spark基于ALS的协同过虑实例

最近在学习spark，网上有很多关于ALS的实例，大多数千篇一例，决定自己动手写一个实例，力求做到可以运行，有结果。

1、数据集准备：

到 http://grouplens.org/datasets/movielens/ 下载电影评分数据，查看一下README里面有对数据集的介绍。

对数据进行预处理：

 

cat u1.base | awk -F "\t" '{print $1"::"$2"::"$3"::"$4}' > ratings.dat  
cat u.item | awk -F "|" '{print $1"\t"$2"\t"$3}' > movies.dat  

 结果如下：

 

 

[root@hongboVM ml-100k]# head -10 ratings.data 
1::1::5::874965758
1::2::3::876893171
1::3::4::878542960
1::4::3::876893119
1::5::3::889751712
1::7::4::875071561
1::8::1::875072484
1::9::5::878543541
1::11::2::875072262
1::13::5::875071805

 

[root@hongboVM ml-100k]# head -10 movies.data 
1::Toy Story (1995)::01-Jan-1995
2::GoldenEye (1995)::01-Jan-1995
3::Four Rooms (1995)::01-Jan-1995
4::Get Shorty (1995)::01-Jan-1995
5::Copycat (1995)::01-Jan-1995
6::Shanghai Triad (Yao a yao yao dao waipo qiao) (1995)::01-Jan-1995
7::Twelve Monkeys (1995)::01-Jan-1995
8::Babe (1995)::01-Jan-1995
9::Dead Man Walking (1995)::01-Jan-1995
10::Richard III (1995)::22-Jan-1996

 

[root@hongboVM ml-100k]# head -10 user.data 
1|24|M|technician|85711
2|53|F|other|94043
3|23|M|writer|32067
4|24|M|technician|43537
5|33|F|other|15213
6|42|M|executive|98101
7|57|M|administrator|91344
8|36|M|administrator|05201
9|29|M|student|01002
10|53|M|lawyer|90703

 

  将数据上传到hdfs上。

2、数据处理基于思路：

  先使用ratings数据训练模型，然后使用模型做预测。打印出推荐信息：

3、代码如下：

 

package com.bohai.mllib

import org.apache.log4j.{Level, Logger}
import org.apache.spark.mllib.recommendation.{ALS, Rating}
import org.apache.spark.{SparkConf, SparkContext}

object MoviesRecommondNew {
  def main(args: Array[String]) {
    //屏蔽日志，由于结果是打印在控制台上的，为了方便查看结果，将spark日志输出关掉
    //解决spark日志输出的问题最好的解决办法是:修改spark日志文件，将日志写入文件中
    Logger.getLogger("org.apache.spark").setLevel(Level.WARN)
    Logger.getLogger("org.eclipse.jetty.server").setLevel(Level.OFF)

    val conf = new SparkConf().setAppName("MoviesRecommondNew")
    val sc = new SparkContext(conf)
    //ratings.data的数据
    val data = sc.textFile("/data/ratings.data")
    val test_data = sc.textFile("/data/test.data")
    //注意这里的分隔符
    val user_data = sc.textFile("/data/user.data").map(x => x.split("[|]") match {
      case Array(userId, age, gender, occupation, zipCode) => Users(userId.toInt, age.toInt, gender)
    })
    val movie_data = sc.textFile("/data/movies.data").map(x =>x.split("::"))

    println("rate data count is : " + data.count())
    val ratings = data.map(x => x.split("::") match {
      case Array(user, item, rate, ts) => Rating(user.toInt, item.toInt, rate.toDouble)
    })

    val test_ratings = test_data.map(x => x.split("::") match {
      case Array(user, item, rate, ts) => Rating(user.toInt, item.toInt, rate.toDouble)
    })
    println("test rate data count is : " +test_ratings.count())
    //println("test rate data is : " + test_ratings.take(2))

    val userIds = user_data.map(_.id)

    println("user data is : " + userIds.count())

    //生成k,v形式，便于通过movieID找到movieName
    val movieDataMap = movie_data.map(x => (x(0).toInt,x(1))).collectAsMap()
    //broadcast进行广播
    val bMovieDataMap = sc.broadcast(movieDataMap)

    val rank = 10
    val numIterations = 10
    val model = ALS.train(ratings, rank, numIterations, 0.001)
    val usersProducts = ratings.map { case Rating(user, prod, rate) => (user, prod) }
    val predictions = model.predict(usersProducts).map { case Rating(user, product, rate) => ((user, product), rate) }
    val ratesAndPreds = ratings.map { case Rating(user, product, rate) => ((user, product), rate) }.join(predictions)
    val MSE = ratesAndPreds.map { case ((user, product), (r1, r2)) =>
      val err = r1 - r2
      err * err
    }.mean()
    println(s"Mean squared Error = $MSE")


