2-Spark高级数据分析-第二章 用Scala和Spark进行数据分析

　　数据清洗时数据科学项目的第一步，往往也是最重要的一步。

　　本章主要做数据统计（总数、最大值、最小值、平均值、标准偏差）和判断记录匹配程度。

Spark编程模型

　　编写Spark程序通常包括一系列相关步骤：
　　　　1. 在输入数据集上定义一组转换。
　　　　2. 调用action，用以将转换后的数据集保存到持久存储上，或者把结果返回到驱动程序的本地内存。
　　　　3. 运行本地计算，本地计算处理分布式计算的结果。本地计算有助于你确定下一步的转换和action。

2.4 小试牛刀：Spark shell和SparkContext

　　加州大学欧文分校机器学习资料库(UC Irvine Machine Learning Repository)，这个资料库为研究和教学提供了大量非常好的数据源，这些数据源非常有意义，并且是免费的。

　　首先，我们从资料库中下载数据（需翻墙，也可以从此处下载(http://pan.baidu.com/s/1c29fBVy)）：

$ mkdir linkage
$ cd linkage/
$ curl -o donation.zip http://bit.ly/1Aoywaq
$ unzip donation.zip
$ unzip 'block_*.zip'

　　我在学习时没有使用Hadoop集群，数据就放到了源码中，主要解压后的数据放到了一个文件夹下。同时，这本书使用的语言是Scala，在学习时，会再写一遍对应的Java版本。

创建RDD：

Scala：

scala> val rawblocks = sc.textFile("D:/Workspace/AnalysisWithSpark/src/main/java/advanced/chapter2/linkage")
rawblocks: org.apache.spark.rdd.RDD[String] = D:/Workspace/AnalysisWithSpark/src/main/java/advanced/chapter2/linkage MapPartitionsRDD[3] at textFile at <console>:27

Java：

SparkConf sc = new SparkConf().setMaster("local").setAppName("Linkage");
System.setProperty("hadoop.home.dir", "D:/Tools/hadoop-2.6.4");
JavaSparkContext jsc = new JavaSparkContext(sc);
JavaRDD<String> lines = jsc.textFile("src/main/java/advanced/chapter2/linkage");

2.5 把数据从集群上获取到客户端

　　使用first向客户端返回RDD的第一个元素：

　　Scala：

scala> rawblocks.first
res2: String = "id_1","id_2","cmp_fname_c1","cmp_fname_c2","cmp_lname_c1","cmp_lname_c2","cmp_sex","cmp_bd","cmp_bm","cmp_by","cmp_plz","is_match"

　　Java：

System.out.println(lines.first());

　　如果RDD只包含少量记录，可以使用collect方法向客户返回一个包含所有RDD内容的数组。
　　take方法可以向客户端返回一个包含指定数量记录的数组。例如：

　　Scala：

scala> val head = rawblocks.take(10)
head: Array[String] = Array("id_1","id_2","cmp_fname_c1","cmp_fname_c2","cmp_lname_c1","cmp_lname_c2","cmp_sex","cmp_bd","cmp_bm","cmp_by","cmp_plz","is_match", 37291,53113,0.833333333333333,?,1,?,1,1,1,1,0,TRUE, 39086,47614,1,?,1,?,1,1,1,1,1,TRUE, 70031,70237,1,?,1,?,1,1,1,1,1,TRUE, 84795,97439,1,?,1,?,1,1,1,1,1,TRUE, 36950,42116,1,?,1,1,1,1,1,1,1,TRUE, 42413,48491,1,?,1,?,1,1,1,1,1,TRUE, 25965,64753,1,?,1,?,1,1,1,1,1,TRUE, 49451,90407,1,?,1,?,1,1,1,1,0,TRUE, 39932,40902,1,?,1,?,1,1,1,1,1,TRUE)

scala> head.length
res4: Int = 10

　　Java：

　　

//RDD前10个元素
List<String> head = lines.take(10);
System.out.println(head);
System.out.println(head.size());

