如何在 Apache Spark 中获取上一行的数据

Question

从 Spark 数据框中查找每个城市的上个月销售额

|City|     Month   |Sale|
+----+----------- +----- +
|  c1|    JAN-2017|  49 |
|  c1|    FEB-2017|  46 |
|  c1|    MAR-2017|  83 |
|  c2|    JAN-2017|  59 |
|  c2|    MAY-2017|  60 |
|  c2|    JUN-2017|  49 |
|  c2|    JUL-2017|  73 |
+----+-----+----+-------

需要的解决方案是

|City|     Month  |Sale   |previous_sale|
+----+-----+-------+-------------+--------
|  c1|    JAN-2017|  49|           NULL  |
|  c1|    FEB-2017|  46|           49    |
|  c1|    MAR-2017|  83|           46    |
|  c2|    JAN-2017|  59|           NULL  |
|  c2|    MAY-2017|  60|           59    |
|  c2|    JUN-2017|  49|           60    |
|  c2|    JUL-2017|  73|           49    |
+----+-----+----+-------------+-----------

请帮帮我

Answer 1

你可以使用lag函数来获取之前的值

如果要按月份排序，则需要转换为正确的日期。对于"JAN-2017" 到"01-01-2017" 是这样的。

import spark.implicits._
val df = spark.sparkContext.parallelize(Seq(
  ("c1", "JAN-2017", 49),
("c1", "FEB-2017", 46),
("c1", "MAR-2017", 83),
("c2", "JAN-2017", 59),
("c2", "MAY-2017", 60),
("c2", "JUN-2017", 49),
("c2", "JUL-2017", 73)
)).toDF("city", "month", "sales")

val window = Window.partitionBy("city").orderBy("month")

df.withColumn("previous_sale", lag($"sales", 1, null).over(window)).show

输出：

+----+--------+-----+----+
|city|   month|sales| previous_sale|
+----+--------+-----+----+
|  c1|FEB-2017|   46|null|
|  c1|JAN-2017|   49|  46|
|  c1|MAR-2017|   83|  49|
|  c2|JAN-2017|   59|null|
|  c2|JUL-2017|   73|  59|
|  c2|JUN-2017|   49|  73|
|  c2|MAY-2017|   60|  49|
+----+--------+-----+----+

您可以使用此 UDF 创建一个默认日期，如 01/月/年，即使它有不同的年份，也会使用日期排序

val fullDate = udf((value :String )=>
{
  val months = List("JAN", "FEB", "MAR", "APR", "MAY", "JUN", "JUL", "AUG", "SEP", "OCT", "NOV", "DEC")
  val splited = value.split("-")
  new Date(splited(1).toInt, months.indexOf(splited(0)) + 1, 1)
})

df.withColumn("month", fullDate($"month")).show()

希望这能帮到你！

Answer 2

package com.incedo.pharma
import org.apache.spark.SparkConf
import org.apache.spark.SparkContext
import org.apache.spark.sql.SQLContext
import org.apache.spark.SparkContext._
import org.apache.spark.sql.functions._ 
import org.apache.spark.sql.functions.unix_timestamp
import org.apache.spark.sql.functions.udf
import org.apache.spark.sql.SparkSession
//import org.apache.spark.sql.expressions.Window
import java.sql.Date

class SecondTask {
  def previousMonthSale(){
    val conf = new SparkConf().setAppName("Excel-read-write").setMaster("local")
    val sc = new SparkContext(conf)
    val sqlContext = new org.apache.spark.sql.SQLContext(sc)
    val ss = SparkSession.builder().master("local").appName("Excel-read-write").getOrCreate()
    import ss.sqlContext.implicits._
    val df = sqlContext.read.format("com.databricks.spark.csv")
             .option("header", "true")
             .option("inferSchema", "true")
             .option("delimiter", "|")
             .load("taskTwoData.csv")
    val df1 = df.withColumn("Timestamp", unix_timestamp(df("Date"), "MM/dd/yyyy").cast("timestamp"))
    val df2 = df1.withColumn("Month",month(df1("Timestamp")))
    val df3 = df2.groupBy("City", "Month").agg(sum(col("Sale")).alias("Current_Month_Total_Sale")).orderBy("City","Month")
    val df4 = df3.withColumn("pre_month",df3("Month")-1)
    val df5 = df4.alias("a").join(df3.alias("b"),$"a.pre_month" === $"b.Month" && $"a.City" === $"b.City","left_outer")
              .select($"a.City",$"a.Month",$"a.Current_Month_Total_Sale",($"b.Current_Month_Total_Sale")
              .alias("Previous_Month_Total_Sale")).na.fill(0,Seq("Previous_Month_Total_Sale"))

    val df6 = df5.withColumn("Percent_Sale_Change",round(((df5("Current_Month_Total_Sale") - df5("Previous_Month_Total_Sale"))/df5("Current_Month_Total_Sale"))*100,2))
    val df7 = df6.groupBy("City").max("Current_Month_Total_Sale").alias("Max_Sale").orderBy("City")
    //df7.show()
    val df8 = df6.join(df7, Seq("City"))
    val df9 = df8.withColumn("Percent_Sale_By_Max_Sale", round(df8("Current_Month_Total_Sale")/df8("max(Current_Month_Total_Sale)"),2))
             .drop("max(Current_Month_Total_Sale)")
    df9.toDF().show()
  }
}

object taskTwo {
  def main(arr: Array[String]) {
    new SecondTask().previousMonthSale()
  }`enter code here`
}