sparkmllib学习之线性回归

import org.apache.log4j.{Level, Logger}
import org.apache.spark.mllib.linalg.Vectors
import org.apache.spark.mllib.regression.{LabeledPoint, LinearRegressionWithSGD}
import org.apache.spark.mllib.tree.DecisionTree
import org.apache.spark.rdd.RDD
import org.apache.spark.{SparkConf, SparkContext}

/**
  * Created by Damon on 17-5-22.
  */
object Regression {
  def main(args: Array[String]): Unit = {
    Logger.getLogger("org.apache.spark").setLevel(Level.WARN)
    Logger.getLogger("org.eclipse.jetty.server").setLevel(Level.OFF)
    val conf =new SparkConf().setAppName("regression").setMaster("local[4]")
    val sc =new SparkContext(conf)
    //文件名
    val file_bike="hour_nohead.csv"
    //调用二元向量化方法
    val labeled_file=labeledFile(file_bike,sc)
    /*/*对目标值取对数*/
    val labeled_file1=labeled_file.map(point => LabeledPoint(math.log(point.label),point.features))
    */
    //构建线性回归模型，注该方法在：spark2.1.0已经抛弃了。。。。
    val model_liner=LinearRegressionWithSGD.train(labeled_file,10,0.1)
    //val categoricalFeaturesInfo = Map[Int,Int]()
    //val model_DT=DecisionTree.trainRegressor(labeled_file,categoricalFeaturesInfo,"variance",5,32)
    val predict_vs_train=labeled_file.map{
      point => (model_liner.predict(point.features),point.label)
//对目标取对数后的，预测方法
     /* point => (math.exp(model_liner.predict(point.features)),math.exp(point.label))*/
    }
    predict_vs_train.take(5).foreach(println(_))
   /*
(135.94648455498356,16.0)
(134.38058174607252,40.0)
(134.1840793861374,32.0)
(133.88699144084515,13.0)
(133.77899037657548,1.0)
   */
  def labeledFile(originFile:String,sc:SparkContext):RDD[LabeledPoint]={
    val file_load=sc.textFile(originFile)
    val file_split=file_load.map(_.split(","))
    /*构建映射类函数的方法:mapping*/
    def mapping(rdd:RDD[Array[String]],index:Int)=
      rdd.map(x => x(index)).distinct.zipWithIndex().collect.toMap
    /*存储每列映射方法mapping的maps集合*/
    var maps:Map[Int,Map[String,Long]] = Map()
    /* 生成maps*/
    for(i <- 2 until 10)
      maps += (i -> mapping(file_split,i))
    /*max_size表示每列的特征之和*/
    val max_size=maps.map(x =>x._2.size).sum
    val file_label=file_split.map{
      x =>
        var num:Int=0
        var size:Int=0
        /*构建长度为max_size+4的特征数组,初始值全为0*/
        val arrayOfDim=Array.ofDim[Double](max_size+4)
        for(j<-2 until 10) {
          num = maps(j)(x(j)).toInt
          if(j==2) size=0 else size += maps(j-1).size
          /*为特征赋值*/
          arrayOfDim(size+num)=1.0
        }
        /*添加后面4列归一化的特征*/
        for(j<-10 until 14)
        arrayOfDim(max_size+(j-10))=x(j).toDouble
        /*生成LabeledPoint类型*/
        LabeledPoint(x(14).toDouble+x(15).toDouble,Vectors.dense(arrayOfDim))
    }
    file_label
  }
}

/*MSE是均方误差*/
    val mse=predict_vs_train.map(x => math.pow(x._1-x._2,2)).mean()
   /* 平均绝对误差（MAE）*/
    val mae=predict_vs_train.map(x => math.abs(x._1-x._2)).mean()
    /*均方根对数误差（RMSLE）*/
    val rmsle=math.sqrt(predict_vs_train.map(x => math.pow(math.log(x._1+1)-math.log(x._2+1),2)).mean())
    println(s"mse is $mse and mae is $mae and rmsle is $rmsle")
/*
mse is 29897.34020145107 and mae is 130.53255991178477 and rmsle is 1.4803867063174845
*/

//划分训练集和测试集
    val labeled_orign = labeled_file.randomSplit(Array(0.8, 0.2), 11L)
    val train_file = labeled_orign(0)
    val test_file = labeled_orign(1)
    /*调节迭代次数*/
    val Iter_Results = Seq(1, 5, 10, 20, 50, 100).map { param =>
      val model = LinearRegressionWithSGD.train(test_file, param, 0.01)
      val scoreAndLabels = test_file.map { point =>
        (model.predict(point.features), point.label)
      }
      val rmsle = math.sqrt(scoreAndLabels.map(x => math.pow(math.log(x._1) - math.log(x._2), 2)).mean)
      (s"$param lambda", rmsle)
    }
    /*迭代次数的结果输出*/
    Iter_Results.foreach { case (param, rmsl) => println(f"$param, rmsle = ${rmsl}")}
    /*调节步长数的大小*/
    val Step_Results = Seq(0.01, 0.025, 0.05, 0.1, 1.0).map { param =>
        val model = LinearRegressionWithSGD.train(test_file, 20, param)
        val scoreAndLabels = test_file.map { point =>
          (model.predict(point.features), point.label)
        }
        val rmsle = math.sqrt(scoreAndLabels.map(x => math.pow(math.log(x._1) - math.log(x._2), 2)).mean)
        (s"$param lambda", rmsle)
      }
    /*步长的结果输出*/
    Step_Results.foreach { case (param, rmsl) => println(f"$param, rmsle = ${rmsl}")}
/*results
1 lambda, rmsle = 2.9033629718241167
5 lambda, rmsle = 2.0102924520366092
10 lambda, rmsle = 1.7548482896314488
20 lambda, rmsle = 1.5785106813100764
50 lambda, rmsle = 1.461748782192306
100 lambda, rmsle = 1.4462810196387068
步长
0.01 lambda, rmsle = 1.5785106813100764
0.025 lambda, rmsle = 1.4478358250917658
0.05 lambda, rmsle = 1.5152549319928832
0.1 lambda, rmsle = 1.5687431700715837
1.0 lambda, rmsle = NaN
*/