ci, ruff: add timeout (closes #30), separate linting, factor strategy/setup + fix lint

Signed-off-by: Élie Goudout <[email protected]>

Author: eliegoudout

.github/workflows/ci.yml

@@ -7,23 +7,21 @@ on:
     branches: [develop, release]
 
 jobs:
-  build:
-    name: continuous-integration
-    runs-on: ${{ matrix.os }}
+  ci:
+    timeout-minutes: 15
     strategy:
       matrix:
+        os: ["ubuntu-latest", "windows-latest"]  # Remove MacOS (billing + github.com/ThalesGroup/scio/issues/2)
         python-version: ["3.12", "3.13"]
-        os: ["ubuntu-latest", "windows-latest", "macos-latest"]
-
-
+    runs-on: ${{ matrix.os }}
     steps:
       - name: Checkout
         uses: actions/checkout@v4
 
-      - name: Install the latest version of uv and set the python version
+      - name: Install uv and set python version
         uses: astral-sh/setup-uv@v6
         with:
-          version: "0.8.9"
+          version: "0.8.17"
           python-version: ${{ matrix.python-version }}
 
       - name: Run ruff
@@ -35,8 +33,13 @@ jobs:
         run: |
           uv -v run mypy
 
+      - name: Build & Install
+        run: |
+          uv -v build
+          uv -v pip install dist/scio_pypi-1.0.0rc3-py3-none-any.whl
+
       - name: Build docs (Posix)
-        if: matrix.os != 'windows-latest' && matrix.os != 'macos-latest'  # Temporary (github.com/ThalesGroup/scio/issues/2)
+        if: matrix.os != 'windows-latest'
         env:
           SPHINXOPTS: --fail-on-warning
         run: |
@@ -49,13 +52,7 @@ jobs:
         run: |
           uv -v run cmd /c "docs\\make.bat"
 
-      - name: Build & Install
-        run: |
-          uv -v build
-          uv -v pip install dist/scio_pypi-1.0.0rc3-py3-none-any.whl
-
       - name: Run pytest
-        if: matrix.os != 'macos-latest'  # Temporary (github.com/ThalesGroup/scio/issues/2)
         id: run-pytest
         continue-on-error: true
         run: |
@@ -67,13 +64,13 @@ jobs:
         with:
           name: pytest-observed-${{ matrix.os }}-py${{ matrix.python-version }}
           path: test/expected/*/*.observed.*
-          retention-days: 7
+          retention-days: 30
 
       - name: Fail the job if pytest failed
         if: ${{ steps.run-pytest.outcome == 'failure' }}
         run: exit 1
 
-      - name: Upload coverage report to Codecov (Ubuntu, python3.13)
+      - name: Upload coverage report to Codecov
         if: matrix.os == 'ubuntu-latest' && matrix.python-version == '3.13'
         uses: codecov/codecov-action@v5
         with:
@@ -82,7 +79,6 @@ jobs:
 
       # TEMPORARILY DISABLED (https://github.com/ThalesGroup/scio/issues/9)
       # - name: Run pytest with lowest resolution
-      #   if: matrix.os != 'macos-latest'  # Temporary (github.com/ThalesGroup/scio/issues/2)
       #   id: run-pytest-lowest
       #   continue-on-error: true
       #   run: |

scio/scores/classification/deepmahalanobis.py

@@ -300,7 +300,7 @@ def compute_precision(residues: Tensor) -> Tensor:
             if not precision.isfinite().all():
                 raise torch.linalg.LinAlgError  # pragma: no cover  # noqa: TRY301
         except torch.linalg.LinAlgError:
-            data_dim = residues.size(1)
+            data_dim = residues.shape[1]
             return torch.full((data_dim, data_dim), torch.nan).to(residues)
 
         return precision

scio/scores/classification/dknn.py

@@ -61,7 +61,7 @@ def _check_params(self, n_calib: int) -> None:
     def calibrate(self, calib_data: Tensor, calib_labels: Tensor) -> None:
         """Calibrate the scoring algorithm with In-Distribution data."""
         self.calib_labels = calib_labels
-        n_samples, n_classes = self.rnet(calib_data).shape  # Records activations
+        n_samples = len(self.rnet(calib_data))  # Records activations
         all_activations = self.activations()
         self.indexes = make_indexes(all_activations, metric=self.index_metric)
 

scio/scores/classification/gram.py

@@ -135,7 +135,7 @@ def calibrate(self, calib_data: Tensor, calib_labels_true: Tensor) -> None:
     def get_conformity(self, inputs: Tensor) -> tuple[Tensor, Tensor]:
         """Compute output and associated conformity at inference."""
         out = self.rnet(inputs)  # Records activations
-        n_samples, n_classes = out.shape
+        n_classes = out.shape[1]
 
         total_deviations = torch.zeros_like(out)
         for activations, low_high, expected_deviation in zip(
@@ -184,7 +184,7 @@ def layer_stats(self, activations: Tensor) -> Tensor:
             )
             raise ValueError(msg)
 
