Keras模型训练报错：unhashable type: 'list'问题求助

问题详情 #

之前可正常运行的代码如今出现报错，不添加回调函数时一切正常。

相关代码 #

checkpoint = ModelCheckpoint('best_model.hdf5' , monitor = ['val_accuracy'] , verbose = 1  , mode = 'max', save_best_only=True)
earlystop = EarlyStopping(monitor='val_accuracy', patience=3, restore_best_weights=True)
callbacks = [checkpoint, earlystop]

...

history = model.fit(train_generator, 
          steps_per_epoch = len(train_generator),
          validation_data=val_generator, 
          validation_steps = len(val_generator),
          epochs=EPOCHS,
          callbacks=checkpoint)

报错信息 #

TypeError                                 Traceback (most recent call last)
<ipython-input-39-805f1a83aa3c> in <module>
      4           validation_steps = len(val_generator),
      5           epochs=EPOCHS,
----> 6           callbacks=checkpoint)

1 frames
/usr/local/lib/python3.7/dist-packages/keras/utils/traceback_utils.py in error_handler(*args, **kwargs)
     65     except Exception as e:  # pylint: disable=broad-except
     66       filtered_tb = _process_traceback_frames(e.__traceback__)
---> 67       raise e.with_traceback(filtered_tb) from None
     68     finally:
     69       del filtered_tb

/usr/local/lib/python3.7/dist-packages/keras/callbacks.py in _save_model(self, epoch, batch, logs)
   1426       try:
   1427         if self.save_best_only:
-> 1428           current = logs.get(self.monitor)
   1429           if current is None:
   1430             logging.warning('Can save best model only with %s available, '

TypeError: unhashable type: 'list'

原因分析 #

• 监控参数类型错误：ModelCheckpoint的monitor参数被传入了列表['val_accuracy']，但该参数要求传入字符串类型的指标名称。logs是字典结构，字典的键必须是可哈希类型（如字符串），用列表当键会触发unhashable type: 'list'错误。
• 回调参数格式错误：model.fit的callbacks参数要求传入回调对象的列表，但代码中直接传了单个checkpoint对象，不符合参数要求。

解决办法 #

修正两处错误即可：

1. 将ModelCheckpoint的monitor参数改为字符串'val_accuracy'
2. model.fit的callbacks传入定义好的callbacks列表（若仅需单个回调，可传[checkpoint]）

修正后的代码示例：

# 修正monitor参数为字符串
checkpoint = ModelCheckpoint('best_model.hdf5' , monitor='val_accuracy' , verbose=1 , mode='max', save_best_only=True)
earlystop = EarlyStopping(monitor='val_accuracy', patience=3, restore_best_weights=True)
callbacks = [checkpoint, earlystop]

...

# 传入回调列表
history = model.fit(train_generator, 
          steps_per_epoch = len(train_generator),
          validation_data=val_generator, 
          validation_steps = len(val_generator),
          epochs=EPOCHS,
          callbacks=callbacks)

内容的提问来源于stack exchange，提问作者Sergei Frolov