使用sklearn2pmml将XGBoost分类器写入pmml时出错

Question

我想使用sklearn2pmml将我的XGBoost模型保存为pmml。我使用的是PythonV3.7.3和Sklearn 0.20.3 & sklearn2pmml V0.53.0。我的数据主要是二进制的，只有3列连续数据，我在Databricks中运行我的笔记本，并将我的Spark数据帧转换为pandas数据帧。下面的代码片段

​

import xgboost as xgb

from sklearn_pandas import DataFrameMapper
from sklearn.compose import ColumnTransformer

from sklearn2pmml import sklearn2pmml
from sklearn2pmml.pipeline import PMMLPipeline
from sklearn2pmml.decoration import ContinuousDomain
from sklearn.preprocessing import StandardScaler

X = pdf[continuous_features + numericCols]
y = pdf["Label"]


mapper = DataFrameMapper(
  [([cont_column], [ContinuousDomain(), StandardScaler()]) for cont_column in continuous_features] +
  [([c for c in numericCols], None)] # no transformation
)

clf = xgb.XGBClassifier(objective='multi:softprob',eval_metric='auc',num_class = 2,
                        n_jobs =6,max_delta_step=1, min_child_weight=14, gamma=1.5, subsample = 0.8,
                        colsample_bytree = 0.5, max_depth=10, learning_rate = 0.1)


pipeline = PMMLPipeline([
  ("mapper", mapper),
  ("estimator", clf)
])

pipeline.fit(X,y.values.reshape(-1,))

sklearn2pmml(pipeline, "xgb_V1.pmml", with_repr = True)

​

管道根据数据进行拟合，使用pipeline.score(X，y)和pipeline.predict(X)生成分数和预测，但当我尝试将其写入pmml时，我得到以下错误：

​

Standard output is empty
Standard error:
Feb 21, 2020 1:53:30 PM org.jpmml.sklearn.Main run
INFO: Parsing PKL..
Feb 21, 2020 1:53:30 PM org.jpmml.sklearn.Main run
INFO: Parsed PKL in 47 ms.
Feb 21, 2020 1:53:30 PM org.jpmml.sklearn.Main run
INFO: Converting..
Feb 21, 2020 1:53:30 PM sklearn2pmml.pipeline.PMMLPipeline initTargetFields
WARNING: Attribute 'sklearn2pmml.pipeline.PMMLPipeline.target_fields' is not set. Assuming y as the name of the target field
Feb 21, 2020 1:53:30 PM org.jpmml.sklearn.Main run
SEVERE: Failed to convert
java.lang.IllegalArgumentException: Attribute 'xgboost.sklearn.XGBClassifier._le' has an unsupported value (Python class xgboost.compat.XGBoostLabelEncoder)
	at org.jpmml.sklearn.CastFunction.apply(CastFunction.java:45)
	at org.jpmml.sklearn.PyClassDict.get(PyClassDict.java:82)
	at sklearn.LabelEncoderClassifier.getLabelEncoder(LabelEncoderClassifier.java:40)
	at sklearn.LabelEncoderClassifier.getClasses(LabelEncoderClassifier.java:34)
	at sklearn.ClassifierUtil.getClasses(ClassifierUtil.java:32)
	at sklearn2pmml.pipeline.PMMLPipeline.encodePMML(PMMLPipeline.java:133)
	at org.jpmml.sklearn.Main.run(Main.java:145)
	at org.jpmml.sklearn.Main.main(Main.java:94)
Caused by: java.lang.ClassCastException: Cannot cast net.razorvine.pickle.objects.ClassDict to sklearn.preprocessing.LabelEncoder
	at java.lang.Class.cast(Class.java:3369)
	at org.jpmml.sklearn.CastFunction.apply(CastFunction.java:43)
	... 7 more

Exception in thread "main" java.lang.IllegalArgumentException: Attribute 'xgboost.sklearn.XGBClassifier._le' has an unsupported value (Python class xgboost.compat.XGBoostLabelEncoder)
	at org.jpmml.sklearn.CastFunction.apply(CastFunction.java:45)
	at org.jpmml.sklearn.PyClassDict.get(PyClassDict.java:82)
	at sklearn.LabelEncoderClassifier.getLabelEncoder(LabelEncoderClassifier.java:40)
	at sklearn.LabelEncoderClassifier.getClasses(LabelEncoderClassifier.java:34)
	at sklearn.ClassifierUtil.getClasses(ClassifierUtil.java:32)
	at sklearn2pmml.pipeline.PMMLPipeline.encodePMML(PMMLPipeline.java:133)
	at org.jpmml.sklearn.Main.run(Main.java:145)
	at org.jpmml.sklearn.Main.main(Main.java:94)
Caused by: java.lang.ClassCastException: Cannot cast net.razorvine.pickle.objects.ClassDict to sklearn.preprocessing.LabelEncoder
	at java.lang.Class.cast(Class.java:3369)
	at org.jpmml.sklearn.CastFunction.apply(CastFunction.java:43)

​

根据这篇文章https://github.com/jpmml/sklearn2pmml/issues/197，我认为这可能是Sklearn和sklearn2pmml之间的版本不兼容问题，但我认为我安装的版本应该是可以的。你知道这是怎么回事吗？提前感谢

Answer 1

这可能是XGBoost包的版本问题。SkLearn2PMML包期望标签编码器(XGBClassifier._le属性)是一个“普通的”Scikit-Learn标签编码器类(sklearn.preprocessing.(label|_label).LabelEncoder)，但在您的例子中是不同的(xgboost.compat.XGBoostLabelEncoder)。

此xgboost.compat.XGBoostLabelEncoder是在哪个XGBOost包版本中引入的？要么是很老的东西，要么是很新的东西。

在任何情况下，请使用JPMML SkLearn项目here打开一个功能请求来解决这个问题。