-        n_channels = activations.size(1) if ndim == 4 else 1  # noqa: PLR2004 (magic value 4)
+        n_channels = activations.shape[1] if ndim == 4 else 1  # noqa: PLR2004 (magic value 4)
         per_channel = activations.shape[ndim // 2 :].numel()
         activations_3d = activations.reshape(len(activations), n_channels, per_channel)
 

scio/scores/classification/isomax.py

@@ -60,7 +60,7 @@ def calibrate(self, calib_data: Tensor, calib_labels: Tensor) -> None:
         with torch.no_grad():
             logits = self.rnet(calib_data, dont_record=True)
 
-        n_classes = logits.size(1)
+        n_classes = logits.shape[1]
         # Init at 0 like [IsoMax]
         self.prototypes = torch.zeros(
             n_classes,

scio/scores/classification/jtla.py

@@ -167,7 +167,7 @@ def calibrate(self, calib_data: Tensor, calib_labels: Tensor) -> None:
         all_activations = self.activations()
         self.calib_labels_true = calib_labels
         self.calib_labels_pred = out.argmax(1) if self.pred_conditional else None
-        self.prepare_tests(all_activations, out.size(1))
+        self.prepare_tests(all_activations, out.shape[1])
         # Next are tensors of shape (n_classes, n_calib_samples, n_layers) (or None)
         self.calib_tests_true, self.calib_tests_pred = self.run_tests(all_activations)
 
@@ -410,7 +410,7 @@ def aggregate_layers(
 
         # Special aK-LPE method
         if self.layer_aggregation_method == "lpe":
-            n_calib = self.lpe_true_calib.size(1)
+            n_calib = self.lpe_true_calib.shape[1]
             similarity_search = self.index_metric == "ip"
 
             normed_true = self.lpe_normalizer(tests_true)
@@ -510,7 +510,7 @@ def slicer(n: int, mask: Tensor) -> Tensor:
             if only_consecutive:
                 return mask.unfold(1, n, 1)
 
-            range_n_layers = torch.arange(mask.size(1)).to(device=device)
+            range_n_layers = torch.arange(mask.shape[1]).to(device=device)
             combs = torch.combinations(range_n_layers, n)
             return mask[:, combs]
 

scio/scores/classification/odds.py

@@ -68,7 +68,7 @@ def calibrate(self, calib_data: Tensor, calib_labels: Tensor) -> None:
             self.rng = torch.Generator(calib_data.device).manual_seed(self.rng_seed)
 
         logits = self.rnet(calib_data, dont_record=True)
-        n_samples, n_classes = logits.shape
+        n_classes = logits.shape[1]
         self.g_stats_avg = torch.full(
             (n_classes,) * 2,
             torch.nan,

test/scio/eval/classification/roc/test_roc.py

@@ -18,13 +18,13 @@ def test_roc_trivial():
 @parametrize_bool("label")
 def test_roc_requires_positive_and_negative(label):
     """Test that both positive and negative samples are required."""
-    with pytest.raises(AssertionError, match="^$"):
+    with pytest.raises(AssertionError, match=r"^$"):
         ROC([label], [0])
 
 
 def test_roc_rejects_nan():
     """Test that ``nan`` are forbidden."""
-    with pytest.raises(AssertionError, match="^$"):
+    with pytest.raises(AssertionError, match=r"^$"):
         ROC([T, F], [0, np.nan])
 
 
@@ -34,7 +34,7 @@ def test_roc_allows_only_posinf(sign):
     if sign > 0:
         ROC([T, F], [0, I])
     else:
-        with pytest.raises(AssertionError, match="^$"):
+        with pytest.raises(AssertionError, match=r"^$"):
             ROC([T, F], [0, -I])
 
 

test/scio/scores/classification/base/test_base.py

@@ -24,7 +24,7 @@ def test_raises_on_unexpected_mode(monkeypatch_enum_new_member, score, test_data
 
 def test_accepts_unique_conformity(score, class_aggregation, test_data):
     """Check that the score can output one score only per sample."""
-    out, conformity = score(test_data)
+    conformity = score(test_data)[1]
     same_conformity_over_classes = (conformity == conformity[:, [0]]).all()
 
     assert conformity.shape == (len(test_data), N_CLASSES)