Merge branch 'abigailgold-dev_1.11.0_label_fixes' into development_maintenance_111

Author: Beat Buesser

art/attacks/inference/attribute_inference/black_box.py

@@ -155,7 +155,7 @@ def fit(self, x: np.ndarray, y: Optional[np.ndarray] = None) -> None:
         if ClassifierMixin in type(self.estimator).__mro__:
             predictions = np.array([np.argmax(arr) for arr in self.estimator.predict(x)]).reshape(-1, 1)
             if y is not None:
-                y = check_and_transform_label_format(y, nb_classes=len(np.unique(y)), return_one_hot=True)
+                y = check_and_transform_label_format(y, nb_classes=self.estimator.nb_classes, return_one_hot=True)
         else:  # Regression model
             if self.scale_range is not None:
                 predictions = minmax_scale(self.estimator.predict(x).reshape(-1, 1), feature_range=self.scale_range)
@@ -237,7 +237,7 @@ def infer(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> np.n
         else:
             x_test = np.concatenate((x, pred), axis=1).astype(np.float32)
             if y is not None:
-                y = check_and_transform_label_format(y, nb_classes=len(np.unique(y)), return_one_hot=True)
+                y = check_and_transform_label_format(y, nb_classes=self.estimator.nb_classes, return_one_hot=True)
 
         if y is not None:
             x_test = np.concatenate((x_test, y), axis=1)

art/attacks/inference/attribute_inference/true_label_baseline.py

@@ -158,7 +158,7 @@ def fit(self, x: np.ndarray, y: np.ndarray) -> None:
                 normalized_labels = y * self.prediction_normal_factor
             normalized_labels = normalized_labels.reshape(-1, 1)
         else:
-            normalized_labels = check_and_transform_label_format(y, nb_classes=len(np.unique(y)), return_one_hot=True)
+            normalized_labels = check_and_transform_label_format(y, return_one_hot=True)
         x_train = np.concatenate((np.delete(x, self.attack_feature, 1), normalized_labels), axis=1).astype(np.float32)
 
         # train attack model
@@ -194,7 +194,7 @@ def infer(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> np.n
                 normalized_labels = y * self.prediction_normal_factor
             normalized_labels = normalized_labels.reshape(-1, 1)
         else:
-            normalized_labels = check_and_transform_label_format(y, nb_classes=len(np.unique(y)), return_one_hot=True)
+            normalized_labels = check_and_transform_label_format(y, return_one_hot=True)
         x_test = np.concatenate((x, normalized_labels), axis=1).astype(np.float32)
 
         predictions = self.attack_model.predict(x_test).astype(np.float32)

art/attacks/inference/membership_inference/black_box.py

@@ -187,8 +187,8 @@ def fit(  # pylint: disable=W0613
                 raise ValueError("Shape of test_x does not match input_shape of estimator")
 
         if not self._regressor_model:
-            y = check_and_transform_label_format(y, len(np.unique(y)), return_one_hot=True)
-            test_y = check_and_transform_label_format(test_y, len(np.unique(test_y)), return_one_hot=True)
+            y = check_and_transform_label_format(y, self.estimator.nb_classes, return_one_hot=True)  # type: ignore
+            test_y = check_and_transform_label_format(test_y, self.estimator.nb_classes, return_one_hot=True)
 
         if y.shape[0] != x.shape[0]:  # pragma: no cover
             raise ValueError("Number of rows in x and y do not match")
@@ -258,7 +258,7 @@ def fit(  # pylint: disable=W0613
                     loss.backward()
                     optimizer.step()
         else:
-            y_ready = check_and_transform_label_format(y_new, len(np.unique(y_new)), return_one_hot=False)
+            y_ready = check_and_transform_label_format(y_new, nb_classes=2, return_one_hot=False)
             self.attack_model.fit(np.c_[x_1, x_2], y_ready.ravel())  # type: ignore
 
     def infer(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> np.ndarray:
@@ -285,7 +285,7 @@ def infer(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> np.n
             probabilities = False
 
         if not self._regressor_model:
-            y = check_and_transform_label_format(y, len(np.unique(y)), return_one_hot=True)
+            y = check_and_transform_label_format(y, self.estimator.nb_classes, return_one_hot=True)
 
         if y is None:
             raise ValueError("None value detected.")

