tristandeleu
diff --git a/Collapse file
‎torchmeta/datasets/bach.py‎
Copy file name to clipboardExpand all lines: torchmeta/datasets/bach.py
+4Lines changed: 4 additions & 0 deletions b/Collapse file
‎torchmeta/datasets/bach.py‎
Copy file name to clipboardExpand all lines: torchmeta/datasets/bach.py
+4Lines changed: 4 additions & 0 deletions
diff --git a/Collapse file
‎torchmeta/tests/datasets/test_datasets_helpers_tabular.py‎
Copy file name to clipboardExpand all lines: torchmeta/tests/datasets/test_datasets_helpers_tabular.py
+9-2Lines changed: 9 additions & 2 deletions b/Collapse file
‎torchmeta/tests/datasets/test_datasets_helpers_tabular.py‎
Copy file name to clipboardExpand all lines: torchmeta/tests/datasets/test_datasets_helpers_tabular.py
+9-2Lines changed: 9 additions & 2 deletions
@@ -172,6 +172,10 @@ def __init__(self, root, meta_train=False, meta_val=False, meta_test=False, meta
         if download:
             self.download(process_features, min_num_samples_per_class)
 
+        if min_num_samples_per_class != self.meta_data["min_num_data_per_class"]:
+            raise ValueError("min_num_samples_per_class given ({0}) does not match existing value"
+                             "({1}).".format(min_num_samples_per_class, self.meta_data["min_num_data_per_class"]))
+
         if not self._check_integrity():
             raise RuntimeError('Bach integrity check failed')
         self._num_classes = len(self.labels)
 
@@ -8,10 +8,17 @@
 
 is_local = (os.getenv('TORCHMETA_DATA_FOLDER') is not None)
 
+list_of_datasets = helpers_tabular.__all__
+# `bach' is a bit more tricky to test as it has classes with very little data which are
+# dropped during pre-processing according to the parameter `min_num_samples_per_class'.
+# Hence the test is only applicable if the downloaded dataset was processed with
+# min_num_samples_per_class = shots + test_shots.
+list_of_datasets.remove('bach')
+
 
 @pytest.mark.skipif(not is_local, reason='Requires datasets downloaded locally')
-@pytest.mark.parametrize('name', helpers_tabular.__all__)
-@pytest.mark.parametrize('shots', [5, 1])  # large number first for `bach' dataset.
+@pytest.mark.parametrize('name', list_of_datasets)
+@pytest.mark.parametrize('shots', [1, 5])
 @pytest.mark.parametrize('split', ['train', 'val', 'test'])
 def test_datasets_helpers_tabular(name, shots, split):
     function = getattr(helpers_tabular, name)