使用检查点在 Keras 中使用自定义损失函数恢复训练

Question

我正在使用 keras（带有 Tensorflow 后端）训练一个模型，它的损失函数由我定义（在代码中命名为 custom_loss），我在训练期间以最佳准确度保存模型：

model = Sequential()
model.add(...)

adam = keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, decay=0.01)
model.compile(loss=custom_loss, optimizer=adam, metrics=['accuracy'])  # better
filepath="weights.best.hdf5"
checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max')
callbacks_list = [checkpoint]
# Fit the model
model.fit(x_train, y_train, validation_split=0.1, epochs=150, batch_size=64, callbacks=callbacks_list)

我想在一些 epoch 后停止训练并重新加载模型以从该保存点恢复训练：

adam = keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, decay=0.01)

#Reload model
model = load_model('weights.best.hdf5')
model.compile(loss=custom_loss, optimizer=adam, metrics=['accuracy']) 

# checkpoint
filepath="weights.best.hdf5"
checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max')
callbacks_list = [checkpoint]

#Continue training
model.fit(x_train, y_train, validation_split=0.1, epochs=150,
      batch_size=64, callbacks=callbacks_list)

但是，在加载模型后，我收到了这个错误，说 custom_loss 是未知的；问题是什么？ custom_loss 在主类中定义：

  File "C:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 2073, in <module>
main()
  File "C:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 2067, in main
  globals = debugger.run(setup['file'], None, None, is_module)
  File "C:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 1418, in run
  return self._exec(is_module, entry_point_fn, module_name, file, globals, locals)
  File "C:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\pydevd.py", line 1425, in _exec 
  pydev_imports.execfile(file, globals, locals)  # execute the script
  File "C:\Program Files\JetBrains\PyCharm 2019.2.3\helpers\pydev\_pydev_imps\_pydev_execfile.py", 
  line 18, in execfile
  exec(compile(contents+"\n", file, 'exec'), glob, loc)
  File "C:/Projects/ML/Test_Yolo.py", line 194, in <module>
  model = load_model('weights.best.hdf5')
  File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site 
  packages\keras\engine\saving.py", line 492, in load_wrapper
  return load_function(*args, **kwargs)
  File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site- 
packages\keras\engine\saving.py", line 584, in load_model
model = _deserialize_model(h5dict, custom_objects, compile)
File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site- 
packages\keras\engine\saving.py", line 369, in _deserialize_model
sample_weight_mode=sample_weight_mode)
File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site- 
packages\keras\engine\training.py", line 119, in compile
self.loss, self.output_names)
File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site- 
packages\keras\engine\training_utils.py", line 822, in prepare_loss_functions
loss_functions = [get_loss_function(loss) for _ in output_names]
File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site- 
packages\keras\engine\training_utils.py", line 822, in <listcomp>
loss_functions = [get_loss_function(loss) for _ in output_names]
File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site- 
packages\keras\engine\training_utils.py", line 705, in get_loss_function
loss_fn = losses.get(loss)
File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site- 
packages\keras\losses.py", line 795, in get
return deserialize(identifier)
File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site-packages\keras\losses.py", line 
776, in deserialize
printable_module_name='loss function')
File "C:\Users\AppData\Local\conda\conda\envs\tensorflow_env\lib\site- 
packages\keras\utils\generic_utils.py", line 167, in deserialize_keras_object
':' + function_name)
ValueError: Unknown loss function:custom_loss

Answer 1

加载模型时，必须使用custom_objects 参数指定自定义损失函数（参见docs）：

custom_objects: 可选字典映射名称（字符串）到自定义 反序列化过程中要考虑的类或函数。

试试这个：

model = load_model('weights.best.hdf5', custom_objects={'loss': custom_loss})

你也应该看看this。