spark LinearRegression 预测缺失字段的值

最近在做金融科技建模的时候，字段里面很多缺少值得时候，模型对于新用户的预测会出现很大的不稳定，即PSI较大的情况。

虽然我们依据字段IV值得大小不断的在调整字段且开发新变量，但是很多IV值很大的字段直接用平均值、或者0代替显然不够合理。

所以，我们在尝试把字段缺失值当作需要预测的值，把该字段不缺失的当作ｙ，用其他字段当作Ｘ，去预测该字段缺失值得值。不同于机器学习的回归和分类预测。

这里的预测结果是一个具体的值，它的范围从负无穷到正无穷都有可能。

　　数据直接读存于Ｈｉｖｅ，代码如下：

import org.apache.spark.sql.{DataFrame, Row, SQLContext, SaveMode}
import org.apache.spark.{SparkConf, SparkContext}
import org.apache.spark.mllib.linalg.{Vector, Vectors}
import org.apache.spark.ml.regression.LinearRegression
import org.apache.spark.mllib.regression.LabeledPoint
import  org.apache.spark.ml.regression.LinearRegressionModel
import org.apache.spark.sql.hive.HiveContext
import org.apache.spark.sql.types.{DoubleType, StringType, StructField, StructType}
import scala.collection.mutable.ArrayBuffer
//  select   corr(cast(p.cnt_addbook_one as double),cast(l.cnt_addbook_one as double))as corrs  from   lkl_card_score.predictcnt_addbook_one20180201  p join lkl_card_score.fieldValuePredictModel3
//l on p.order_id=l.order_src  where l.cnt_addbook_one<>0
//
object predictcnt_addbook_one20180201 {
  def main(args: Array[String]): Unit = {
    val cf = new SparkConf().setAppName("ass").setMaster("local")
    val sc = new SparkContext(cf)
    val sqlContext = new SQLContext(sc)
    val hc = new HiveContext(sc)
    import sqlContext.implicits._

     val data = hc.sql(s"select * from lkl_card_score.fieldValuePredictModel3 where cnt_addbook_one<>0   and cnt_addbook_one%2=1").map {
          row =>
            val arr = new ArrayBuffer[Double]()
            //剔除label、phone字段
            for (i <- 4 until row.size) {
              if (row.isNullAt(i)) {
                arr += 0.0
              }
              else if (row.get(i).isInstanceOf[Int])
                arr += row.getInt(i).toDouble
              else if (row.get(i).isInstanceOf[Double])
                arr += row.getDouble(i)
              else if (row.get(i).isInstanceOf[Long])
                arr += row.getLong(i).toDouble
              else if (row.get(i).isInstanceOf[String])
                arr += 0.0
            }
            LabeledPoint(row.getLong(0).toDouble,Vectors.dense(arr.toArray))
        }.toDF("Murder","features")

    // 建立模型，预测谋杀率Murder
    // 设置线性回归参数

      val lr1 = new LinearRegression()
     val lr2 = lr1.setFeaturesCol("features").setLabelCol("Murder").setFitIntercept(true)
    // RegParam：正则化
    val lr3 = lr2.setMaxIter(50).setRegParam(0.3).setElasticNetParam(0.8)
    // 将训练集合代入模型进行训练

      val lr = lr3
    val lrModel = lr.fit(data)
    // 输出模型全部参数
    lrModel.extractParamMap()
    println(s"Coefficients: ${lrModel.coefficients} Intercept: ${lrModel.intercept}")
    lrModel.write.overwrite().save(s"hdfs://ns1/user/songchunlin/model/predictcnt_addbook_one20180202")
    // 模型进行评价
    val trainingSummary = lrModel.summary
    println(s"numIterations: ${trainingSummary.totalIterations}")
    println(s"objectiveHistory: ${trainingSummary.objectiveHistory.toList}")
    trainingSummary.residuals.show()
    println(s"RMSE: ${trainingSummary.rootMeanSquaredError}")
    println(s"r2: ${trainingSummary.r2}")
    val predict = hc.sql(s"select * from lkl_card_score.fieldValuePredictModel3 where cnt_addbook_one<>0   and cnt_addbook_one%2=0").map {
      row =>
        val arr = new ArrayBuffer[Double]()
        //剔除label、phone字段
        for (i <- 4 until row.size) {
          if (row.isNullAt(i)) {
            arr += 0.0
          }
          else if (row.get(i).isInstanceOf[Int])
            arr += row.getInt(i).toDouble
          else if (row.get(i).isInstanceOf[Double])
            arr += row.getDouble(i)
          else if (row.get(i).isInstanceOf[Long])
            arr += row.getLong(i).toDouble
          else if (row.get(i).isInstanceOf[String])
            arr += 0.0
        }
        (row.getString(2),Vectors.dense(arr.toArray))
    }.toDF("order_src","features")


   val models=LinearRegressionModel.load("hdfs://ns1/user/songchunlin/model/predictcnt_addbook_one20180202")
    val prediction =models.transform(predict)


    //    val predictions = lrModel.transform(vecDF)
    println("输出预测结果")
    val predict_result: DataFrame =prediction.selectExpr("order_src","prediction")
    val pre2=prediction.map(row=>Row(row.get(0).toString,row.get(2).toString))
    val schema = StructType(
      List(
        StructField("order_id", StringType, true),
        StructField("cnt_addbook_one", StringType, true)
      )
    )
    val scoreDataFrame = hc.createDataFrame(pre2,schema)
    scoreDataFrame.count()
    scoreDataFrame.write.mode(SaveMode.Overwrite).saveAsTable("lkl_card_score.predictcnt_addbook_one20180202")

//    predict_result.write.mode(SaveMode.Overwrite).saveAsTable("lkl_card_score.fieldValuePredictModel3_prediction20180131")
//    predict_result.foreach(println(_))
//    sc.stop()



  }
}

   用模型预测未参加训练的数据，计算预测的数据和真实数据相关性为0.99553818714507836，有很大的价值。

select  corr(cast(l.cnt_addbook_one as double),cast(p.cnt_addbook_one as double)) from    lkl_card_score.predictcnt_addbook_one20180202  l
join lkl_card_score.fieldValuePredictModel3 p  on l.order_id=p.order_src
;