Improve coverage of data module

.github/codecov.yml

@@ -3,7 +3,4 @@ coverage:
     project: false
 
 ignore:
-  - "setup.py"
   - "torchio/external/due.py"
-  - "torchio/datasets"
-  - "torchio/download.py"

.github/workflows/tests.yml

@@ -46,21 +46,57 @@ jobs:
         shell: bash
 
       - name: Setup test suite
-        run: tox run -v --notest --skip-missing-interpreters false -e ${{ matrix.python }}
+        run: |
+          if [ "${{ matrix.os }}" == "ubuntu-latest" ] && [ "${{ matrix.python }}" == "3.14" ]; then
+            tox_env=coverage
+          else
+            tox_env=pytest
+          fi
+          tox run -v --notest --skip-missing-interpreters false -e "$tox_env"
+        shell: bash # this wouldn't work on Powershell
 
-      # Run all tests on schedule, but only non-slow tests on push
-      - name: Run pytest
+      # Run all tests on schedule, but only non-slow tests otherwise
+      - name: Run unit tests
         run: |
+          if [ "${{ matrix.os }}" == "ubuntu-latest" ] && [ "${{ matrix.python }}" == "3.14" ]; then
+            tox_env=coverage
+          else
+            tox_env=pytest
+          fi
+
           if [ "${{ github.event_name }}" == "schedule" ]; then
-            tox -e pytest
+            tox -e "$tox_env"
           else
-            tox -e pytest -- -m "not slow"
+            tox -e "$tox_env" -- -m "not slow"
           fi
         shell: bash # this wouldn't work on Powershell
 
+      - name: Summarize coverage
+        if: ${{ matrix.os == 'ubuntu-latest' && matrix.python == '3.14' }}
+        run: |
+          python3 - <<'PY'
+          import os
+          import xml.etree.ElementTree as ET
+
+          root = ET.parse('coverage.xml').getroot()
+          line_rate = float(root.attrib['line-rate']) * 100
+          lines_covered = int(root.attrib['lines-covered'])
+          lines_valid = int(root.attrib['lines-valid'])
+
+          with open(os.environ['GITHUB_STEP_SUMMARY'], 'a', encoding='utf-8') as f:
+              f.write('## Coverage report\n\n')
+              f.write(
+                  f'- Line coverage: **{line_rate:.2f}%** '
+                  f'({lines_covered}/{lines_valid} lines)\n'
+              )
+              f.write('- Full coverage details are uploaded to Codecov.\n')
+          PY
+
       - name: Upload coverage reports to Codecov
         if: ${{ matrix.os == 'ubuntu-latest' && matrix.python == '3.14' }}
         uses: codecov/codecov-action@v5.5.2
+        with:
+          files: ./coverage.xml
         env:
           CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
 

tests/data/inference/test_aggregator.py

@@ -69,6 +69,10 @@ def test_overlap_hann(self):
         )
         self.aggregate('hann', fixture)
 
+    def test_bad_overlap_mode(self):
+        with pytest.raises(ValueError, match='Overlap mode must be'):
+            tio.data.GridAggregator._parse_overlap_mode('bad')
+
     def run_sampler_aggregator(self, overlap_mode='crop'):
         patch_size = 10
         patch_overlap = 2
@@ -93,6 +97,11 @@ def test_warning_int64(self):
         with pytest.warns(RuntimeWarning):
             aggregator.get_output_tensor()
 
+    def test_overlap_hann_casts_int_tensors_to_float(self):
+        aggregator = self.run_sampler_aggregator(overlap_mode='hann')
+        output = aggregator.get_output_tensor()
+        assert output.dtype == torch.float32
+
     def run_patch_crop_issue(self, *, padding_mode):
         # https://github.com/TorchIO-project/torchio/issues/813
         pao, pas, ims, bb1, bb2 = 4, 102, 320, 100, 120

tests/data/inference/test_grid_sampler.py

@@ -65,3 +65,10 @@ def test_subject_shape(self):
         )
         final_shape = self.sample_subject.shape
         assert initial_shape == final_shape
+
+    def test_num_patches_not_supported(self):
+        sampler = tio.GridSampler(self.sample_subject, 5)
+        with pytest.raises(
+            ValueError, match='does not support limiting the number of patches'
+        ):
+            next(sampler(num_patches=1))