    val userID = 384
    val moviesForUser = ratings.keyBy(_.user).lookup(userID)
    println(s"用户$userID 评价过的电影：\n")
    for (movieID <- moviesForUser.map(f => f.product)) {
      //movie_data.filter{x  => x(0).toInt == movieID}.map(x => x(1)).collect().foreach(println)
      println(bMovieDataMap.value.getOrElse(movieID,""))
    }

    println(s"用户$userID 推荐的电影：\n")
    val recommendProds:Array[Rating] = model.recommendProducts(userID,20)
    for (recommend <- recommendProds) {
      //println(recommend.user + "," + recommend.product + "," + recommend.rating)
      //movie_data.filter{x  => x(0).toInt == recommend.product}.map(x => x(1)).collect().foreach(println)
      println(bMovieDataMap.value.getOrElse(recommend.product,""))
    }

    println("为每个用户推荐10个电影：\n")
    val allRecommendations = model.recommendProductsForUsers(10).map{
      case (userId,recommends) =>
        val str = new StringBuilder()
        for (r <- recommends) {
          if (str.nonEmpty) {
            str.append("::")
          }
          str.append(r.product)
        }
        (userId,str.toString())
    }
    allRecommendations.take(10).foreach(println)
  }

  //样例类，用作SparkSQL隐式转换
  case class Ratings(userId: Int, movieId: Int, rating: Int)

  case class Movies(id: Int, moveTitle: String, releaseDate: String)

  case class Users(id: Int, age: Int, gender: String)

}

   提交到spark上进行测试：

 

 

spark-submit --master spark://172.4.23.99:7077 --num-executors 4 --executor-cores 2 --class com.bohai.mllib.MoviesRecommondNew  ./simple-project_2.10-1.0.jar

 运行结果如下 ：

 

 

rate data count is : 80000                                                      
test rate data count is : 20000
user data is : 943
Mean squared Error = 0.44838904095188975                                        
用户384 评价过的电影：

Contact (1997)
Starship Troopers (1997)
English Patient, The (1996)
Evita (1996)
Air Force One (1997)
L.A. Confidential (1997)
Titanic (1997)
As Good As It Gets (1997)
Cop Land (1997)
Conspiracy Theory (1997)
Desperate Measures (1998)
Game, The (1997)
Tomorrow Never Dies (1997)
That Darn Cat! (1997)
Peacemaker, The (1997)
Cats Don't Dance (1997)
用户384 推荐的电影：

Amos & Andrew (1993)
I'm Not Rappaport (1996)
Ruling Class, The (1972)
Amateur (1994)
Englishman Who Went Up a Hill, But Came Down a Mountain, The (1995)
To Live (Huozhe) (1994)
Stupids, The (1996)
Cemetery Man (Dellamorte Dellamore) (1994)
Eye for an Eye (1996)
M. Butterfly (1993)
Crooklyn (1994)
8 1/2 (1963)
Herbie Rides Again (1974)
City of Lost Children, The (1995)
Vanya on 42nd Street (1994)
Afterglow (1997)
Addiction, The (1995)
Die xue shuang xiong (Killer, The) (1989)
Haunted World of Edward D. Wood Jr., The (1995)
Mute Witness (1994)
为每个用户推荐10个电影：

(656,1313::998::1480::1206::149::253::1411::1451::974::401)                     
(692,1192::1058::1286::1425::1483::703::1113::1404::960::753)
(932,1313::57::1643::947::601::1128::1131::954::1224::965)
(772,1131::860::1192::1512::1205::1129::967::1128::57::445)
(324,1019::1192::904::1022::982::1262::320::1404::786::1298)
(180,1426::394::1195::1184::793::1389::1069::1208::1245::1120)
(340,860::1192::974::998::1131::1273::440::1178::296::1483)
(320,1286::1056::115::1425::320::916::534::962::960::703)
(752,1286::1643::115::906::800::1160::296::1129::767::1049)
(744,1131::1380::624::1211::1137::782::860::1192::1113::630)

 总结：（1）下载下来的数据是使用 \t 进行分隔的，需要进行预处理；（2）为方便通过movieID找到movieName，我们将movieID、movieName封装成map，并对map进行 broadcast ，提交检索的性能。（3）没有计算预测的结果与实际评分的误 差，有空补上；

参考：

http://blog.csdn.net/oopsoom/article/details/34462329

http://blog.javachen.com/2015/06/01/how-to-implement-collaborative-filtering-using-spark-als.html