Spark购物篮分析:关联规则挖掘

1、浅谈数据挖掘中的关联规则挖掘

2、Hadoop/MapReduce购物篮分析:关联规则挖掘

3、Spark购物篮分析

过程分析:

import org.apache.spark.SparkConf
import org.apache.spark.SparkContext
import scala.collection.mutable.ListBuffer


object FindAssociationRules {

  def main(args: Array[String]): Unit = {
    val sparkConf = new SparkConf().setAppName("market-basket-analysis").setMaster("local")
    val sc = new SparkContext(sparkConf)
    val input = "file:///media/chenjie/0009418200012FF3/ubuntu/mba2.txt"
    val output = "file:///media/chenjie/0009418200012FF3/ubuntu/mba2"
    val transactions = sc.textFile(input)
    /*
    * a,b,c
      a,b,d
      b,c
      b,c
    * */

   /* val tests = transactions.flatMap(line => {
      println("line=" + line)
      val items = line.split(",").toList
      // Converting to List is required because Spark doesn't partition on Array (as returned by split method)
      //(0 to items.size) flatMap items.combinations filter (xs => !xs.isEmpty)
      println("result=" + items.combinations(2).mkString(","))
      val list = ListBuffer.empty[List[String]]
      for(i <- 0 to items.size){
        list.++= (items.combinations(i).toBuffer)
      }
      list.toList.filter(xs => !xs.isEmpty)
    })
    tests.foreach(println)*/

    val patterns = transactions.flatMap(line => {
      val items = line.split(",").toList
      // Converting to List is required because Spark doesn't partition on Array (as returned by split method)
      (0 to items.size) flatMap items.combinations filter (xs => !xs.isEmpty)
      /*
        combinations(n: Int): Iterator[List[A]] 取列表中的n个元素进行组合，返回不重复的组合列表，结果一个迭代器
       */
      /*
      * 上句话等价于：
      * val list = ListBuffer.empty[List[String]]
        for(i <- 0 to items.size){
          list.++= (items.combinations(i).toBuffer)
        }
        list.toList.filter(xs => !xs.isEmpty)
      * 即对a,b,c
      * 先取0个元素进行组合，得到不重复的组合列表[]，加入list中，list为[[]]
      * 再取1个元素进行组合，得到不重复的组合列表[[a],[b],[c]]，加入list中，list为[[a],[b],[c]]
      * 再取2个元素进行组合，得到不重复的组合列表[[a,b],[a,c],[b,c]]，加入list中,list为[[],[a],[b],[c],[a,b],[a,c],[b,c]]
      * 再取3个元素进行组合，得到不重复的组合列表[[a,b,c]]，加入list中,list为[[],[a],[b],[c],[a,b],[a,c],[b,c],[a,b,c]]
      * 然后对其进行过滤，去掉其中为空的列表
      * list为[[a],[b],[c],[a,b],[a,c],[b,c],[a,b,c]]
      * 最后回到外层的flatMap，会将列表的列表拍扁成列表：
      * [a],[b],[c],[a,b],[a,c],[b,c],[a,b,c]
      * */
    }).map((_, 1))
    //到最外面的map，将列表映射为(列表，1)的键值对
    /*
    * (List(a),1)
      (List(b),1)
      (List(c),1)
      (List(a, b),1)
      (List(a, c),1)
      (List(b, c),1)
      (List(a, b, c),1)
      (List(a),1)
      (List(b),1)
      (List(d),1)
      (List(a, b),1)
      (List(a, d),1)
      (List(b, d),1)
      (List(a, b, d),1)
      (List(b),1)
      (List(c),1)
      (List(b, c),1)
      (List(b),1)
      (List(c),1)
      (List(b, c),1)
    * */

    val combined = patterns.reduceByKey(_ + _)//合并key值相同的键值对
    /*
    * (List(a, b, c),1)
      (List(b),4)
      (List(a, b, d),1)
      (List(b, d),1)
      (List(a, b),2)
      (List(a),2)
      (List(a, d),1)
      (List(b, c),3)
      (List(a, c),1)
      (List(c),3)
      (List(d),1)
    *
    * */

    /*下面开始生成子模式
    给定一个频繁模式：(K=List<A1,A2,...,An>,V=Frequency)
    创建如下的子模式(K2,V2)
    (K2=K=List<A1,A2,...,An>,V2=Tuple(null,V))
    即把K作为K2，Tuple(null,V))作为V2
    (K2=List<A1,A2,...,An-1>),V2=Tuple(K,V))
    (K2=List<A1,A2,...,An-2,An>),V2=Tuple(K,V))
    ...
    (K2=List<A2,...,An-1,An>),V2=Tuple(K,V))
    即把K的每一个元素拿掉一次作为K2,Tuple(K,V))作为V2
    */
    val subpatterns = combined.flatMap(pattern => {
      //pattern:(List(a, b, c),1)
      val result = ListBuffer.empty[Tuple2[List[String], Tuple2[List[String], Int]]]
      result += ((pattern._1, (Nil, pattern._2)))//即把K作为K2，Tuple(null,V))作为V2