　　可以使用foreach方法并结合println来打印出数组中的每个值，并且每一行打印一个值：

　　Scala：
　　

scala> head.foreach(println)
"id_1","id_2","cmp_fname_c1","cmp_fname_c2","cmp_lname_c1","cmp_lname_c2","cmp_sex","cmp_bd","cmp_bm","cmp_by","cmp_plz","is_match"
37291,53113,0.833333333333333,?,1,?,1,1,1,1,0,TRUE
39086,47614,1,?,1,?,1,1,1,1,1,TRUE
70031,70237,1,?,1,?,1,1,1,1,1,TRUE
84795,97439,1,?,1,?,1,1,1,1,1,TRUE
36950,42116,1,?,1,1,1,1,1,1,1,TRUE
42413,48491,1,?,1,?,1,1,1,1,1,TRUE
25965,64753,1,?,1,?,1,1,1,1,1,TRUE
49451,90407,1,?,1,?,1,1,1,1,0,TRUE
39932,40902,1,?,1,?,1,1,1,1,1,TRUE

　　Java：

//分行打印前10个元素
for (String string : head) {
    System.out.println(string);
}

　　由于CSV文件有一个标题行需要过滤，以免影响后续分析，书上对此部分有详细介绍，我在这里使用下面的方法：

　　Scala：

scala> head.filter(x => !x.contains("id_1")).length
res6: Int = 9

　　或者
　　Scala：

scala> head.filterNot(_.contains("id_1")).length
res9: Int = 9

　　或者
　　Scala：

scala> head.filter(!_.contains("id_1")).length
res8: Int = 9

　　Java：

//输出不包含id_1的前10行
List<String> headWithout = lines.filter(line -> !line.contains("id_1")).take(10);
for (String string : headWithout) {
    System.out.println(string);
}

　　其中Java代码其实是先过滤再取的前10行，其实是在集群进行的处理，如果在客户端进行处理只需要对返回List进行过滤，这个就不写了。

　　Java中似乎没有filterNot。

2.6 把代码从客户端发送到集群

　　Scala：

scala> val noheader = rawblocks.filter(!_.contains("id_1"))
noheader: org.apache.spark.rdd.RDD[String] = MapPartitionsRDD[5] at filter at <console>:29

scala> noheader.first
res11: String = 37291,53113,0.833333333333333,?,1,?,1,1,1,1,0,TRUE

　　Java：

JavaRDD<String> noheader = lines.filter(line -> !line.contains("id_1"));
System.out.println(noheader.first());

2.7 用元组和case class对数据进行结构化

　　使用了Scala的隐式类型转换。

　　Scala：

scala> val line = head(5)
line: String = 36950,42116,1,?,1,1,1,1,1,1,1,TRUE

scala> val pieces = line.split(',')
pieces: Array[String] = Array(36950, 42116, 1, ?, 1, 1, 1, 1, 1, 1, 1, TRUE)

scala> val id1 = pieces(0).toInt
id1: Int = 36950

scala> val id2 = pieces(1).toInt
id2: Int = 42116

scala> val matched = pieces(11).toBoolean
matched: Boolean = true

scala> val rawscores = pieces.slice(2, 11)
rawscores: Array[String] = Array(1, ?, 1, 1, 1, 1, 1, 1, 1)

scala> rawscores.map(s => s.toDouble)
java.lang.NumberFormatException: For input string: "?"
at sun.misc.FloatingDecimal.readJavaFormatString(Unknown Source)
at sun.misc.FloatingDecimal.parseDouble(Unknown Source)

　　遇到错误，此时需要将rawscores 中的"?"在转换成double类型时转换为Double.NaN:

　　Scala：

scala> def toDouble(s: String) = {
if ("?".equals(s)) Double.NaN else s.toDouble
}
toDouble: (s: String)Double

scala> val scores = rawscores.map(toDouble)
scores: Array[Double] = Array(1.0, NaN, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)

　　Java:

