Skip to content

fix(aggregation): fix nan_stategy=disable dropping weights #3433

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鈥檒l occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
s22chan wants to merge 1 commit into
base: master
Choose a base branch
from
Open

fix(aggregation): fix nan_stategy=disable dropping weights #3433

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 1 addition & 1 deletion src/torchmetrics/aggregation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -103,7 +103,7 @@ def _cast_and_nan_check_input(
raise ValueError(f"`nan_strategy` shall be float but you pass {self.nan_strategy}")
x[nans | nans_weight] = self.nan_strategy
weight[nans | nans_weight] = 1
else:
elif weight is None:
weight = torch.ones_like(x)
return x.to(self.dtype), weight.to(self.dtype)

Expand Down
8 changes: 8 additions & 0 deletions tests/unittests/bases/test_aggregation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -163,6 +163,14 @@ def test_error_on_wrong_nan_strategy(metric_class):
metric_class(nan_strategy=[])


def test_disable_preserves_weight():
"""`nan_strategy="disable"` should only skip nan checks, not also discarding `weight`."""
metric = MeanMetric(nan_strategy="disable")
metric.update(torch.tensor(1.0), weight=torch.tensor(3.0))
metric.update(torch.tensor(0.0), weight=torch.tensor(1.0))
assert torch.allclose(metric.compute(), torch.tensor(0.75))


@pytest.mark.skipif(not hasattr(torch, "broadcast_to"), reason="PyTorch <1.8 does not have broadcast_to")
@pytest.mark.parametrize(
("weights", "expected"), [(1, 11.5), (torch.ones(2, 1, 1), 11.5), (torch.tensor([1, 2]).reshape(2, 1, 1), 13.5)]
Expand Down