix weighted AUROC on gpu #606

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
9fe12e2
fix
SkafteNicki Nov 4, 2021
d07a192
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Nov 4, 2021
b7084d7
changelog
SkafteNicki Nov 4, 2021
13e84e6
Merge branch 'fix_auroc' of https://github.com/PyTorchLightning/metri…
SkafteNicki Nov 4, 2021
9b3af99
Merge branch 'master' into fix_auroc
Borda Nov 8, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
- Fix empty predictions in MAP metric ([#594](https://github.com/PyTorchLightning/metrics/pull/594))


- Fix edge case of AUROC with `average=weighted` on GPU ([#606](https://github.com/PyTorchLightning/metrics/pull/606))


## [0.6.0] - 2021-10-28

### Added
Expand Down
10 changes: 7 additions & 3 deletions tests/classification/test_auroc.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -185,11 +185,15 @@ def test_error_multiclass_no_num_classes():
_ = auroc(torch.randn(20, 3).softmax(dim=-1), torch.randint(3, (20,)))


def test_weighted_with_empty_classes():
@pytest.mark.parametrize("device", ["cpu", "cuda"])
def test_weighted_with_empty_classes(device):
"""Tests that weighted multiclass AUROC calculation yields the same results if a new but empty class exists.

Tests that the proper warnings and errors are raised
"""
if not torch.cuda.is_available() and device == "cuda":
pytest.skip("Test requires gpu to run")

preds = torch.tensor(
[
[0.90, 0.05, 0.05],
Expand All @@ -198,8 +202,8 @@ def test_weighted_with_empty_classes():
[0.85, 0.05, 0.10],
[0.10, 0.10, 0.80],
]
)
target = torch.tensor([0, 1, 1, 2, 2])
).to(device)
target = torch.tensor([0, 1, 1, 2, 2]).to(device)
num_classes = 3
_auroc = auroc(preds, target, average="weighted", num_classes=num_classes)

Expand Down
2 changes: 1 addition & 1 deletion torchmetrics/functional/classification/auroc.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -135,7 +135,7 @@ def _auroc_compute(
raise ValueError("Detected input to `multiclass` but you did not provide `num_classes` argument")
if average == AverageMethod.WEIGHTED and len(torch.unique(target)) < num_classes:
# If one or more classes has 0 observations, we should exclude them, as its weight will be 0
target_bool_mat = torch.zeros((len(target), num_classes), dtype=bool)
target_bool_mat = torch.zeros((len(target), num_classes), dtype=bool, device=target.device)
target_bool_mat[torch.arange(len(target)), target.long()] = 1
class_observed = target_bool_mat.sum(axis=0) > 0
for c in range(num_classes):
Expand Down