Add build os to docs yml

Signed-off-by: Beat Buesser <[email protected]>

art/attacks/evasion/auto_attack.py

@@ -286,9 +286,10 @@ def generate(self, x: np.ndarray, y: Optional[np.ndarray] = None, **kwargs) -> n
                     logger.warning("Error completing attack: %s}", str(error))
 
         if self.parallel:
-            with multiprocess.get_context("spawn").Pool() as pool:
+            with multiprocess.get_context("spawn").Pool(processes=1) as pool:
                 # Results come back in the order that they were issued
                 results = pool.starmap(run_attack, args)
+                asdfdasd
             perturbations = []
             is_robust = []
             for img_idx in range(len(x)):
@@ -364,6 +365,7 @@ def run_attack(
     :param eps: Maximum perturbation that the attacker can introduce.
     :return: An array holding the adversarial examples.
     """
+    eeeeee
     # Attack only correctly classified samples
     x_robust = x[sample_is_robust]
     y_robust = y[sample_is_robust]
@@ -394,4 +396,6 @@ def run_attack(
 
     sample_is_robust[sample_is_robust] = np.invert(sample_is_not_robust)
 
+    print("Attack done:", attack)
+
     return x, sample_is_robust

art/defences/detector/poison/clustering_analyzer.py

@@ -328,7 +328,7 @@ def analyze_by_silhouette_score(
                     logger.info("computed silhouette score: %s", silhouette_avg)
                     dict_i.update(suspicious=True)
                 else:
-                    poison_clusters = (np.array([[]]), )
+                    poison_clusters = (np.array([[]]),)
                     clean_clusters = np.where(percentages >= 0)
                     dict_i.update(suspicious=False)
             else:

tests/attacks/evasion/test_auto_attack.py

@@ -193,7 +193,9 @@ def test_classifier_type_check_fail(art_warning):
         art_warning(e)
 
 
-@pytest.mark.skip_framework("tensorflow1", "tensorflow2v1", "keras", "non_dl_frameworks", "mxnet", "kerastf")
+@pytest.mark.skip_framework(
+    "tensorflow1", "tensorflow2v1", "tensorflow2", "keras", "non_dl_frameworks", "mxnet", "kerastf"
+)
 def test_generate_parallel(art_warning, fix_get_mnist_subset, image_dl_estimator, framework):
     try:
         classifier, _ = image_dl_estimator(from_logits=True)
@@ -286,35 +288,12 @@ def test_generate_parallel(art_warning, fix_get_mnist_subset, image_dl_estimator
             parallel=False,
         )
 
-        from tensorflow.keras.utils import CustomObjectScope
-
-        # Copy
-        from tests.utils import _tf_weights_loader
-
-        _tf_initializer_W_CONV2D_MNIST = _tf_weights_loader("MNIST", "W", "CONV2D", 2)
-        # _tf_initializer_MNIST_W_CONV2D.__name__ = "_tf_initializer_MNIST_W_CONV2D"
-        _tf_initializer_B_CONV2D_MNIST = _tf_weights_loader("MNIST", "B", "CONV2D", 2)
-        # _tf_initializer_MNIST_B_CONV2D.__name__ = "_tf_initializer_MNIST_B_CONV2D"
-
-        _tf_initializer_W_DENSE_MNIST = _tf_weights_loader("MNIST", "W", "DENSE", 2)
-        # _tf_initializer_MNIST_W_DENSE.__name__ = "_tf_initializer_MNIST_W_DENSE"
-        _tf_initializer_B_DENSE_MNIST = _tf_weights_loader("MNIST", "B", "DENSE", 2)
-        # _tf_initializer_MNIST_B_DENSE.__name__ = "_tf_initializer_MNIST_B_DENSE"
-
-        custom_objects = {
-            "_tf_initializer_W_CONV2D_MNIST": _tf_initializer_W_CONV2D_MNIST,
-            "_tf_initializer_B_CONV2D_MNIST": _tf_initializer_B_CONV2D_MNIST,
-            "_tf_initializer_W_DENSE_MNIST": _tf_initializer_W_DENSE_MNIST,
-            "_tf_initializer_B_DENSE_MNIST": _tf_initializer_B_DENSE_MNIST,
-        }
-
-        with CustomObjectScope(custom_objects):
-            x_train_mnist_adv = attack.generate(x=x_train_mnist, y=y_train_mnist)
+        x_train_mnist_adv = attack.generate(x=x_train_mnist, y=y_train_mnist)
 
         x_train_mnist_adv_nop = attack_noparallel.generate(x=x_train_mnist, y=y_train_mnist)
 
-        assert np.mean(np.abs(x_train_mnist_adv - x_train_mnist)) == pytest.approx(0.0182, abs=0.105)
-        assert np.max(np.abs(x_train_mnist_adv - x_train_mnist)) == pytest.approx(0.3, abs=0.05)
+        assert np.mean(np.abs(x_train_mnist_adv - x_train_mnist)) == pytest.approx(expected=0.0182, abs=0.105)
+        assert np.max(np.abs(x_train_mnist_adv - x_train_mnist)) == pytest.approx(expected=0.3, abs=0.05)
 
         noparallel_perturbation = np.linalg.norm(x_train_mnist[[2]] - x_train_mnist_adv_nop[[2]])
         parallel_perturbation = np.linalg.norm(x_train_mnist[[2]] - x_train_mnist_adv[[2]])
@@ -334,10 +313,10 @@ def test_generate_parallel(art_warning, fix_get_mnist_subset, image_dl_estimator
             parallel=True,
         )
 