String line = head.get(5);
String pieces[] = line.split(",");
Integer id1 = Integer.parseInt(pieces[0]);
Double scores[] = new Double[pieces.length-3];
Integer id2 = Integer.parseInt(pieces[1]);
for (int i = 2,j = 0; i < pieces.length-1; i++, j++) {
    String piece = pieces[i];
    if(piece.equals("?")){
        scores[j] = Double.NaN;
    } else {
        scores[j] = Double.parseDouble(pieces[i]);
    }
}
Boolean matched = Boolean.parseBoolean(pieces[1]);
MatchData matchData = new MatchData();
matchData.setId1(id1);
matchData.setId2(id2);
matchData.setMatched(matched);
matchData.setMatched(matched);
matchData.toString();

　　将上面的代码合并成一个函数：
　　Scala：

scala> val scores = rawscores.map(toDouble)
scores: Array[Double] = Array(1.0, NaN, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)

scala> def parse(line: String) = {
val pieces = line.split(',')
val id1 = pieces(0).toInt
val id2 = pieces(1).toInt
val scores = pieces.slice(2, 11).map(toDouble)
val matched = pieces(11).toBoolean
(id1, id2, scores, matched)
}
parse: (line: String)(Int, Int, Array[Double], Boolean)

scala> val tup = parse(line)
tup: (Int, Int, Array[Double], Boolean) = (36950,42116,Array(1.0, NaN, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0),true)

　　获取元组单个字段的值可以通过下标函数，从_1开始，或者用productElement方法，它是从0开始计数的。可用productArity方法获得元组大小。

　　通过定义一个case class来通过有意义的名称来访问记录元素：
　　Scala：

scala> case class MatchData(id1: Int, id2: Int,
scores: Array[Double], matched: Boolean)
defined class MatchData

scala> def parse(line: String) = {
val pieces = line.split(',')
val id1 = pieces(0).toInt
val id2 = pieces(1).toInt
val scores = pieces.slice(2, 11).map(toDouble)
val matched = pieces(11).toBoolean
MatchData(id1, id2, scores, matched)
}
parse: (line: String)MatchData

scala> val md = parse(line)
md: MatchData = MatchData(36950,42116,[D@3be5eccd,true)

scala> md.matched
res5: Boolean = true

scala> md.id1
res6: Int = 36950

　　Java:

private static MatchData parseLine(String line) {
    String pieces[] = line.split(",");
    Integer id1 = Integer.parseInt(pieces[0]);
    Double[] scores = new Double[pieces.length-3];
    Integer id2 = Integer.parseInt(pieces[1]);
    for (int i = 2,j = 0; i < pieces.length-1; i++, j++) {
        String piece = pieces[i];
        if(piece.equals("?")){
            scores[j] = Double.NaN;
        } else {
            scores[j] = Double.parseDouble(pieces[i]);
        }
    }
    Boolean matched = Boolean.parseBoolean(pieces[11]);
    MatchData matchData = new MatchData();
    matchData.setId1(id1);
    matchData.setId2(id2);
    matchData.setScores(scores);
    matchData.setMatched(matched);
    matchData.toString();
    return matchData;
}

　　其中MatchData定义如下：
　　Java:

import java.io.Serializable;
import java.util.Arrays;

public class MatchData implements Serializable{
    private static final long serialVersionUID = 1L;
    private Integer id1;
    private Integer id2;
    private Double[] scores; 
    private Boolean matched;
    public Integer getId1() {
        return id1;
    }
    public void setId1(Integer id1) {
        this.id1 = id1;
    }
    public Integer getId2() {
        return id2;
    }
    public void setId2(Integer id2) {
        this.id2 = id2;
    }
    public Double[] getScores() {
        return scores;
    }
    public void setScores(Double[] scores) {
        this.scores = scores;
    }
    public Boolean getMatched() {
        return matched;
    }
    public void setMatched(Boolean matched) {
        this.matched = matched;
    }
    
    @Override
    public String toString() {
        return "MatchData [id1=" + id1 + ", id2=" + id2 + ", scores=" + Arrays.toString(scores) + ", matched=" + matched
                + "]";
    }
}

