拿到某超市的銷售資料，將資料整理後得到一年三千萬條交易記錄，想試試用spark中的推薦系統做一下預測

先把資料匯入到HDFS中，資料需要使用者id，商品id，和購買次數，這裡我拿購買次數當作電影推薦系統中的電影評分

HDFS中的資料用":"分割開。如下：

461365:22535:1.0
461365:5059:1.0
461365:5420:4.0
461366:1987:4.0
461366:31911:1.0

進入spark-shell

引入需要的mllib包和日誌的設定

import org.apache.spark.mllib.recommendation.{ALS, Rating,MatrixFactorizationModel}
import org.apache.spark.sql.hive.HiveContext
import org.apache.log4j.{Logger,Level}
import org.apache.spark.mllib.evaluation.{RankingMetrics, RegressionMetrics}
Logger.getLogger("org.apache.spark").setLevel(Level.WARN)
Logger.getLogger("org.eclipse.jetty.server").setLevel(Level.OFF)
 

val data = sc.textFile("/input/rate")
val ratings = data.map(_.split(':') match { case Array(user, item, rate) =>  Rating(user.toInt, item.toInt, rate.toDouble)})

scala> val users = ratings.map(_.user).distinct()
users: org.apache.spark.rdd.RDD[Int] = MapPartitionsRDD[1356] at distinct at <console>:35

scala> val products = ratings.map(_.product).distinct()
products: org.apache.spark.rdd.RDD[Int] = MapPartitionsRDD[1360] at distinct at <console>:35

scala> println("Got "+ratings.count()+" ratings from "+users.count+" users on "+products.count+" products.")
Got 30299054 ratings from 354172 users on 45786 products.   
 

val splits = ratings.randomSplit(Array(0.8, 0.2))
val training = splits(0)
val test = splits(1)

進行訓練，並設定引數

Rank： 對應ALS模型中的因子個數，即矩陣分解出的兩個矩陣的新的行/列數

numIterations：模型迭代最大次數
引數0.01： 控制模型的正則化過程，從而控制模型的過擬合情況。

val rank = 30
val numIterations = 12
val model = ALS.train(training, rank, numIterations, 0.01)
 

val testUsersProducts = test.map { case Rating(user, product, rate) =>
           (user, product)
          }
val predictions = model.predict(testUsersProducts).map { case Rating(user, product, rate) =>
    ((user, product), rate)
}
val ratesAndPreds = ratings.map { case Rating(user, product, rate) =>
  ((user, product), rate)
   }.join(predictions)
val rmse= math.sqrt(ratesAndPreds.map { case ((user, product), (r1, r2)) =>
  val err = (r1 - r2)
  err * err
}.mean())