tests/data/sampler/test_label_sampler.py

@@ -1,3 +1,4 @@
+import numpy as np
 import pytest
 import torch
 
@@ -79,3 +80,18 @@ def test_empty_map(self):
         sampler = tio.LabelSampler(4)
         with pytest.raises(RuntimeError):
             next(sampler(subject))
+
+    def test_missing_label_name(self):
+        subject = tio.Subject(image=tio.ScalarImage(tensor=torch.rand(1, 1, 1, 1)))
+        sampler = tio.LabelSampler(1, 'missing')
+        with pytest.raises(KeyError):
+            sampler.get_probability_map_image(subject)
+
+    def test_patch_size_larger_than_label_map(self):
+        label_map = torch.ones(1, 5, 5, 5)
+        with pytest.raises(RuntimeError, match='Patch size .*larger than label map'):
+            tio.LabelSampler.get_probabilities_from_label_map(
+                label_map,
+                {1: 1},
+                np.array((8, 8, 8)),
+            )

tests/data/sampler/test_patch_sampler.py

@@ -21,3 +21,8 @@ def test_bad_patch_size(self):
 
     def test_extract_patch(self):
         PatchSampler(1).extract_patch(self.sample_subject, (3, 4, 5))
+
+    def test_patch_size_larger_than_image(self):
+        sampler = PatchSampler(100)
+        with pytest.raises(RuntimeError, match='cannot be larger than image size'):
+            sampler(self.sample_subject)

tests/data/sampler/test_uniform_sampler.py

@@ -24,3 +24,8 @@ def test_incosistent_shape(self):
         patch_size = 2
         sampler = UniformSampler(patch_size)
         next(sampler(subject))
+
+    def test_num_patches(self):
+        sampler = UniformSampler(5)
+        patches = list(sampler(self.sample_subject, num_patches=3))
+        assert len(patches) == 3

tests/data/sampler/test_weighted_sampler.py

@@ -1,3 +1,7 @@
+from unittest.mock import patch
+
+import numpy as np
+import pytest
 import torch
 
 import torchio as tio
@@ -37,3 +41,44 @@ def test_inconsistent_shape(self):
         patch_size = 2
         sampler = tio.data.WeightedSampler(patch_size, 'im1')
         next(sampler(subject))
+
+    def test_missing_probability_map(self):
+        subject = self.get_sample((1, 7, 7, 7))
+        sampler = WeightedSampler(5, 'missing')
+        with pytest.raises(KeyError):
+            sampler.get_probability_map_image(subject)
+
+    def test_negative_probability_map(self):
+        subject = self.get_sample((1, 7, 7, 7))
+        subject.prob.set_data(subject.prob.data - 2)
+        sampler = WeightedSampler(5, 'prob')
+        with pytest.raises(ValueError, match='Negative values found'):
+            sampler.get_probability_map(subject)
+
+    def test_extract_patch_without_cdf(self):
+        subject = self.get_sample((1, 7, 7, 7))
+        sampler = WeightedSampler(5, 'prob')
+        patch = sampler.extract_patch(subject, np.array((1, 1, 1)))
+        assert patch.spatial_shape == (5, 5, 5)
+
+    def test_extract_patch_tuple_without_cdf(self):
+        subject = self.get_sample((1, 7, 7, 7))
+        sampler = WeightedSampler(5, 'prob')
+        patch = sampler.extract_patch(subject, (1, 1, 1))
+        assert patch.spatial_shape == (5, 5, 5)
+
+    def test_extract_patch_invalid_type_with_cdf(self):
+        subject = self.get_sample((1, 7, 7, 7))
+        sampler = WeightedSampler(5, 'prob')
+        with pytest.raises(TypeError, match='Probability map must be a NumPy array'):
+            sampler.extract_patch(subject, (1, 1, 1), cdf=np.array((1.0,)))
+
+    def test_probability_zero_raises(self):
+        probability_map = np.array(((0, 1), (1, 1)), dtype=np.float32)
+        cdf = WeightedSampler.get_cumulative_distribution_function(probability_map)
+
+        with (
+            patch('torchio.data.sampler.weighted.np.searchsorted', return_value=0),
+            pytest.raises(RuntimeError, match='Error retrieving probability'),
+        ):
+            WeightedSampler.sample_probability_map(probability_map, cdf)