　　现在单条记录测试完成了，我们将其用在head数组的除了标题的所有元素上：
　　Scala：

scala> val mds = head.filter(!_.contains("id_1")).map(x => parse(x))
mds: Array[MatchData] = Array(MatchData(37291,53113,[D@5b555e90,true), MatchData(39086,47614,[D@4305d2ec,true), MatchData(70031,70237,[D@cb06460,true), MatchData(84795,97439,[D@1cfa2146,true), MatchData(36950,42116,[D@6daa5d97,true), MatchData(42413,48491,[D@55607c,true), MatchData(25965,64753,[D@fc98772,true), MatchData(49451,90407,[D@42f477dc,true), MatchData(39932,40902,[D@7d0ddcf0,true))

　　将解析函数用于集群数据：
　　Scala：

scala> val parsed = noheader.map(line => parse(line))
parsed: org.apache.spark.rdd.RDD[MatchData] = MapPartitionsRDD[3] at map at <console>:53

　　Java:

JavaRDD<MatchData> parsed = noheader.map(line -> parseLine(line));

　　缓存RDD：
　　Scala：

scala> parsed.cache()
res8: parsed.type = MapPartitionsRDD[3] at map at <console>:53

　　Java:

parsed.cache();

　　Spark为持久化RDD定义了几种不同的机制，用不同的StorageLevel值表示。
　　Rdd.cache()是rdd.persist(StorageLevel.MEMORY)的简写，他将RDD存储为未序列化的Java对象。占用很大的内存。内存空间不够时就不存储，使用时需要重新计算。
　　MEMORY_SER(Spark 1.6之后为MEMORY_ONLY_SER)存放序列化后的内容。使用得当时，序列化数据占用的空间比未序列化数据少两到五倍。
　　MEMORY_AND_DISK和MEMORY_AND_DISK_SER类似于MEMORY和MEMORY_SER，如果分区在内存中放不下，Spark会将其溢写到磁盘上。

2.8 聚合

　　对集群数据进行聚合时，一定要时刻记住我们分析的数据是存放在多台机器上的，并且聚合需要通过连接机器的网络来移动数据。跨网络移动数据需要许多计算资源，包括确定每条记录要传到哪些服务器、数据序列化、数据压缩，通过网络发送数据、解压缩，接着序列化结果，最后在聚合后的数据上执行运算。为了提高速度，我们需要尽可能少的移动数据。在聚合前能过滤掉的数据越多，就能越快得到问题的答案。

2.9 创建直方图

　　使用RDD的countByValue创建直方图，该动作对计数类运算效率非常高，它向客户端返回Map[T,Long]类型的结果。安照MatchData的matched字段：

　　Scala：

scala> val matchCounts = parsed.map(md => md.matched).countByValue()
matchCounts: scala.collection.Map[Boolean,Long] = Map(true -> 20931, false -> 5728201)

scala> val matchCountsSeq = matchCounts.toSeq
matchCountsSeq: Seq[(Boolean, Long)] = ArrayBuffer((true,20931), (false,5728201))

scala> matchCountsSeq.sortBy(_._1).foreach(println)
(false,5728201)
(true,20931)

scala> matchCountsSeq.sortBy(_._2).foreach(println)
(true,20931)
(false,5728201)

　　Java的排序需要使用TreeMap和Comparator，不写了。

2.10 连续变量的概要统计

　　查看scores的值：
　　Scala：

scala> parsed.map(md => md.scores(0)).stats()
res2: org.apache.spark.util.StatCounter = (count: 5749132, mean: NaN, stdev: NaN, max: NaN, min: NaN)

scala> import java.lang.Double.isNaN
import java.lang.Double.isNaN

scala> parsed.map(md => md.scores(0)).filter(!isNaN(_)).stats()
res3: org.apache.spark.util.StatCounter = (count: 5748125, mean: 0.712902, stdev: 0.388758, max: 1.000000, min: 0.000000)

