在Python 3.8中调用sklearn2pmml()函数抛出RuntimeError求助

问题复现 #

代码示例 #

from sklearn2pmml import sklearn2pmml
from sklearn2pmml.pipeline import PMMLPipeline
from sklearn.linear_model import LogisticRegression

pipe_pmml = PMMLPipeline(steps=[('mapper', mapper),
                                ('estimator', LogisticRegression(C = 0.01, 
                                  penalty = 'l1', 
                                  solver = 'liblinear', 
                                  random_state = 1))
                                ])
pipe_pmml.fit(X_small, y)

sklearn2pmml(pipe_pmml, pmml_filename, with_repr = True)

注：mapper为sklearn_pandas.DataFrameMapper实例

报错信息 #

Standard output is empty
Standard error:
Exception in thread "main" net.razorvine.pickle.InvalidOpcodeException: invalid pickle opcode: 0
    at net.razorvine.pickle.Unpickler.dispatch(Unpickler.java:366)
    at org.jpmml.python.CustomUnpickler.dispatch(CustomUnpickler.java:31)
    at org.jpmml.python.PickleUtil$1.dispatch(PickleUtil.java:64)
    at net.razorvine.pickle.Unpickler.load(Unpickler.java:109)
    at org.jpmml.python.PickleUtil.unpickle(PickleUtil.java:85)
    at com.sklearn2pmml.Main.run(Main.java:78)
    at com.sklearn2pmml.Main.main(Main.java:6

当前依赖版本 #

• sklearn==0.0
• scikit-learn==1.1.2
• sklearn-pandas==2.2.0
• sklearn2pmml==0.86.3

解决思路 #

• 移除冲突的占位依赖：sklearn==0.0是无实际功能的占位包，与scikit-learn==1.1.2存在依赖冲突，会干扰pickle序列化流程。执行以下命令卸载：

  pip uninstall -y sklearn
  

• 对齐版本兼容性：sklearn2pmml对scikit-learn版本有严格适配要求，0.86.3版本仅支持到scikit-learn 1.0.x，而当前使用的1.1.2超出适配范围，导致底层解析出错。两种调整方案：
  • 降级scikit-learn到兼容版本：

    pip install scikit-learn==1.0.2
    

  • 升级sklearn2pmml到支持1.1.x的版本（如0.96.0及以上）：

    pip install sklearn2pmml==0.96.0
    

• 简化序列化参数：去掉sklearn2pmml函数中的with_repr=True参数，该参数会额外序列化对象的repr信息，增加解析风险，修改为：

  sklearn2pmml(pipe_pmml, pmml_filename)
  

• 验证第三方Mapper兼容性：若调整后仍有问题，可尝试将DataFrameMapper替换为scikit-learn原生的ColumnTransformer，避免第三方组件的序列化兼容性问题。

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