Keras - LeakyReLU 保存模型时没有属性名称错误

Question

我在我的模型中使用 LeakyReLU 激活。我可以训练它。但是当我训练保存模型时，

discriminator_model.save(os.path.join(output_folder_path, 'discriminator_model_{0}.h5'.format(iteration_no)))

我收到以下错误

AttributeError: 'LeakyReLU' object has no attribute '__name__'

我将 keras-gpu 2.2.4 与 tensorflow-gpu 1.12.0 后端一起使用。这是我的模型：

discriminator_model = Sequential()
discriminator_model.add(Conv2D(64, 5, strides=2, input_shape=(28, 28, 1), padding='same', activation=LeakyReLU(alpha=0.2)))
discriminator_model.add(Dropout(0.4))
discriminator_model.add(Conv2D(128, 5, strides=2, padding='same'))
discriminator_model.add(LeakyReLU(alpha=0.2))
discriminator_model.add(Dropout(0.4))
discriminator_model.add(Conv2D(256, 5, strides=2, padding='same'))
discriminator_model.add(LeakyReLU(alpha=0.2))
discriminator_model.add(Dropout(0.4))
discriminator_model.add(Conv2D(512, 5, strides=2, padding='same'))
discriminator_model.add(LeakyReLU(alpha=0.2))
discriminator_model.add(Dropout(0.4))
discriminator_model.add(Flatten())
discriminator_model.add(Dense(1))
discriminator_model.add(Activation('sigmoid'))
discriminator_model.summary()

最初，我正在使用

discriminator_model.add(Conv2D(128, 5, strides=2, padding='same', activation=LeakyReLU(alpha=0.2)))

但是这里和这里都建议添加 LeakyReLU 作为单独的激活层。即使在尝试之后也没有运气。

完整的堆栈跟踪：

Traceback (most recent call last):
  File "/opt/PyCharm/pycharm-community-2018.3.3/helpers/pydev/pydevd.py", line 1741, in <module>
    main()
  File "/opt/PyCharm/pycharm-community-2018.3.3/helpers/pydev/pydevd.py", line 1735, in main
    globals = debugger.run(setup['file'], None, None, is_module)
  File "/opt/PyCharm/pycharm-community-2018.3.3/helpers/pydev/pydevd.py", line 1135, in run
    pydev_imports.execfile(file, globals, locals)  # execute the script
  File "/opt/PyCharm/pycharm-community-2018.3.3/helpers/pydev/_pydev_imps/_pydev_execfile.py", line 18, in execfile
    exec(compile(contents+"\n", file, 'exec'), glob, loc)
  File "..../Workspace/src/v01/MnistTrainer.py", line 100, in <module>
    main()
  File "..../Workspace/src/v01/MnistTrainer.py", line 92, in main
    mnist_trainer.train(train_steps=100, log_interval=1, save_interval=1)
  File "..../Workspace/src/v01/MnistTrainer.py", line 56, in train
    self.save_models(output_folder_path, i + 1)
  File "..../Workspace/src/v01/MnistTrainer.py", line 69, in save_models
    os.path.join(output_folder_path, 'discriminator_model_{0}.h5'.format(iteration_no)))
  File "..../.conda/envs/e0_270_ml_3.6/lib/python3.6/site-packages/keras/engine/network.py", line 1090, in save
    save_model(self, filepath, overwrite, include_optimizer)
  File "..../.conda/envs/e0_270_ml_3.6/lib/python3.6/site-packages/keras/engine/saving.py", line 382, in save_model
    _serialize_model(model, f, include_optimizer)
  File "..../.conda/envs/e0_270_ml_3.6/lib/python3.6/site-packages/keras/engine/saving.py", line 83, in _serialize_model
    model_config['config'] = model.get_config()
  File "..../.conda/envs/e0_270_ml_3.6/lib/python3.6/site-packages/keras/engine/sequential.py", line 278, in get_config
    'config': layer.get_config()
  File "..../.conda/envs/e0_270_ml_3.6/lib/python3.6/site-packages/keras/layers/convolutional.py", line 493, in get_config
    config = super(Conv2D, self).get_config()
  File "..../.conda/envs/e0_270_ml_3.6/lib/python3.6/site-packages/keras/layers/convolutional.py", line 226, in get_config
    'activation': activations.serialize(self.activation),
  File "..../.conda/envs/e0_270_ml_3.6/lib/python3.6/site-packages/keras/activations.py", line 176, in serialize
    return activation.__name__
AttributeError: 'LeakyReLU' object has no attribute '__name__'

Answer 1

编辑部分（感谢@NagabhushanSN 提及剩余问题）

我们仍然有一行代码discriminator_model.add(Conv2D(64, 5, strides=2, input_shape=(28, 28, 1), padding='same', activation=LeakyReLU(alpha=0.2)))，它是代码的第二行。

如果我们修改该行，最终更正后的代码应该是这样的：

discriminator_model = Sequential()
discriminator_model.add(Conv2D(64, 5, strides=2, input_shape=(28, 28, 1), padding='same'))
discriminator_model.add(LeakyReLU(alpha=0.2))
discriminator_model.add(Dropout(0.4))
discriminator_model.add(Conv2D(128, 5, strides=2, padding='same'))
discriminator_model.add(LeakyReLU(alpha=0.2))
discriminator_model.add(Dropout(0.4))
discriminator_model.add(Conv2D(256, 5, strides=2, padding='same'))
discriminator_model.add(LeakyReLU(alpha=0.2))
discriminator_model.add(Dropout(0.4))
discriminator_model.add(Conv2D(512, 5, strides=2, padding='same'))
discriminator_model.add(LeakyReLU(alpha=0.2))
discriminator_model.add(Dropout(0.4))
discriminator_model.add(Flatten())
discriminator_model.add(Dense(1))
discriminator_model.add(Activation('sigmoid'))
discriminator_model.summary()

这个应该在最新版本的tensroflow上运行良好，我在1.8.0上测试过，它运行得很好。但是，如果使用旧版本（例如 tesnorflow1.1.0）检查您的代码，我们会得到相同的错误。

对于这种情况，我建议将tensorflow更新到更高版本

• 要检查 python 使用的当前 Tensorflow 版本，请按照此处执行操作。
• 要更新张量流，这篇文章似乎足以展示如何做到这一点。