scala> val stats = (0 until 9).map(i => {
parsed.map(md => md.scores(i)).filter(!isNaN(_)).stats()
})
stats: scala.collection.immutable.IndexedSeq[org.apache.spark.util.StatCounter] = Vector((count: 5748125, mean: 0.712902, stdev: 0.388758, max: 1.000000, min: 0.000000), (count: 103698, mean: 0.900018, stdev: 0.271316, max: 1.000000, min: 0.000000), (count: 5749132, mean: 0.315628, stdev: 0.334234, max: 1.000000, min: 0.000000), (count: 2464, mean: 0.318413, stdev: 0.368492, max: 1.000000, min: 0.000000), (count: 5749132, mean: 0.955001, stdev: 0.207301, max: 1.000000, min: 0.000000), (count: 5748337, mean: 0.224465, stdev: 0.417230, max: 1.000000, min: 0.000000), (count: 5748337, mean: 0.488855, stdev: 0.499876, max: 1.000000, min: 0.000000), (count: 5748337, mean: 0.222749, stdev: 0.416091, max: 1.000000, min: 0.000000), (count: 5736289, mean: 0.005529, stdev: 0.074149, max: 1.000000,...

　　Java：

for (int i = 0; i < 9; i++) {
    final Integer innerI = new Integer(i);
    StatCounter statCounter = parsed.mapToDouble(md -> md.getScores()[innerI]).filter(score -> !Double.isNaN(score)).stats();
    System.out.println(statCounter);
}

2.11 为计算概要信息创建可重用代码

　　上面的方法能够完成工作，但是为了得到所有统计信息必须重复处理parsed RDD的所有记录9次，是每列分别统计一次。现在需要将其改进为一次处理所有记录。先将所有行聚合，然后再处理。
　　Scala的代码参考书上或者书中附带的源码(https://github.com/sryza/aas/tree/master/ch02-intro/src/main/scala/com/cloudera/datascience/intro)：

　　Scala：

/*
 * Copyright 2015 Sanford Ryza, Uri Laserson, Sean Owen and Joshua Wills
 *
 * See LICENSE file for further information.
 */

package com.cloudera.datascience.intro

import org.apache.spark.{SparkConf, SparkContext}
import org.apache.spark.SparkContext._
import org.apache.spark.rdd.RDD
import org.apache.spark.util.StatCounter

case class MatchData(id1: Int, id2: Int,
  scores: Array[Double], matched: Boolean)
case class Scored(md: MatchData, score: Double)

object RunIntro extends Serializable {
  def main(args: Array[String]): Unit = {
    val sc = new SparkContext(new SparkConf().setAppName("Intro"))
   
    val rawblocks = sc.textFile("hdfs:///user/ds/linkage")
    def isHeader(line: String) = line.contains("id_1")
    
    val noheader = rawblocks.filter(x => !isHeader(x))
    def toDouble(s: String) = {
     if ("?".equals(s)) Double.NaN else s.toDouble
    }

    def parse(line: String) = {
      val pieces = line.split(',')
      val id1 = pieces(0).toInt
      val id2 = pieces(1).toInt
      val scores = pieces.slice(2, 11).map(toDouble)
      val matched = pieces(11).toBoolean
      MatchData(id1, id2, scores, matched)
    }

    val parsed = noheader.map(line => parse(line))
    parsed.cache()

    val matchCounts = parsed.map(md => md.matched).countByValue()
    val matchCountsSeq = matchCounts.toSeq
    matchCountsSeq.sortBy(_._2).reverse.foreach(println)

    val stats = (0 until 9).map(i => {
      parsed.map(_.scores(i)).filter(!_.isNaN).stats()
    })
    stats.foreach(println)

    val nasRDD = parsed.map(md => {
      md.scores.map(d => NAStatCounter(d))
    })
    val reduced = nasRDD.reduce((n1, n2) => {
      n1.zip(n2).map { case (a, b) => a.merge(b) }
    })
    reduced.foreach(println)

    val statsm = statsWithMissing(parsed.filter(_.matched).map(_.scores))
    val statsn = statsWithMissing(parsed.filter(!_.matched).map(_.scores))
    statsm.zip(statsn).map { case(m, n) =>
      (m.missing + n.missing, m.stats.mean - n.stats.mean)
    }.foreach(println)

    def naz(d: Double) = if (Double.NaN.equals(d)) 0.0 else d
    val ct = parsed.map(md => {
      val score = Array(2, 5, 6, 7, 8).map(i => naz(md.scores(i))).sum
      Scored(md, score)
    })

