python - sklearn2pmml 错误 : expected zero arguments for construction of ClassDict (for pandas. _libs.interval.Interval)

Question

我在训练 LR 模型时使用 sklearn2pmml.preprocessing.CutTransformer 和 sklearn.preprocessing.LabelEncoder 对目标进行编码。

这是我的代码:

from sklearn2pmml.preprocessing import CutTransformer
from sklearn.preprocessing.label import LabelEncoder
income_bins = [-np.inf, 10000, 50000, 100000, 300000, 500000, 1000000, 3000000, 5000000, 10000000, np.inf]

targetDiscretizer = PMMLPipeline([('target', 
                               DataFrameMapper([
                                   (['income'], [CutTransformer(bins=income_bins), LabelEncoder()])
                               ])
                              )])
dataset['target_income_lvl'] = targetDiscretizer.fit_transform(dataset)
sklearn2pmml(targetDiscretizer, '../model/targetDiscretizer.pmml', with_repr=True )

但是我收到一个错误:

net.razorvine.pickle.PickleException: expected zero arguments for construction of ClassDict (for pandas._libs.interval.Interval)
    at net.razorvine.pickle.objects.ClassDictConstructor.construct(ClassDictConstructor.java:23)
    at net.razorvine.pickle.Unpickler.load_reduce(Unpickler.java:732)
    at net.razorvine.pickle.Unpickler.dispatch(Unpickler.java:200)
    at net.razorvine.pickle.Unpickler.load(Unpickler.java:122)
    at numpy.core.NDArrayUtil.readObject(NDArrayUtil.java:384)
    at numpy.core.NDArrayUtil.access$700(NDArrayUtil.java:42)
    at numpy.core.NDArrayUtil$TypeDescriptor.read(NDArrayUtil.java:542)
    at numpy.core.NDArrayUtil.parseArray(NDArrayUtil.java:215)
    at numpy.core.NDArrayUtil.parseData(NDArrayUtil.java:190)
    at joblib.NumpyArrayWrapper.toArray(NumpyArrayWrapper.java:43)
    at org.jpmml.sklearn.PickleUtil$1.dispatch(PickleUtil.java:88)
    at net.razorvine.pickle.Unpickler.load(Unpickler.java:122)
    at org.jpmml.sklearn.PickleUtil.unpickle(PickleUtil.java:98)
    at org.jpmml.sklearn.Main.run(Main.java:104)
    at org.jpmml.sklearn.Main.main(Main.java:94)

Exception in thread "main" net.razorvine.pickle.PickleException: expected zero arguments for construction of ClassDict (for pandas._libs.interval.Interval)
at net.razorvine.pickle.objects.ClassDictConstructor.construct(ClassDictConstructor.java:23)
at net.razorvine.pickle.Unpickler.load_reduce(Unpickler.java:732)
at net.razorvine.pickle.Unpickler.dispatch(Unpickler.java:200)
at net.razorvine.pickle.Unpickler.load(Unpickler.java:122)
at numpy.core.NDArrayUtil.readObject(NDArrayUtil.java:384)
at numpy.core.NDArrayUtil.access$700(NDArrayUtil.java:42)
at numpy.core.NDArrayUtil$TypeDescriptor.read(NDArrayUtil.java:542)
at numpy.core.NDArrayUtil.parseArray(NDArrayUtil.java:215)
at numpy.core.NDArrayUtil.parseData(NDArrayUtil.java:190)
at joblib.NumpyArrayWrapper.toArray(NumpyArrayWrapper.java:43)
at org.jpmml.sklearn.PickleUtil$1.dispatch(PickleUtil.java:88)
at net.razorvine.pickle.Unpickler.load(Unpickler.java:122)
at org.jpmml.sklearn.PickleUtil.unpickle(PickleUtil.java:98)
at org.jpmml.sklearn.Main.run(Main.java:104)
at org.jpmml.sklearn.Main.main(Main.java:94)

我对此一无所知。有人可以帮助我吗？

Answer 1

默认情况下，Python pickle 文件的 Java 解析器不知道非标准 CPython 类，例如 pandas._libs.interval.Interval。它需要由每个 CPython 类(class)单独教授。例如，SkLearn2PMML问题跟踪器中有一个相关的错误报告:https://github.com/jpmml/sklearn2pmml/issues/115

如果您(至少暂时)设法抑制 pandas._libs.interval.Interval 对象的生成，则转换应该有效。最可能的来源是自动生成的垃圾箱标签。因此，请尝试使用 labels 参数显式提供 bin 标签:CutTransformer(bins = Income_bins, labels = Income_bin_labels)。