Supports slicing for multi-label data.

PiperOrigin-RevId: 523846333

Author: shs037

tensorflow_privacy/privacy/privacy_tests/membership_inference_attack/data_structures.py

@@ -84,9 +84,10 @@ class SlicingSpec:
   # When is set to true, one of the slices is the whole dataset.
   entire_dataset: bool = True
 
-  # Used in classification tasks for slicing by classes. It is assumed that
-  # classes are integers 0, 1, ... number of classes. When true one slice per
-  # each class is generated.
+  # Used in classification tasks for slicing by classes. When true one slice per
+  # each class is generated. Classes can either be
+  #  - integers 0, 1, ..., (for single label) or
+  #  - an array of integers (for multi-label).
   by_class: Union[bool, Iterable[int], int] = False
 
   # if true, it generates 10 slices for percentiles of the loss - 0-10%, 10-20%,
@@ -238,8 +239,10 @@ class AttackInputData:
   probs_train: Optional[np.ndarray] = None
   probs_test: Optional[np.ndarray] = None
 
-  # Contains ground-truth classes. Classes are assumed to be integers starting
-  # from 0.
+  # Contains ground-truth classes. For single-label classification, classes are
+  # assumed to be integers starting from 0. For multi-label classification,
+  # label is assumed to be multi-hot, i.e., labels is a binary array of shape
+  # (num_examples, num_classes).
   labels_train: Optional[np.ndarray] = None
   labels_test: Optional[np.ndarray] = None
 
@@ -290,7 +293,9 @@ def num_classes(self):
       raise ValueError(
           'Can\'t identify the number of classes as no labels were provided. '
           'Please set labels_train and labels_test')
-    return int(max(np.max(self.labels_train), np.max(self.labels_test))) + 1
+    if not self.multilabel_data:
+      return int(max(np.max(self.labels_train), np.max(self.labels_test))) + 1
+    return self.labels_train.shape[1]
 
   @property
   def logits_or_probs_train(self):
@@ -586,6 +591,8 @@ def validate(self):
       _is_array_two_dimensional(self.entropy_test, 'entropy_test')
       _is_array_two_dimensional(self.labels_train, 'labels_train')
       _is_array_two_dimensional(self.labels_test, 'labels_test')
+      self.is_multihot_labels(self.labels_train, 'labels_train')
+      self.is_multihot_labels(self.labels_test, 'labels_test')
     else:
       _is_array_one_dimensional(self.loss_train, 'loss_train')
       _is_array_one_dimensional(self.loss_test, 'loss_test')

tensorflow_privacy/privacy/privacy_tests/membership_inference_attack/dataset_slicing.py

@@ -72,10 +72,18 @@ def _slice_data_by_indices(data: AttackInputData, idx_train,
 
 
 def _slice_by_class(data: AttackInputData, class_value: int) -> AttackInputData:
-  if data.is_multilabel_data():
-    raise ValueError("Slicing by class not supported for multilabel data.")
-  idx_train = data.labels_train == class_value
-  idx_test = data.labels_test == class_value
+  """Gets the indices (boolean) for examples belonging to the given class."""
+  if not data.is_multilabel_data():
+    idx_train = data.labels_train == class_value
+    idx_test = data.labels_test == class_value
+  else:
+    if class_value >= data.num_classes:
+      raise ValueError(
+          f"class_value ({class_value}) is larger than the number of classes"
+          " (data.num_classes)."
+      )
+    idx_train = data.labels_train[:, class_value].astype(bool)
+    idx_test = data.labels_test[:, class_value].astype(bool)
   return _slice_data_by_indices(data, idx_train, idx_test)
 
 

tensorflow_privacy/privacy/privacy_tests/membership_inference_attack/dataset_slicing_test.py

@@ -358,10 +358,20 @@ def test_slice_entire_dataset(self):
     expected.slice_spec = entire_dataset_slice
     self.assertTrue(_are_all_fields_equal(output, self.input_data))
 
-  def test_slice_by_class_fails(self):
+  def test_slice_by_class(self):
     class_index = 1
     class_slice = SingleSliceSpec(SlicingFeature.CLASS, class_index)
-    self.assertRaises(ValueError, get_slice, self.input_data, class_slice)
+    output = get_slice(self.input_data, class_slice)
+    expected_indices_train = np.array([0, 2, 3])
+    expected_indices_test = np.array([1, 2])
+
+    np.testing.assert_array_equal(
+        output.logits_train,
+        self.input_data.logits_train[expected_indices_train],
+    )
+    np.testing.assert_array_equal(
+        output.logits_test, self.input_data.logits_test[expected_indices_test]
+    )
 
   @mock.patch('logging.Logger.info', wraps=logging.Logger)
   def test_slice_by_percentile_logs_multilabel_data(self, mock_logger):