    ct.filter(s => s.score >= 4.0).
      map(s => s.md.matched).countByValue().foreach(println)
    ct.filter(s => s.score >= 2.0).
      map(s => s.md.matched).countByValue().foreach(println)
  }

  def statsWithMissing(rdd: RDD[Array[Double]]): Array[NAStatCounter] = {
    val nastats = rdd.mapPartitions((iter: Iterator[Array[Double]]) => {
      val nas: Array[NAStatCounter] = iter.next().map(d => NAStatCounter(d))
      iter.foreach(arr => {
        nas.zip(arr).foreach { case (n, d) => n.add(d) }
      })
      Iterator(nas)
    })
    nastats.reduce((n1, n2) => {
      n1.zip(n2).map { case (a, b) => a.merge(b) }
    })
  }
}

class NAStatCounter extends Serializable {
  val stats: StatCounter = new StatCounter()
  var missing: Long = 0

  def add(x: Double): NAStatCounter = {
    if (x.isNaN) {
      missing += 1
    } else {
      stats.merge(x)
    }
    this
  }

  def merge(other: NAStatCounter): NAStatCounter = {
    stats.merge(other.stats)
    missing += other.missing
    this
  }

  override def toString: String = {
    "stats: " + stats.toString + " NaN: " + missing
  }
}

object NAStatCounter extends Serializable {
  def apply(x: Double) = new NAStatCounter().add(x)
}

　　Java测试相关代码
　　Java:

public static void main(String[] args) {
    //测试NAStatCounter类
    NAStatCounter nas1 = new NAStatCounter(10.0);
    nas1.add(2.1);
    NAStatCounter nas2 = new NAStatCounter(Double.NaN);
    nas1.merge(nas2);
    System.out.println("nas1:");
    System.out.println(nas1.toString());
    
    //测试List统计
    NAStatCounter nas3 = new NAStatCounter();
    Arrays.asList(new Double[]{1.0, Double.NaN, 17.29}).forEach(d -> nas3.add(d));
    System.out.println("nas3:");
    System.out.println(nas3);
    NAStatCounter nas4 = new NAStatCounter();
    Arrays.asList(new Double[]{Double.NaN, 15.39, 2.0}).forEach(d -> nas4.add(d));
    System.out.println("nas4:");
    System.out.println(nas4);
    
    //测试聚合
    List<NAStatCounter> list1 = Arrays.asList(new Double[]{1.0, Double.NaN, 17.29})
                                     .stream()
                                     .map(d -> new NAStatCounter(d))
                                     .collect(Collectors.toList());
    System.out.println("list1:");
    for (NAStatCounter naStatCounter : list1) {
        System.out.println(naStatCounter);
    }
    List<NAStatCounter> list2 = Arrays.asList(new Double[]{Double.NaN, 15.39, 2.0})
            .stream()
            .map(d -> new NAStatCounter(d))
            .collect(Collectors.toList());
    System.out.println("list2:");
    for (NAStatCounter naStatCounter : list2) {
        System.out.println(naStatCounter);
    }
    List<Tuple2<NAStatCounter, NAStatCounter>> n0 = new ArrayList<Tuple2<NAStatCounter, NAStatCounter>>();
    for (int i = 0; i < list1.size(); i++) {
        n0.add(new Tuple2<NAStatCounter, NAStatCounter>(list1.get(i), list2.get(i)));
    }
    List<NAStatCounter> merged = n0.stream().map(p -> p._1.merge(p._2)).collect(Collectors.toList());
    System.out.println("merged:");
    merged.forEach(System.out::println);
}

　　Java最终代码：

　　Java：

public static List<NAStatCounter> ststsWithMissing(JavaRDD<MatchData> parsed){
    JavaRDD<List<NAStatCounter>> nasRDD = parsed.map(md -> {
        return Arrays.asList(md.getScores()).stream().map(d -> new NAStatCounter(d)).collect(Collectors.toList());
    });
    List<NAStatCounter> reduced = nasRDD.reduce((List<NAStatCounter> n1, List<NAStatCounter>n2) -> {
        List<Tuple2<NAStatCounter, NAStatCounter>> n0 = new ArrayList<Tuple2<NAStatCounter, NAStatCounter>>();
        for (int i = 0; i < n1.size(); i++) {
            n0.add(new Tuple2<NAStatCounter, NAStatCounter>(n1.get(i), n2.get(i)));
        }
        return n0.stream().map(p -> p._1.merge(p._2)).collect(Collectors.toList());
    });
    reduced.forEach(System.out::println);
    return reduced;
}