-        with CustomObjectScope(custom_objects):
-            x_train_mnist_adv = attack.generate(x=x_train_mnist, y=y_train_mnist)
+        x_train_mnist_adv = attack.generate(x=x_train_mnist, y=y_train_mnist)
+
+        assert np.mean(x_train_mnist_adv - x_train_mnist) == pytest.approx(expected=0.0, abs=0.0075)
+        assert np.max(np.abs(x_train_mnist_adv - x_train_mnist)) == pytest.approx(expected=eps, abs=0.005)
 
-        assert np.mean(x_train_mnist_adv - x_train_mnist) == pytest.approx(0.0, abs=0.0075)
-        assert np.max(np.abs(x_train_mnist_adv - x_train_mnist)) == pytest.approx(eps, abs=0.005)
     except ARTTestException as e:
         art_warning(e)

tests/defences/detector/poison/test_provenance_defence.py

@@ -99,6 +99,7 @@ def setUpClass(cls):
             eps=1.0,
             x_val=valid_data,
             y_val=valid_labels,
+            max_iter=100,
             verbose=False,
         )
 

tests/utils.py

@@ -29,6 +29,7 @@
 import warnings
 
 import numpy as np
+import tensorflow.python.keras.initializers.initializers_v2
 
 from art.estimators.classification.tensorflow import TensorFlowV2Classifier
 from art.estimators.encoding.tensorflow import TensorFlowEncoder
@@ -165,38 +166,61 @@ def is_valid_framework(framework):
     return True
 
 
-def _tf_weights_loader(dataset, weights_type, layer="DENSE", tf_version=1):
+class _tf_weights_loader(tensorflow.keras.initializers.Initializer):
 
-    filename = str(weights_type) + "_" + str(layer) + "_" + str(dataset) + ".npy"
+    import tensorflow as tf
 
-    # pylint: disable=W0613
-    # disable pylint because of API requirements for function
-    if tf_version == 1:
+    def __init__(self, dataset, weights_type, layer="DENSE", tf_version=1):
+        self.dataset = dataset
+        self.weights_type = weights_type
+        self.layer = layer
+        self.tf_version = tf_version
 
-        def _tf_initializer(_, dtype, partition_info):
-            import tensorflow as tf
+    def get_config(self):
+        return {
+            "dataset": self.dataset,
+            "weights_type": self.weights_type,
+            "layer": self.layer,
+            "tf_version": self.tf_version,
+        }
 
-            weights = np.load(
-                os.path.join(os.path.dirname(os.path.dirname(__file__)), "utils/resources/models", filename)
-            )
-            return tf.constant(weights, dtype)
+    def __call__(self, shape, dtype=None, **kwargs):
 
-    elif tf_version == 2:
+        import tensorflow as tf
 
-        def _tf_initializer(_, dtype):
-            import tensorflow as tf
+        filename = str(self.weights_type) + "_" + str(self.layer) + "_" + str(self.dataset) + ".npy"
 
-            weights = np.load(
-                os.path.join(os.path.dirname(os.path.dirname(__file__)), "utils/resources/models", filename)
-            )
-            return tf.constant(weights, dtype)
-
-    else:
-        raise ValueError("The TensorFlow version tf_version has to be either 1 or 2.")
+        # pylint: disable=W0613
+        # disable pylint because of API requirements for function
+        # if self.tf_version == 1:
+        #
+        #     def _tf_initializer(_, dtype, partition_info):
+        #         import tensorflow as tf
+        #
+        #         weights = np.load(
+        #             os.path.join(os.path.dirname(os.path.dirname(__file__)), "utils/resources/models", filename)
+        #         )
+        #         return tf.constant(weights, dtype)
+
+        # elif self.tf_version == 2:
+        #
+        #     def _tf_initializer(_, dtype):
+        #         import tensorflow as tf
+        #
+        #         weights = np.load(
+        #             os.path.join(os.path.dirname(os.path.dirname(__file__)), "utils/resources/models", filename)
+        #         )
+        #         return tf.constant(weights, dtype)
+        #
+        # else:
+        #     raise ValueError("The TensorFlow version tf_version has to be either 1 or 2.")
+        #
+        # _tf_initializer.__name__ = "_tf_initializer_" + str(self.weights_type) + "_" + str(self.layer) + "_" + str(self.dataset)
 
-    _tf_initializer.__name__ = "_tf_initializer_" + str(weights_type) + "_" + str(layer) + "_" + str(dataset)
+        weights = np.load(os.path.join(os.path.dirname(os.path.dirname(__file__)), "utils/resources/models", filename))
+        return tf.convert_to_tensor(weights, dtype)
 
-    return _tf_initializer
+        # return _tf_initializer
 
 
 def _kr_weights_loader(dataset, weights_type, layer="DENSE"):
@@ -468,6 +492,10 @@ def get_image_classifier_tf_v2(from_logits=False):
     _tf_initializer_W_DENSE_MNIST = _tf_weights_loader("MNIST", "W", "DENSE", 2)
     _tf_initializer_B_DENSE_MNIST = _tf_weights_loader("MNIST", "B", "DENSE", 2)
 
+    print("W", _tf_initializer_W_CONV2D_MNIST)
+    print("B", _tf_initializer_B_CONV2D_MNIST)
+    # sdf
+
     model = Sequential()
     model.add(
         Conv2D(