如何在ApacheSpark中获取上一行的数据

从Spark Data框架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 |
+----+-----+----+-------

所需的解决方案是

|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    |
+----+-----+----+-------------+-----------

请帮我

您可以使用滞后函数获取先前的值

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

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创建默认日期，例如01 / month / year，该日期将用于日期排序，即使年份不同

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()

希望这个帮助！