Remove support for MXNet and TensorFlow v1

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

Author: beat-buesser

.github/ISSUE_TEMPLATE/bug_report.md

@@ -24,4 +24,4 @@ If applicable, add screenshots to help explain your problem.
  - OS
  - Python version
  - ART version or commit number
- - TensorFlow / Keras / PyTorch / MXNet version
+ - TensorFlow / Keras / PyTorch version

.github/workflows/ci-mxnet.yml

@@ -7,7 +7,7 @@ RUN apt-get install -y python3 python3-distutils python3-pip
 RUN pip3 install tensorflow==2.9.1 keras==2.9.0 numpy==1.22.4 scipy==1.8.1 matplotlib==3.5.2 scikit-learn==1.1.2 \
                  six==1.15.0 Pillow==9.2.0 pytest-cov==3.0.0 tqdm==4.64.0 statsmodels==0.13.2 pydub==0.25.1 \
                  resampy==0.3.1 ffmpeg-python==0.2.0 cma==3.2.2 pandas==1.4.3 h5py==3.7.0 tensorflow-addons==0.17.1 \
-                 mxnet==1.6.0 torch==1.12.0 torchaudio==0.12.0 torchvision==0.13.0 catboost==1.0.6 GPy==1.10.0 \
+                 torch==1.12.0 torchaudio==0.12.0 torchvision==0.13.0 catboost==1.0.6 GPy==1.10.0 \
                  lightgbm==3.3.2 xgboost==1.6.1 kornia==0.6.6 lief==0.12.1 pytest==7.1.2 pytest-pep8==1.0.6 \
                  pytest-mock==3.8.2 requests==2.28.1
 

.github/workflows/ci-tensorflow-v1.yml

@@ -25,7 +25,7 @@ Please describe the tests that you ran to verify your changes. Consider listing
 - OS
 - Python version
 - ART version or commit number
-- TensorFlow / Keras / PyTorch / MXNet version
+- TensorFlow / Keras / PyTorch version
 
 # Checklist
 

Dockerfile

@@ -24,7 +24,7 @@
 [Linux Foundation AI & Data Foundation](https://lfaidata.foundation) (LF AI & Data)。 ART提供的工具可
 帮助开发人员和研究人员针对以下方面捍卫和评估机器学习模型和应用程序：
 逃逸，数据污染，模型提取和推断的对抗性威胁。ART支持所有流行的机器学习框架
-（TensorFlow，Keras，PyTorch，MXNet，scikit-learn，XGBoost，LightGBM，CatBoost，GPy等），所有数据类型
+（TensorFlow，Keras，PyTorch，scikit-learn，XGBoost，LightGBM，CatBoost，GPy等），所有数据类型
 （图像，表格，音频，视频等）和机器学习任务（分类，物体检测，语音识别，
 生成模型，认证等）。
 

PULL_REQUEST_TEMPLATE.md

@@ -26,7 +26,7 @@ Adversarial Robustness Toolbox (ART) is a Python library for Machine Learning Se
 [Linux Foundation AI & Data Foundation](https://lfaidata.foundation) (LF AI & Data). ART provides tools that enable
 developers and researchers to defend and evaluate Machine Learning models and applications against the
 adversarial threats of Evasion, Poisoning, Extraction, and Inference. ART supports all popular machine learning frameworks
-(TensorFlow, Keras, PyTorch, MXNet, scikit-learn, XGBoost, LightGBM, CatBoost, GPy, etc.), all data types
+(TensorFlow, Keras, PyTorch, scikit-learn, XGBoost, LightGBM, CatBoost, GPy, etc.), all data types
 (images, tables, audio, video, etc.) and machine learning tasks (classification, object detection, speech recognition,
 generation, certification, etc.).
 

README-cn.md

@@ -92,10 +92,7 @@ def __init__(
             "cross_entropy", or "difference_logits_ratio"
         :param verbose: Show progress bars.
         """
-        from art.estimators.classification import TensorFlowClassifier, TensorFlowV2Classifier, PyTorchClassifier
-
-        if isinstance(estimator, TensorFlowClassifier):
-            raise ValueError("This attack does not support TensorFlow  v1.")
+        from art.estimators.classification import  TensorFlowV2Classifier, PyTorchClassifier
 
         if loss_type not in self._predefined_losses:
             raise ValueError(

README.md

@@ -301,8 +301,7 @@ def tensor_norm(tensor, norm_type: int | float | str = 2):  # pylint: disable=in
         "tensorflow.python.framework.ops.SymbolicTensor",
     )
     torch_tensor_types = ("torch.Tensor", "torch.float", "torch.double", "torch.long")
-    mxnet_tensor_types = ()
-    supported_types = tf_tensor_types + torch_tensor_types + mxnet_tensor_types
+    supported_types = tf_tensor_types + torch_tensor_types
     tensor_type = get_class_name(tensor)
     if tensor_type not in supported_types:  # pragma: no cover
         raise TypeError("Tensor type `" + tensor_type + "` is not supported")
@@ -316,8 +315,3 @@ def tensor_norm(tensor, norm_type: int | float | str = 2):  # pylint: disable=in
         import torch
 
         return torch.norm
-
-    if tensor_type in mxnet_tensor_types:  # pragma: no cover
-        import mxnet
-
-        return mxnet.ndarray.norm(tensor, ord=norm_type)

art/attacks/evasion/auto_projected_gradient_descent.py

@@ -33,7 +33,6 @@
 
 if TYPE_CHECKING:
     import keras
-    import mxnet
     import tensorflow as tf
     import torch
 
@@ -258,47 +257,6 @@ def get_batch(self) -> tuple:
         return tuple(batch)
 
 
-class MXDataGenerator(DataGenerator):
-    """
-    Wrapper class on top of the MXNet/Gluon native data loader :class:`mxnet.gluon.data.DataLoader`.
-    """
-
-    def __init__(self, iterator: "mxnet.gluon.data.DataLoader", size: int, batch_size: int) -> None:
-        """
-        Create a data generator wrapper on top of an MXNet :class:`DataLoader`.
-
-        :param iterator: A MXNet DataLoader instance.
-        :param size: Total size of the dataset.
-        :param batch_size: Size of the minibatches.
-        """
-        import mxnet
-
-        super().__init__(size=size, batch_size=batch_size)
-        if not isinstance(iterator, mxnet.gluon.data.DataLoader):
-            raise TypeError(f"Expected instance of Gluon `DataLoader, received {type(iterator)} instead.`")
-
-        self._iterator = iterator
-        self._current = iter(self.iterator)
-
-    def get_batch(self) -> tuple:
-        """
-        Provide the next batch for training in the form of a tuple `(x, y)`. The generator should loop over the data
-        indefinitely.
-
-        :return: A tuple containing a batch of data `(x, y)`.
-        """
-        try:
-            batch = list(next(self._current))
-        except StopIteration:
-            self._current = iter(self.iterator)
-            batch = list(next(self._current))
-
-        for i, item in enumerate(batch):
-            batch[i] = item.asnumpy()
-
-        return tuple(batch)
-
-
 class TensorFlowDataGenerator(DataGenerator):  # pragma: no cover
     """
     Wrapper class on top of the TensorFlow native iterators :class:`tf.data.Iterator`.