如何基于Spark Scala中的现有列添加新列

Question

Halo

i'm done to build recommendation using Mllib ALS in apache spark, with output

user | product | rating
    1 | 20 | 0.002
    1 | 30 | 0.001
    1 | 10 | 0.003
    2 | 20 | 0.002
    2 | 30 | 0.001
    2 | 10 | 0.003

but i need to change data structure based on sort by rating, like that :

user | product | rating | number_rangking
    1 | 10 | 0.003 | 1
    1 | 20 | 0.002 | 2 
    1 | 30 | 0.001 | 3
    2 | 10 | 0.002 | 1
    2 | 20 | 0.001 | 2
    2 | 30 | 0.003 | 3

how can i do that? maybe any one can give me a clue...

thx

Answer 1

All you need is a window functions depending on details you choose either rank or rowNumber

import org.apache.spark.sql.expressions.Window
import org.apache.spark.sql.functions.rank

val w = Window.partitionBy($"user").orderBy($"rating".desc)

df.select($"*", rank.over(w).alias("number_rangking")).show
// +----+-------+------+---------------+
// |user|product|rating|number_rangking|
// +----+-------+------+---------------+
// |   1|     10| 0.003|              1|
// |   1|     20| 0.002|              2|
// |   1|     30| 0.001|              3|
// |   2|     10| 0.003|              1|
// |   2|     20| 0.002|              2|
// |   2|     30| 0.001|              3|
// +----+-------+------+---------------+

Using plain RDD you can groupByKey , process locally and flatMap :

rdd
  // Convert to PairRDD
  .map{case (user, product, rating) => (user, (product, rating))}
  .groupByKey 
  .flatMap{case (user, vals) => vals.toArray
    .sortBy(-_._2) // Sort by rating
    .zipWithIndex // Add index
    // Yield final values
    .map{case ((product, rating), idx) => (user, product, rating, idx + 1)}}