object RecommendationScores {

  private def precisionAtK(k: Int, actual: Array[Int], predicted: Array[Int]): Double = {
    //val pred = predicted.take(k)
    val a = actual.length
    //actual.intersect(pred).length / a : Double
    a
  }

  def averagePrecisionAtK(k: Int) = {
    new Score[Array[Int]] with PerformanceScore {
      //TODO
      override def evaluate(data: DataSet[(Array[Int], Array[Int])]): DataSet[Double] = {
        val perUser = data.map { trueAndPred => precisionAtK(k, trueAndPred._1, trueAndPred._2)
          //val pred = trueAndPred._2.take(k)
          //val a = trueAndPred._1.length
          //trueAndPred._1.intersect(pred).length / a : Double}
          //a
        }
          //val perUser = data.map(trueAndPred => 2)
        perUser.mean()
      }
    }
  }

  def meanAveragePrecisionAtK(k: Int) = {
    new Score[Array[Int]] with PerformanceScore {
      //TODO
      override def evaluate(data: DataSet[(Array[Int], Array[Int])]): DataSet[Double] = {
        val perUser = data.map{ trueAndPred =>
          val precisions = 1 to k map (_k => precisionAtK(_k, trueAndPred._1, trueAndPred._2))
          precisions.sum / k
        }
        perUser.mean()
      }
    }
  }

  def averageRecallAtK(k: Int) = {
    new Score[Array[Int]] with PerformanceScore {
      //TODO
      override def evaluate(data: DataSet[(Array[Int], Array[Int])]): DataSet[Double] = {
        val perUser = data.map { trueAndPred =>
          val act = trueAndPred._1.toSet
          val pred = trueAndPred._2.take(k).toSet
          act.intersect(pred).size / k : Double
        }
        perUser.mean()
      }
    }
  }

  private def iDCG(k: Int) = {

  }
}