使用新版joblib加载旧版保存模型出现KeyError:0的解决方法

问题描述 #

我安装了最新版本的joblib，但忘记保存模型时使用的joblib版本，使用Python加载该旧版本保存的机器学习模型时出现KeyError:0错误，请问是否有解决办法？

相关代码 #

import joblib
# Load the saved model
model = joblib.load("/content/drive/MyDrive/dataset/Base Paper dataset results/Classifers/Linear_SVC_classifier.sav")

# Check the version of Joblib used to save the model
print("Joblib version used to save the model:", joblib.__version__)

报错信息 #

KeyError                                  Traceback (most recent call last)
<ipython-input-117-f85138883bf7> in <cell line: 3>()
      1 import joblib
      2 # Load the saved model
----> 3 model = joblib.load("/content/drive/MyDrive/dataset/Base Paper dataset results/Classifers/Linear_SVC_classifier.sav")
      4 
      5 # Check the version of Joblib used to save the model

2 frames
/usr/local/lib/python3.9/dist-packages/joblib/numpy_pickle.py in load(filename, mmap_mode)
    585                 obj = _unpickle(fobj, filename, mmap_mode)
    586     return obj

/usr/local/lib/python3.9/dist-packages/joblib/numpy_pickle.py in _unpickle(fobj, filename, mmap_mode)
    504         obj = unpickler.load()
    505         if unpickler.compat_mode:
--> 506             warnings.warn("The file '%s' has been generated with a "
    507                           "joblib version less than 0.10. "
    508                           "Please regenerate this pickle file."

/usr/lib/python3.9/pickle.py in load(self)
   1210                     raise EOFError
   1211                 assert isinstance(key, bytes_types)
--> 1212                 dispatch[key[0]](self)
   1213         except _Stop as stopinst:
   1214             return stopinst.value

KeyError: 0

解决办法 #

• 降级Joblib到兼容版本：
  从报错提示可知，模型是用joblib 0.10之前的版本保存的。先尝试安装旧版本，比如0.9.0：

  pip install joblib==0.9.0
  

  安装完成后重新运行加载代码，若仍报错，可尝试更早的版本如0.8.4。

• 尝试用标准库pickle加载：
  Joblib基于pickle开发，部分旧版本joblib保存的模型可直接用pickle读取：

  import pickle
  with open("/content/drive/MyDrive/dataset/Base Paper dataset results/Classifers/Linear_SVC_classifier.sav", 'rb') as f:
      model = pickle.load(f)
  

• 还原保存模型时的环境：
  若能找到保存模型时的完整Python环境（包括joblib、scikit-learn等依赖的版本），在该环境下加载模型后，用新版本joblib重新保存一次，后续即可用新版本正常加载。

内容的提问来源于stack exchange，提问作者Umer