基于Spark实现随机森林代码

public class RandomForestClassficationTest extends TestCase implements Serializable
{
 
  /**
  * 
  */
  private static final long serialVersionUID = 7802523720751354318L;
  
  class PredictResult implements Serializable{
    /**
    * 
    */
    private static final long serialVersionUID = -168308887976477219L;
    double label;
    double prediction;
    
    public PredictResult(double label,double prediction){
    this.label = label;
    this.prediction = prediction;
    }
    
    @Override
    public String toString(){
      return this.label + " : " + this.prediction ;
    }
  }
  
  
  public void test_randomForest() throws JAXBException{
  
    SparkConf sparkConf = new SparkConf();
    sparkConf.setAppName("RandomForest");
    sparkConf.setMaster("local");
    
    SparkContext sc = new SparkContext(sparkConf);
    String dataPath = RandomForestClassficationTest.class.getResource("/").getPath() + "/sample_libsvm_data.txt";
    
    RDD dataSet = MLUtils.loadLibSVMFile(sc, dataPath);
    RDD[] rddList = dataSet.randomSplit(new double[]{0.7,0.3},1);
    
    RDD trainingData = rddList[0];
    RDD testData = rddList[1];
    
    ClassTag labelPointClassTag = trainingData.elementClassTag();
    
    JavaRDD trainingJavaData = new JavaRDD(trainingData,labelPointClassTag);
    
    int numClasses = 2;
    Map categoricalFeatureInfos = new HashMap();
    int numTrees = 3;
    String featureSubsetStrategy = "auto";
    String impurity = "gini";
    int maxDepth = 4;
    int maxBins = 32;
    
    /**
    * 1 numClasses分类个数为2
    * 2 numTrees 表示的是随机森林中树的个数
    * 3 featureSubsetStrategy
    * 4 
    */
    final RandomForestModel model = RandomForest.trainClassifier(trainingJavaData,
    numClasses,
    categoricalFeatureInfos,
    numTrees,
    featureSubsetStrategy,
    impurity,
    maxDepth,
    maxBins,
    1);
 
    JavaRDD testJavaData = new JavaRDD(testData,testData.elementClassTag());
    
    JavaRDD predictRddResult = testJavaData.map(new Function(){
 
 
    /**
    * 
    */
    private static final long serialVersionUID = 1L;
    
    
    public PredictResult call(LabeledPoint point) throws Exception {
      // TODO Auto-generated method stub
      double pointLabel = point.label();
      double prediction = model.predict(point.features());
      PredictResult result = new PredictResult(pointLabel,prediction);
      return result;
    }
    
    });
 
    List predictResultList = predictRddResult.collect();
    for(PredictResult result:predictResultList){
      System.out.println(result.toString());
    }
    
      System.out.println(model.toDebugString());
    }
}

TreeEnsembleModel classifier with 3 trees
 
 
Tree 0:
If (feature 435 <= 0.0)
If (feature 516 <= 0.0)
Predict: 0.0
Else (feature 516 > 0.0)
Predict: 1.0
Else (feature 435 > 0.0)
Predict: 1.0
Tree 1:
If (feature 512 <= 0.0)
Predict: 1.0
Else (feature 512 > 0.0)
Predict: 0.0
Tree 2:
If (feature 377 <= 1.0)
Predict: 0.0
Else (feature 377 > 1.0)
If (feature 455 <= 0.0)
Predict: 1.0
Else (feature 455 > 0.0)
Predict: 0.0