      val sublist = for {
        i <- 0 until pattern._1.size
        xs = pattern._1.take(i) ++ pattern._1.drop(i + 1)
        if xs.size > 0
      } yield (xs, (pattern._1, pattern._2))
      //上段代码等价于：
      /*
      for(i <- 0 to pattern._1.size){
        val sublist = pattern._1.take(i) ++ pattern._1.drop(i + 1)
        if(sublist.size > 0)
          result += new Tuple2(sublist,new Tuple2(pattern._1,pattern._2))
      }
      即每次去掉一个元素，将剩下的元素集合作为K2
      */
      result ++= sublist
      result.toList
    })
    /*
    * (List(a, b, c),(List(),1))
      (List(b, c),(List(a, b, c),1))
      (List(a, c),(List(a, b, c),1))
      (List(a, b),(List(a, b, c),1))
      (List(b),(List(),4))
      (List(a, b, d),(List(),1))
      (List(b, d),(List(a, b, d),1))
      (List(a, d),(List(a, b, d),1))
      (List(a, b),(List(a, b, d),1))
      (List(b, d),(List(),1))
      (List(d),(List(b, d),1))
      (List(b),(List(b, d),1))
      (List(a, b),(List(),2))
      (List(b),(List(a, b),2))
      (List(a),(List(a, b),2))
      (List(a),(List(),2))
      (List(a, d),(List(),1))
      (List(d),(List(a, d),1))
      (List(a),(List(a, d),1))
      (List(b, c),(List(),3))
      (List(c),(List(b, c),3))
      (List(b),(List(b, c),3))
      (List(a, c),(List(),1))
      (List(c),(List(a, c),1))
      (List(a),(List(a, c),1))
      (List(c),(List(),3))
      (List(d),(List(),1))
    * */
    val rules = subpatterns.groupByKey()
    /*
    * (List(a, b, c),CompactBuffer((List(),1)))
      (List(b),CompactBuffer((List(),4), (List(b, d),1), (List(a, b),2), (List(b, c),3)))
      (List(a, b),CompactBuffer((List(a, b, c),1), (List(a, b, d),1), (List(),2)))
      (List(b, d),CompactBuffer((List(a, b, d),1), (List(),1)))
      (List(a, b, d),CompactBuffer((List(),1)))
      (List(a),CompactBuffer((List(a, b),2), (List(),2), (List(a, d),1), (List(a, c),1)))
      (List(a, d),CompactBuffer((List(a, b, d),1), (List(),1)))
      (List(b, c),CompactBuffer((List(a, b, c),1), (List(),3)))
      (List(a, c),CompactBuffer((List(a, b, c),1), (List(),1)))
      (List(c),CompactBuffer((List(b, c),3), (List(a, c),1), (List(),3)))
      (List(d),CompactBuffer((List(b, d),1), (List(a, d),1), (List(),1)))
    * */
    val assocRules = rules.map(in => {
      println("in=" + in)
      //in:(List(b),CompactBuffer((List(),4), (List(b, d),1), (List(a, b),2), (List(b, c),3)))
      val fromCount = in._2.find(p => p._1 == Nil).get//找到[b]的frequency：即(List(),4)
      println("fromCount=" + fromCount)
      val toList = in._2.filter(p => p._1 != Nil).toList//将规则集合去掉空的
      println("toList=" + toList)
      //toList:CompactBuffer((List(b, d),1), (List(a, b),2), (List(b, c),3))
      if (toList.isEmpty) Nil
      else {
        val result =
          for {
            t2 <- toList
            confidence = t2._2.toDouble / fromCount._2.toDouble
            difference = t2._1 diff in._1
            //diff(that: collection.Seq[A]): List[A] 保存列表中那些不在另外一个列表中的元素，即从集合中减去与另外一个集合的交集
          } yield (((in._1, difference, confidence)))
        result
      }
      //等价于
      /*if (toList.isEmpty) Nil
      else {
        val result = ListBuffer.empty[Tuple3[List[String],List[String],Double]]
        for(t2 <- toList){
          println("t2=" + t2)
          //t2:(List(b, d),1)
          val confidence = t2._2.toDouble / fromCount._2.toDouble
          val difference = t2._1 diff in._1
          println(Tuple3(in._1, difference, confidence))
          result.+=(Tuple3(in._1, difference, confidence))
        }
        result
      }*/
    })
    assocRules.foreach(println)
    /*
    List()
    List((List(b),List(d),0.25), (List(b),List(a),0.5), (List(b),List(c),0.75))
    List((List(a, b),List(c),0.5), (List(a, b),List(d),0.5))
    List((List(b, d),List(a),1.0))
    List()
    List((List(a),List(b),1.0), (List(a),List(d),0.5), (List(a),List(c),0.5))
    List((List(a, d),List(b),1.0))
    List((List(b, c),List(a),0.3333333333333333))
    List((List(a, c),List(b),1.0))
    List((List(c),List(b),1.0), (List(c),List(a),0.3333333333333333))
    List((List(d),List(b),1.0), (List(d),List(a),1.0))
    * */
    val formatResult = assocRules.flatMap(f => {
      f.map(s => (s._1.mkString("[", ",", "]"), s._2.mkString("[", ",", "]"), s._3))
    })
    /*
    * ([b],[d],0.25)
      ([b],[a],0.5)
      ([b],[c],0.75)
      ([a,b],[c],0.5)
      ([a,b],[d],0.5)
      ([b,d],[a],1.0)
      ([a],[b],1.0)
      ([a],[d],0.5)
      ([a],[c],0.5)
      ([a,d],[b],1.0)
      ([b,c],[a],0.3333333333333333)
      ([a,c],[b],1.0)
      ([c],[b],1.0)
      ([c],[a],0.3333333333333333)
      ([d],[b],1.0)
      ([d],[a],1.0)
    * */
    formatResult.saveAsTextFile(output)
    sc.stop()
  }
}