在Java Web中使用Spark MLlib訓練的模型

PMML是一種通用的配置文件，只要遵循標準的配置文件，就能夠在Spark中訓練機器學習模型，而後再web接口端去使用。目前應用最廣的就是基於Jpmml來加載模型在javaweb中應用，這樣就能夠實現跨平臺的機器學習應用了。

訓練模型

首先在spark MLlib中使用mllib包下的邏輯迴歸訓練模型：

import org.apache.spark.mllib.classification.{LogisticRegressionModel, LogisticRegressionWithLBFGS}
import org.apache.spark.mllib.evaluation.MulticlassMetrics
import org.apache.spark.mllib.regression.LabeledPoint
import org.apache.spark.mllib.util.MLUtils
    
val training = spark.sparkContext
  .parallelize(Seq("0,1 2 3 1", "1,2 4 1 5", "0,7 8 3 6", "1,2 5 6 9").map( line => LabeledPoint.parse(line)))

// Run training algorithm to build the model
val model = new LogisticRegressionWithLBFGS()
  .setNumClasses(2)
  .run(training)

val test = spark.sparkContext
  .parallelize(Seq("0,1 2 3 1").map( line => LabeledPoint.parse(line)))


// Compute raw scores on the test set.
val predictionAndLabels = test.map { case LabeledPoint(label, features) =>
  val prediction = model.predict(features)
  (prediction, label)
}

// Get evaluation metrics.
val metrics = new MulticlassMetrics(predictionAndLabels)
val accuracy = metrics.accuracy
println(s"Accuracy = $accuracy")

// Save and load model
//    model.save(spark.sparkContext, "target/tmp/scalaLogisticRegressionWithLBFGSModel")
//    val sameModel = LogisticRegressionModel.load(spark.sparkContext,"target/tmp/scalaLogisticRegressionWithLBFGSModel")

model.toPMML(spark.sparkContext, "/tmp/xhl/data/test2")

訓練獲得的模型保存到hdfs。

PMML模型文件

模型下載到本地，從新命名爲xml。
能夠看到默認四個特徵分別叫作feild_0，field_1...目標爲target

<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
<PMML version="4.2" xmlns="http://www.dmg.org/PMML-4_2">
    <Header description="logistic regression">
        <Application name="Apache Spark MLlib" version="2.2.0"/>
        <Timestamp>2018-11-15T10:22:25</Timestamp>
    </Header>
    <DataDictionary numberOfFields="5">
        <DataField name="field_0" optype="continuous" dataType="double"/>
        <DataField name="field_1" optype="continuous" dataType="double"/>
        <DataField name="field_2" optype="continuous" dataType="double"/>
        <DataField name="field_3" optype="continuous" dataType="double"/>
        <DataField name="target" optype="categorical" dataType="string"/>
    </DataDictionary>
    <RegressionModel modelName="logistic regression" functionName="classification" normalizationMethod="logit">
        <MiningSchema>
            <MiningField name="field_0" usageType="active"/>
            <MiningField name="field_1" usageType="active"/>
            <MiningField name="field_2" usageType="active"/>
            <MiningField name="field_3" usageType="active"/>
            <MiningField name="target" usageType="target"/>
        </MiningSchema>
        <RegressionTable intercept="0.0" targetCategory="1">
            <NumericPredictor name="field_0" coefficient="-5.552297758753701"/>
            <NumericPredictor name="field_1" coefficient="-1.4863480719075117"/>
            <NumericPredictor name="field_2" coefficient="-5.7232298850417855"/>
            <NumericPredictor name="field_3" coefficient="8.134075057437393"/>
        </RegressionTable>
        <RegressionTable intercept="-0.0" targetCategory="0"/>
    </RegressionModel>
</PMML>

接口使用

在接口的web工程中引入maven jar:

<!-- https://mvnrepository.com/artifact/org.jpmml/pmml-evaluator -->
<dependency>
    <groupId>org.jpmml</groupId>
    <artifactId>pmml-evaluator</artifactId>
    <version>1.4.3</version>
</dependency>
<!-- https://mvnrepository.com/artifact/org.jpmml/pmml-evaluator-extension -->
<dependency>
    <groupId>org.jpmml</groupId>
    <artifactId>pmml-evaluator-extension</artifactId>
    <version>1.4.3</version>
</dependency>

接口代碼中直接讀取pmml，使用模型進行預測：

package soundsystem;

import org.dmg.pmml.FieldName;
import org.dmg.pmml.PMML;
import org.jpmml.evaluator.*;

import java.io.FileInputStream;
import java.io.InputStream;
import java.util.HashMap;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map;

public class PMMLDemo2 {
    private Evaluator loadPmml(){
        PMML pmml = new PMML();
        try(InputStream inputStream = new FileInputStream("/Users/xingoo/Desktop/test2.xml")){
            pmml = org.jpmml.model.PMMLUtil.unmarshal(inputStream);
        } catch (Exception e) {
            e.printStackTrace();
        }
        ModelEvaluatorFactory modelEvaluatorFactory = ModelEvaluatorFactory.newInstance();
        return modelEvaluatorFactory.newModelEvaluator(pmml);
    }
    private Object predict(Evaluator evaluator,int a, int b, int c, int d) {
        Map<String, Integer> data = new HashMap<String, Integer>();
        data.put("field_0", a);
        data.put("field_1", b);
        data.put("field_2", c);
        data.put("field_3", d);
        List<InputField> inputFields = evaluator.getInputFields();
        //過模型的原始特徵，從畫像中獲取數據，做爲模型輸入
        Map<FieldName, FieldValue> arguments = new LinkedHashMap<FieldName, FieldValue>();
        for (InputField inputField : inputFields) {
            FieldName inputFieldName = inputField.getName();
            Object rawValue = data.get(inputFieldName.getValue());
            FieldValue inputFieldValue = inputField.prepare(rawValue);
            arguments.put(inputFieldName, inputFieldValue);
        }

        Map<FieldName, ?> results = evaluator.evaluate(arguments);

        List<TargetField> targetFields = evaluator.getTargetFields();
        TargetField targetField = targetFields.get(0);
        FieldName targetFieldName = targetField.getName();
        ProbabilityDistribution target = (ProbabilityDistribution) results.get(targetFieldName);
        System.out.println(a + " " + b + " " + c + " " + d + ":" + target);
        return target;
    }
    public static void main(String args[]){
        PMMLDemo2 demo = new PMMLDemo2();
        Evaluator model = demo.loadPmml();
        demo.predict(model,2,5,6,8);
        demo.predict(model,7,9,3,6);
        demo.predict(model,1,2,3,1);
        demo.predict(model,2,4,1,5);
    }
}

獲得輸出內容：

2 5 6 8:ProbabilityDistribution{result=1, probability_entries=[1=0.9999949538769296, 0=5.046123070395758E-6]}
7 9 3 6:ProbabilityDistribution{result=0, probability_entries=[1=1.1216598160542013E-9, 0=0.9999999988783402]}
1 2 3 1:ProbabilityDistribution{result=0, probability_entries=[1=2.363331367481431E-8, 0=0.9999999763666864]}
2 4 1 5:ProbabilityDistribution{result=1, probability_entries=[1=0.9999999831203591, 0=1.6879640907241367E-8]}

其中result爲LR最終的結果，機率爲二分類的機率。

參考資料

• 官方文檔：https://openscoring.io/
• JPMML官方文檔:https://github.com/jpmml/jpmml-evaluator
• jpmml-sklearn:https://github.com/jpmml/jpmml-sklearn
• jpmml-sparkml:https://github.com/jpmml/jpmml-sparkml/tree/master
• 用PMML實現機器學習模型的跨平臺上線：http://www.cnblogs.com/pinard/p/9220199.html
• PMML模型文件在機器學習的實踐經驗:https://blog.csdn.net/hopeztm/article/details/78321700