Fix LGTM warnings

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

Author: Beat Buesser

art/estimators/pytorch.py

@@ -52,7 +52,7 @@ def __init__(self, device_type: str = "gpu", **kwargs) -> None:
                be divided by the second one.
         :param device_type: Type of device on which the classifier is run, either `gpu` or `cpu`.
         """
-        import torch
+        import torch  # lgtm [py/repeated-import]
 
         preprocessing = kwargs.get("preprocessing")
         if isinstance(preprocessing, tuple):
@@ -150,7 +150,7 @@ def _apply_preprocessing(self, x, y, fit: bool = False, no_grad=True) -> Tuple[A
         :return: Tuple of `x` and `y` after applying the defences and standardisation.
         :rtype: Format as expected by the `model`
         """
-        import torch
+        import torch  # lgtm [py/repeated-import]
 
         from art.preprocessing.standardisation_mean_std.standardisation_mean_std import StandardisationMeanStd
         from art.preprocessing.standardisation_mean_std.standardisation_mean_std_pytorch import (
@@ -227,7 +227,7 @@ def _apply_preprocessing_gradient(self, x, gradients, fit=False):
         :return: Gradients after backward pass through preprocessing defences.
         :rtype: Format as expected by the `model`
         """
-        import torch
+        import torch  # lgtm [py/repeated-import]
 
         from art.preprocessing.standardisation_mean_std.standardisation_mean_std import StandardisationMeanStd
         from art.preprocessing.standardisation_mean_std.standardisation_mean_std_pytorch import (
@@ -290,9 +290,9 @@ def _set_layer(self, train: bool, layerinfo: List["torch.nn.modules.Module"]) ->
         :param train: False for evaluation mode.
         :param layerinfo: List of module types.
         """
-        from torch import nn
+        import torch  # lgtm [py/repeated-import]
 
-        assert all([issubclass(l, nn.modules.Module) for l in layerinfo])
+        assert all([issubclass(l, torch.nn.modules.Module) for l in layerinfo])
 
         def set_train(layer, layerinfo=layerinfo):
             "Set layer into training mode if instance of `layerinfo`."
@@ -315,16 +315,16 @@ def set_dropout(self, train: bool) -> None:
 
         :param train: False for evaluation mode.
         """
-        from torch import nn
+        import torch  # lgtm [py/repeated-import]
 
-        self._set_layer(train=train, layerinfo=[nn.modules.dropout._DropoutNd])
+        self._set_layer(train=train, layerinfo=[torch.nn.modules.dropout._DropoutNd])
 
     def set_batchnorm(self, train: bool) -> None:
         """
         Set all batch normalization layers into train or eval mode.
 
         :param train: False for evaluation mode.
         """
-        from torch import nn
+        import torch  # lgtm [py/repeated-import]
 
-        self._set_layer(train=train, layerinfo=[nn.modules.batchnorm._BatchNorm])
+        self._set_layer(train=train, layerinfo=[torch.nn.modules.batchnorm._BatchNorm])