ALS

ML API

• 功能描述

  传入dataSet格式的样本数据，调用fit接口，输出ALS推荐模型。

• 输入输出
  1. 包名：package org.apache.spark.ml.recommendation
  2. 类名：ALS
  3. 方法名：fit
  4. 输入：dataSet[_]，训练样本数据
  5. 算法参数

     算法参数

     def setRank(value: Int): ALS.this.type def setNumUserBlocks(value: Int): ALS.this.type def setNumItemBlocks(value: Int): ALS.this.type def setImplicitPrefs(value: Boolean): ALS.this.type def setAlpha(value: Double): ALS.this.type def setUserCol(value: String): ALS.this.type def setItemCol(value: String): ALS.this.type def setRatingCol(value: String): ALS.this.type def setPredictionCol(value: String): ALS.this.type def setMaxIter(value: Int): ALS.this.type def setRegParam(value: Double): ALS.this.type def setNonnegative(value: Boolean): ALS.this.type def setCheckpointInterval(value: Int): ALS.this.type def setSeed(value: Long): ALS.this.type def setIntermediateStorageLevel(value: String): ALS.this.type def setFinalStorageLevel(value: String): ALS.this.type def setColdStartStrategy(value: String): ALS.this.type def setNumBlocks(value: Int): ALS.type

  6. 新增算法参数。

     参数名称	参数含义	取值类型
     spark.sophon.ALS.bloc kMaxRow	计算格莱姆矩阵行分块大小（不建议修改）	正整数
     spark.sophon.ALS.unp ersistCycle	srcFactorRDD反持久化周期（不建议修改）	正整数

     

     参数及fit代码接口示例：

     val als = new ALS()
         .setMaxIter(numIterations)
         .setUserCol("user")
         .setItemCol("product")
         .setRatingCol("rating")
         .setNonnegative(nonnegative)
         .setImplicitPrefs(implicitPrefs)
         .setNumItemBlocks(numItemBlocks)
         .setNumUserBlocks(numUserBlocks)
         .setRegParam(regParam)
         .setAlpha(alpha)
     val model = als.fit(ratings)

  7. 输出：ALSModel，ALS推荐模型

• 使用样例

  val model = als.fit(ratings)
  val predictions = model.transform(ratings)
  val p = predictions.select("rating", "prediction").rdd   .map { case Row(prediction: Float, label: Float) => (prediction, label) }   .map{t =>     val err = (t._1 - t._2)     err * err   }.mean() println("Mean Squared Error = " + p)

• 结果样例

  Mean Squared Error = 0.9962046583156793