AttributeError：“str”对象没有属性“sc”Pyspark PMML

Question

第一次在这里发帖！我正在尝试通过 pyspark2pmml 保存我的逻辑回归模型。但是，我不断收到标题中所述的错误。我将发布我的管道和模型代码。

from pyspark.ml.feature import Binarizer

binarizer = Binarizer(threshold=10000, inputCol="traffic_count", outputCol="label")
stages = []
stages = [binarizer]


from pyspark.ml import Pipeline
from pyspark.ml.feature import StringIndexer, OneHotEncoder, VectorAssembler


SI_roadname = StringIndexer(inputCol='road_name',outputCol='road_Index')
SI_suburb = StringIndexer(inputCol='suburb',outputCol='suburb_Index')
SI_cardinal = StringIndexer(inputCol='cardinal_direction_name',outputCol='cardinal_Index')
SI_period = StringIndexer(inputCol='period',outputCol='period_Index')
SI_label = StringIndexer(inputCol='label',outputCol='label_index')

stages = []
stages += [SI_roadname, SI_suburb, SI_cardinal, SI_period, SI_label]

OHE = OneHotEncoder(inputCols['road_Index','suburb_Index','cardinal_Index','period_Index','label_index'],outputCols=['road_OHE','suburb_OHE','cardinal_OHE','period_OHE','label_OHE'])
stages += [OHE]


assembler = VectorAssembler(inputCols=['wgs84_latitude','wgs84_longitude'],outputCol='features')
stages += [assembler]

pipeline = Pipeline(stages=stages)
pipelineModel = pipeline.fit(df)
model = pipelineModel.transform(df)

from pyspark.ml.linalg import DenseVector
input_data = model.rdd.map(lambda x: (x["label"], DenseVector(x["features"])))
df_train = sqlContext.createDataFrame(input_data, ["label", "features"])


train, test = df_train.randomSplit([0.7,0.3])

lr = LogisticRegression(labelCol='label')
lr_model = lr.fit(train)


pred_labels = lr_model.evaluate(test)
pred_labels.predictions.show()

所以我得到的特定错误来自这一行

from pyspark2pmml import PMMLBuilder

PMMLBuilder(spark, df, pipelineModel)
PMMLBuilder.buildFile("lr_model.pmml","path")

我对使用 Pyspark 很陌生，所以我希望有人能帮帮我。我也会发布一些屏幕截图作为上下文。

The Dataframe

The error

model.take(1)

Predictions

Answer 1

您正在用字符串“lr_model.pmml”覆盖对象的 self 参数。这就是您收到错误AttributeError: 'str' object has no attribute 'sc' Pyspark PMML 的原因。 您必须调用 buildFile 作为参数传递路径 see。

    def buildFile(self, path):
        javaFile = self.sc._jvm.java.io.File(path)
        javaFile = self.javaPmmlBuilder.buildFile(javaFile)
        return javaFile.getAbsolutePath()

来自库的自述文件：

from pyspark2pmml import PMMLBuilder

pmmlBuilder = PMMLBuilder(sc, df, pipelineModel)

pmmlBuilder.buildFile("DecisionTreeIris.pmml")