Keras問題“AttributeError: 'NoneType' object has no attribute 'update”解決

BUG

在使用Keras訓練模型時，在每個epoch完成後save_model時會報錯 “AttributeError: 'NoneType' object has no attribute 'update'”

具體異常打印信息如下，主要原因是模型中有自定義的class，Keras不知道怎麼進行deep_copy()

File "train.py", line 88, in <module>
    build_model()
  File "train.py", line 80, in build_model
    CSVLogger(log_path),
  File "/usr/python/lib/python3.5/site-packages/keras/legacy/interfaces.py", line 91, in wrapper
    return func(*args, **kwargs)
  File "/usr/python/lib/python3.5/site-packages/keras/engine/training.py", line 2268, in fit_generator
    callbacks.on_epoch_end(epoch, epoch_logs)
  File "/usr/python/lib/python3.5/site-packages/keras/callbacks.py", line 77, in on_epoch_end
    callback.on_epoch_end(epoch, logs)
  File "/usr/python/lib/python3.5/site-packages/keras/callbacks.py", line 447, in on_epoch_end
    self.model.save(filepath, overwrite=True)
  File "/usr/python/lib/python3.5/site-packages/keras/engine/topology.py", line 2591, in save
    save_model(self, filepath, overwrite, include_optimizer)
  File "/usr/python/lib/python3.5/site-packages/keras/models.py", line 126, in save_model
    'config': model.get_config()
  File "/usr/python/lib/python3.5/site-packages/keras/engine/topology.py", line 2432, in get_config
    return copy.deepcopy(config)

解決辦法

方法一

在回調函數callbacks中加入save_weights_only=True，加載模型時new Model()， 然後載入weights，這樣避免deep_copy()

callbacks=[ModelCheckpoint(model_path, save_weights_only=True, 
monitor='val_loss', mode='min', save_best_only=True)]

方法二

對無法deep_copy()的class，自定義複製方法。

class YourClass(object):
   # ...
   def __deepcopy__(self):
      return Your_deep_copy()