　　没有找到Java中可以代替Scala zip函数的好方式，先自己拼了一个，如果好的方式，还请大家不吝赐教。

2.12 变量的选择和评分简介

　　有了ststsWithMissing函数就可以分享parsed RDD中匹配和不匹配记录的匹配分值数组的分布差异了。

　　Scala：

val statsm = statsWithMissing(parsed.filter(_.matched).map(_.scores))
val statsn = statsWithMissing(parsed.filter(!_.matched).map(_.scores))
statsm.zip(statsn).map { case(m, n) =>
(m.missing + n.missing, m.stats.mean - n.stats.mean)
}.foreach(println)

　　Java：

List<NAStatCounter> ststsm = ststsWithMissing(parsed.filter(md -> md.getMatched()));
List<NAStatCounter> ststsn = ststsWithMissing(parsed.filter(md -> !md.getMatched()));

List<Tuple2<NAStatCounter, NAStatCounter>> ststs = new ArrayList<Tuple2<NAStatCounter, NAStatCounter>>();
for (int i = 0; i < ststsm.size(); i++) {
    ststs.add(new Tuple2<NAStatCounter, NAStatCounter>(ststsm.get(i), ststsn.get(i)));
}
System.out.println("评分：");
ststs.stream().forEach(p -> System.out.println((p._1.missing+p._2.missing)+","+(p._1.stats.mean()-p._2.stats.mean())));

　　结果：

1007,0.285452905746686
5645434,0.09104268062279897
0,0.6838772482597568
5746668,0.8064147192926266
0,0.03240818525033473
795,0.7754423117834044
795,0.5109496938298719
795,0.7762059675300523
12843,0.9563812499852176

　　评分模型：
　　Scala：

def naz(d: Double) = if (Double.NaN.equals(d)) 0.0 else d
case class Scored(md: MatchData, score: Double)
val ct = parsed.map(md => {
val score = Array(2, 5, 6, 7, 8).map(i => naz(md.scores(i))).sum
Scored(md, score)
})
ct.filter(s => s.score >= 4.0).map(s => s.md.matched).countByValue()
ct.filter(s => s.score >= 2.0).map(s => s.md.matched).countByValue()

　　Java：
　　计算ct:

public static JavaRDD<Scored> getCT(JavaRDD<MatchData> parsed){
    JavaRDD<Scored> ct = parsed.map(md -> {
        List<Double> scoreList = Arrays.asList(new Integer[]{2, 5, 6, 7, 8}).stream().map(i -> {
            if(md.getScores()[i].equals(Double.NaN)){
                return 0.0;
            } else {
                return md.getScores()[i];
            }
        }).collect(Collectors.toList());
        Scored scored = new Scored();
        scored.score = scoreList.stream().reduce((r, e) -> r = r + e ).get();
        scored.md = md;
        return scored;
    });
    return ct;
}

　　调用ct:

JavaRDD<Scored> ct = getCT(parsed);

Map<Boolean, Long> ct0 = ct.filter(s -> s.score >= 4.0).map(s -> s.md.getMatched()).countByValue();
for (Entry<Boolean, Long> ctTmp : ct0.entrySet()) {
    System.out.println("Key:" + ctTmp.getKey());
    System.out.println("Value:" + ctTmp.getValue());
}
Map<Boolean, Long> ct1 = ct.filter(s -> s.score >= 2.0).map(s -> s.md.getMatched()).countByValue();
for (Entry<Boolean, Long> ctTmp : ct1.entrySet()) {
    System.out.println("Key:" + ctTmp.getKey());
    System.out.println("Value:" + ctTmp.getValue());
}

　　结果：

Key:true
Value:20871
Key:false
Value:637


Key:true
Value:20931
Key:false
Value:596414