Prevent TextDataset objects from containing None (#73)

Signed-off-by: Christopher Schröder <chschroeder@users.noreply.github.com>

CHANGELOG.md

@@ -18,6 +18,8 @@ On the other hand, this also allowed us to deal with further issues that contain
   - Added environment variables `SMALL_TEXT_PROGRESS_BARS` and `SMALL_TEXT_OFFLINE` to control the default behavior for progress bars and model downloading.
 - PoolBasedActiveLearner:
   - `initialize_data()` has been replaced by `initialize()` which can now also be used to provide an initial model in cold start scenarios. ([#10](https://github.com/webis-de/small-text/pull/10))
+- Datasets
+  - Validation has been added to prevent TextDataset objects from containing None items (instead of str) either during initialization or when setting the x property. ([#73](https://github.com/webis-de/small-text/pull/73))
 - Classification:
   - All PyTorch-classifiers (KimCNN, TransformerBasedClassification, SetFitClassification) now support `torch.compile()` which can be enabled on demand. (Requires PyTorch >= 2.0.0). 
   - All PyTorch-classifiers (KimCNN, TransformerBasedClassification, SetFitClassification) now support Automatic Mixed Precision. 

small_text/data/datasets.py

@@ -40,6 +40,12 @@ def check_dataset_and_labels(x, y):
                              f'x = ({len_x},), y.shape= ({y.shape[0]},) ### {type(x)} / {type(y)}')
 
 
+def check_text_data(x):
+    for i, item in enumerate(x):
+        if item is None:
+            raise ValueError(f'instance #{i} is None which is not allowed.')
+
+
 def check_target_labels(target_labels):
     if target_labels is None:
         warnings.warn('Passing target_labels=None is discouraged as it can lead to '
@@ -573,6 +579,7 @@ def __init__(self, x, y, target_labels=None):
         target_labels : numpy.ndarray[int] or None, default=None
             List of possible labels. Will be inferred from `y` if `None` is passed."""
         check_dataset_and_labels(x, y)
+        check_text_data(x)
         check_target_labels(target_labels)
 
         if isinstance(x, np.ndarray):
@@ -603,6 +610,7 @@ def x(self):
 
     @x.setter
     def x(self, x_new):
+        check_text_data(x_new)
         check_dataset_and_labels(x_new, self._y)
         self._x = x_new
 

tests/unit/small_text/data/test_datasets.py

@@ -424,6 +424,19 @@ def test_init_with_dimension_mismatch(self):
         with self.assertRaisesRegex(ValueError, 'Feature and label dimensions do not match'):
             TextDataset(x, y, target_labels=ds.target_labels)
 
+    def test_init_when_samples_are_none(self):
+        x = [None] * self.NUM_SAMPLES
+        y = random_labels(self.NUM_SAMPLES, self.NUM_LABELS,
+                          multi_label=self.labels_type == 'sparse')
+
+        if self.target_labels not in ['explicit', 'inferred']:
+            raise ValueError('Invalid test parameter value for target_labels:' + self.target_labels)
+
+        target_labels = None if self.target_labels == 'inferred' else np.arange(5)
+
+        with self.assertRaisesRegex(ValueError, 'instance #0 is None which is not allowed.'):
+            TextDataset(x, y, target_labels=target_labels)
+
     def test_init_when_some_samples_are_unlabeled(self):
         x = random_text_data(self.NUM_SAMPLES)
         y = random_labels(self.NUM_SAMPLES, self.NUM_LABELS,
@@ -480,6 +493,11 @@ def test_set_features(self):
 
         assert_array_equal(ds.x, ds_new.x)
 
+    def test_set_features_when_samples_are_none(self):
+        ds, x, _ = self._dataset(num_samples=self.NUM_SAMPLES, return_data=True)
+        with self.assertRaisesRegex(ValueError, 'instance #0 is None which is not allowed.'):
+            ds.x = [None] * len(x)
+
     def test_set_features_with_dimension_mismatch(self):
         ds, x, _ = self._dataset(num_samples=self.NUM_SAMPLES, return_data=True)
         x = x[1:]