art/utils.py

@@ -686,26 +686,33 @@ def check_and_transform_label_format(
         if not return_one_hot:
             labels_return = np.argmax(labels, axis=1)
             labels_return = np.expand_dims(labels_return, axis=1)
-    elif (
-        len(labels.shape) == 2 and labels.shape[1] == 1 and nb_classes is not None and nb_classes > 2
-    ):  # multi-class, index labels
-        if return_one_hot:
-            labels_return = to_categorical(labels, nb_classes)
+    elif len(labels.shape) == 2 and labels.shape[1] == 1:
+        if nb_classes is None:
+            nb_classes = np.max(labels) + 1
+        if nb_classes > 2:  # multi-class, index labels
+            if return_one_hot:
+                labels_return = to_categorical(labels, nb_classes)
+            else:
+                labels_return = np.expand_dims(labels, axis=1)
+        elif nb_classes == 2:  # binary, index labels
+            if return_one_hot:
+                labels_return = to_categorical(labels, nb_classes)
         else:
-            labels_return = np.expand_dims(labels, axis=1)
-    elif (
-        len(labels.shape) == 2 and labels.shape[1] == 1 and nb_classes is not None and nb_classes == 2
-    ):  # binary, index labels
-        if return_one_hot:
-            labels_return = to_categorical(labels, nb_classes)
+            raise ValueError(
+                "Shape of labels not recognised."
+                "Please provide labels in shape (nb_samples,) or (nb_samples, "
+                "nb_classes)"
+            )
     elif len(labels.shape) == 1:  # index labels
         if return_one_hot:
             labels_return = to_categorical(labels, nb_classes)
         else:
             labels_return = np.expand_dims(labels, axis=1)
     else:
         raise ValueError(
-            "Shape of labels not recognised." "Please provide labels in shape (nb_samples,) or (nb_samples, nb_classes)"
+            "Shape of labels not recognised."
+            "Please provide labels in shape (nb_samples,) or (nb_samples, "
+            "nb_classes)"
         )
 
     return labels_return

tests/test_utils.py

@@ -210,6 +210,45 @@ def test_check_and_transform_label_format(self):
             labels_transformed, np.expand_dims(np.argmax(labels_expected_binary, axis=1), axis=1)
         )
 
+        # with no nb_classes
+
+        # test input shape (nb_samples,)
+        labels = np.array([3, 1, 4])
+        labels_transformed = check_and_transform_label_format(labels, return_one_hot=True)
+        np.testing.assert_array_equal(labels_transformed, labels_expected)
+
+        # test input shape (nb_samples, 1)
+        labels = np.array([[3], [1], [4]])
+        labels_transformed = check_and_transform_label_format(labels, return_one_hot=True)
+        np.testing.assert_array_equal(labels_transformed, labels_expected)
+
+        # test input shape (nb_samples, 1) - binary
+        labels = np.array([[1], [0], [1]])
+        labels_transformed = check_and_transform_label_format(labels, return_one_hot=True)
+        np.testing.assert_array_equal(labels_transformed, labels_expected_binary)
+
+        # test input shape (nb_samples, 1) - binary
+        labels = np.array([[0, 1], [1, 0], [0, 1]])
+        labels_transformed = check_and_transform_label_format(labels, return_one_hot=True)
+        np.testing.assert_array_equal(labels_transformed, labels_expected_binary)
+
+        # test input shape (nb_samples, nb_classes)
+        labels = np.array([[0, 0, 0, 1, 0], [0, 1, 0, 0, 0], [0, 0, 0, 0, 1]])
+        labels_transformed = check_and_transform_label_format(labels, return_one_hot=True)
+        np.testing.assert_array_equal(labels_transformed, labels_expected)
+
+        # test input shape (nb_samples, nb_classes) with return_one_hot=False
+        labels = np.array([[0, 0, 0, 1, 0], [0, 1, 0, 0, 0], [0, 0, 0, 0, 1]])
+        labels_transformed = check_and_transform_label_format(labels, return_one_hot=False)
+        np.testing.assert_array_equal(labels_transformed, np.expand_dims(np.argmax(labels_expected, axis=1), axis=1))
+
+        # test input shape (nb_samples, 1) - binary
+        labels = np.array([[1], [0], [1]])
+        labels_transformed = check_and_transform_label_format(labels, return_one_hot=False)
+        np.testing.assert_array_equal(
+            labels_transformed, np.expand_dims(np.argmax(labels_expected_binary, axis=1), axis=1)
+        )
+
         # ValueError for len(labels.shape) > 2
         labels = np.array([[[0, 0, 0, 1, 0], [0, 1, 0, 0, 0], [0, 0, 0, 0, 1]]])
         with self.assertRaises(ValueError):