Fix LGTM warnings

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

Author: Beat Buesser

art/attacks/poisoning/gradient_matching_attack.py

@@ -248,7 +248,7 @@ def __initialize_poison_pytorch(
         x_poison: np.ndarray,
         y_poison: np.ndarray,  # pylint: disable=unused-argument
     ):
-        import torch
+        import torch  # lgtm [py/import-and-import-from]
         from torch import nn
         from art.estimators.classification.pytorch import PyTorchClassifier
 
@@ -454,7 +454,7 @@ def __poison__pytorch(self, x_poison: np.ndarray, y_poison: np.ndarray) -> Tuple
         :return: A pair of poisoned samples, B-score (cosine similarity of the gradients).
         """
 
-        import torch
+        import torch  # lgtm [py/import-and-import-from]
 
         device = "cuda" if torch.cuda.is_available() else "cpu"
 

art/estimators/certification/deep_z/deep_z.py

@@ -25,8 +25,6 @@
 import numpy as np
 import torch
 
-from torch import nn
-
 
 class ZonoDenseLayer(torch.nn.Module):
     """
@@ -211,7 +209,7 @@ def __init__(
         padding: Union[int, Tuple[int, int]] = 0,
     ):
         super().__init__()
-        self.conv = nn.Conv2d(
+        self.conv = torch.nn.Conv2d(
             in_channels=in_channels,
             out_channels=out_channels,
             kernel_size=kernel_size,

art/estimators/certification/deep_z/pytorch.py

@@ -26,7 +26,6 @@
 import warnings
 import numpy as np
 import torch
-from torch import nn
 
 from art.estimators.certification.deep_z.deep_z import ZonoConv, ZonoDenseLayer, ZonoReLU, ZonoBounds
 from art.estimators.classification.pytorch import PyTorchClassifier
@@ -125,7 +124,7 @@ def forward_hook(input_module, hook_input, hook_output):
         input_for_hook = torch.unsqueeze(input_for_hook, dim=0)
         model(input_for_hook)  # hooks are fired sequentially from model input to the output
 
-        self.ops = nn.ModuleList()
+        self.ops = torch.nn.ModuleList()
         for module in modules:
             print("registered", type(module))
             if isinstance(module, torch.nn.modules.conv.Conv2d):

art/preprocessing/expectation_over_transformation/image_rotation/pytorch.py

@@ -211,7 +211,7 @@ def _check_params(self) -> None:
                     """For `label_type="object_detection"` only a list of multiples of 90 degrees is supported."""
                 )
             if isinstance(self.angles, list):
-                for angle in self.angles:
+                for angle in self.angles:  # lgtm [py/non-iterable-in-for-loop]
                     if divmod(angle, 90)[1] != 0:
                         raise ValueError(
                             """For `label_type="object_detection"` only a list of multiples of 90 degrees is

tests/estimators/certification/test_deepz.py

@@ -20,10 +20,8 @@
 import os
 import pytest
 
-import torch
 import numpy as np
-
-from torch import nn
+import torch
 
 from art.utils import load_dataset
 from art.estimators.certification.deep_z import PytorchDeepZ
@@ -76,7 +74,7 @@ def test_mnist_certification(art_warning, fix_get_mnist_data):
     ptc = get_image_classifier_pt(from_logits=True, use_maxpool=False)
 
     zonotope_model = PytorchDeepZ(
-        model=ptc.model, clip_values=(0, 1), loss=nn.CrossEntropyLoss(), input_shape=(1, 28, 28), nb_classes=10
+        model=ptc.model, clip_values=(0, 1), loss=torch.nn.CrossEntropyLoss(), input_shape=(1, 28, 28), nb_classes=10
     )
 
     correct_upper_bounds = np.asarray(
@@ -173,7 +171,7 @@ def test_cifar_certification(art_warning, fix_get_cifar10_data):
 
     ptc = get_cifar10_image_classifier_pt(from_logits=True)
     zonotope_model = PytorchDeepZ(
-        model=ptc.model, clip_values=(0, 1), loss=nn.CrossEntropyLoss(), input_shape=(3, 32, 32), nb_classes=10
+        model=ptc.model, clip_values=(0, 1), loss=torch.nn.CrossEntropyLoss(), input_shape=(3, 32, 32), nb_classes=10
     )
 
     correct_upper_bounds = np.asarray(

tests/utils.py

@@ -336,7 +336,7 @@ def get_image_classifier_tf_v1(from_logits=False, load_init=True, sess=None):
 
 
 def get_image_generator_tf_v2(capacity: int, z_dim: int):
-    import tensorflow as tf
+    import tensorflow as tf  # lgtm [py/import-and-import-from]
 
     def make_image_generator_model(capacity: int, z_dim: int) -> tf.keras.Sequential():
         model = tf.keras.Sequential()
@@ -372,7 +372,7 @@ def make_image_generator_model(capacity: int, z_dim: int) -> tf.keras.Sequential
 
 
 def get_image_gan_tf_v2():
-    import tensorflow as tf
+    import tensorflow as tf  # lgtm [py/import-and-import-from]
 
     noise_dim = 100
     capacity = 64