Skip to content

fix(training): raise when a class has no samples instead of training on NaN centroids #21

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

Open
AnKiT-GaRG2 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix(training): raise when a class has no samples instead of training on NaN centroids #21

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
7 changes: 6 additions & 1 deletion src/dlens/tools/_training.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,8 +66,13 @@ def train(
self, dataset: DatasetRef, architecture: ArchitectureSpec, config: TrainingConfig
) -> TrainResult:
x, y = load_dataset(dataset)
xn, mu, sd = _standardize(x)
k = dataset.num_classes
counts = np.bincount(y, minlength=k)
missing = [name for name, count in zip(dataset.class_names, counts) if count == 0]
if missing:
# An empty class would get a NaN centroid, which silently wins every argmin.
raise RuntimeError(f"No samples found under {dataset.root!r} for class(es) {missing}.")
xn, mu, sd = _standardize(x)
centroids = np.stack([xn[y == c].mean(axis=0) for c in range(k)])
# nearest-centroid train accuracy
dists = np.linalg.norm(xn[:, None, :] - centroids[None, :, :], axis=2)
Expand Down
8 changes: 8 additions & 0 deletions tests/test_downstream_agents.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,6 +58,14 @@ def _ref(**kw) -> DatasetRef:
return DatasetRef(**base)


def test_centroid_train_rejects_class_without_samples(tmp_path: Path):
ds = _make_dataset(tmp_path)
ds = ds.model_copy(update={"class_names": [*ds.class_names, "c"]})
with pytest.raises(RuntimeError, match=r"class\(es\) \['c'\]"):
CentroidTrainBackend(output_root=str(tmp_path / "models")).train(ds, _arch(), _cfg())
assert not (tmp_path / "models").exists()


def test_real_backends_chain(tmp_path: Path):
ds = _make_dataset(tmp_path)
tr = CentroidTrainBackend(output_root=str(tmp_path / "models")).train(ds, _arch(), _cfg())
Expand Down