如何选择每组的第一行？

Question

我有一个DataFrame生成如下:

df.groupBy($"Hour", $"Category")
  .agg(sum($"value") as "TotalValue")
  .sort($"Hour".asc, $"TotalValue".desc))

结果如下:

+----+--------+----------+
|Hour|Category|TotalValue|
+----+--------+----------+
|   0|   cat26|      30.9|
|   0|   cat13|      22.1|
|   0|   cat95|      19.6|
|   0|  cat105|       1.3|
|   1|   cat67|      28.5|
|   1|    cat4|      26.8|
|   1|   cat13|      12.6|
|   1|   cat23|       5.3|
|   2|   cat56|      39.6|
|   2|   cat40|      29.7|
|   2|  cat187|      27.9|
|   2|   cat68|       9.8|
|   3|    cat8|      35.6|
| ...|    ....|      ....|
+----+--------+----------+

如您所见,DataFrame按Hour递增顺序排序,然后按TotalValue降序排序.

我想选择每组的顶行,即

• 来自小时组== 0选择(0,cat26,30.9)
• 来自小时组== 1选择(1,cat67,28.5)
• 来自小时组== 2选择(2,cat56,39.6)
• 等等

所以期望的输出将是:

+----+--------+----------+
|Hour|Category|TotalValue|
+----+--------+----------+
|   0|   cat26|      30.9|
|   1|   cat67|      28.5|
|   2|   cat56|      39.6|
|   3|    cat8|      35.6|
| ...|     ...|       ...|
+----+--------+----------+

能够选择每组的前N行可能很方便.

任何帮助都非常感谢.

Answer 1

窗口功能:

像这样的东西应该做的伎俩:

import org.apache.spark.sql.functions.{row_number, max, broadcast}
import org.apache.spark.sql.expressions.Window

val df = sc.parallelize(Seq(
  (0,"cat26",30.9), (0,"cat13",22.1), (0,"cat95",19.6), (0,"cat105",1.3),
  (1,"cat67",28.5), (1,"cat4",26.8), (1,"cat13",12.6), (1,"cat23",5.3),
  (2,"cat56",39.6), (2,"cat40",29.7), (2,"cat187",27.9), (2,"cat68",9.8),
  (3,"cat8",35.6))).toDF("Hour", "Category", "TotalValue")

val w = Window.partitionBy($"hour").orderBy($"TotalValue".desc)

val dfTop = df.withColumn("rn", row_number.over(w)).where($"rn" === 1).drop("rn")

dfTop.show
// +----+--------+----------+
// |Hour|Category|TotalValue|
// +----+--------+----------+
// |   0|   cat26|      30.9|
// |   1|   cat67|      28.5|
// |   2|   cat56|      39.6|
// |   3|    cat8|      35.6|
// +----+--------+----------+

在数据严重偏斜的情况下,此方法效率低下.

简单的SQL聚合后跟join:

或者,您可以加入聚合数据框:

val dfMax = df.groupBy($"hour".as("max_hour")).agg(max($"TotalValue").as("max_value"))

val dfTopByJoin = df.join(broadcast(dfMax),
    ($"hour" === $"max_hour") && ($"TotalValue" === $"max_value"))
  .drop("max_hour")
  .drop("max_value")

dfTopByJoin.show

// +----+--------+----------+
// |Hour|Category|TotalValue|
// +----+--------+----------+
// |   0|   cat26|      30.9|
// |   1|   cat67|      28.5|
// |   2|   cat56|      39.6|
// |   3|    cat8|      35.6|
// +----+--------+----------+

它将保留重复值(如果每小时有多个类别具有相同的总值).您可以按如下方式删除它们:

dfTopByJoin
  .groupBy($"hour")
  .agg(
    first("category").alias("category"),
    first("TotalValue").alias("TotalValue"))

使用订购structs:

虽然没有经过很好的测试,但是不需要连接或窗口功能的技巧很好:

val dfTop = df.select($"Hour", struct($"TotalValue", $"Category").alias("vs"))
  .groupBy($"hour")
  .agg(max("vs").alias("vs"))
  .select($"Hour", $"vs.Category", $"vs.TotalValue")

dfTop.show
// +----+--------+----------+
// |Hour|Category|TotalValue|
// +----+--------+----------+
// |   0|   cat26|      30.9|
// |   1|   cat67|      28.5|
// |   2|   cat56|      39.6|
// |   3|    cat8|      35.6|
// +----+--------+----------+

使用DataSet API(Spark 1.6 +,2.0 +):

Spark 1.6:

case class Record(Hour: Integer, Category: String, TotalValue: Double)

df.as[Record]
  .groupBy($"hour")
  .reduce((x, y) => if (x.TotalValue > y.TotalValue) x else y)
  .show

// +---+--------------+
// | _1|            _2|
// +---+--------------+
// |[0]|[0,cat26,30.9]|
// |[1]|[1,cat67,28.5]|
// |[2]|[2,cat56,39.6]|
// |[3]| [3,cat8,35.6]|
// +---+--------------+

Spark 2.0或更高版本:

df.as[Record]
  .groupByKey(_.Hour)
  .reduceGroups((x, y) => if (x.TotalValue > y.TotalValue) x else y)

最后两种方法可以利用map side combine并且不需要完全shuffle,因此与窗口函数和连接相比,大多数时候应该表现出更好的性能.这些手杖也可以在completed输出模式下与结构化流媒体一起使用.

不要使用:

df.orderBy(...).groupBy(...).agg(first(...), ...)

它似乎有效(特别是在local模式下),但它不可靠(SPARK-16207).学分Tzach琐用于连接相关的JIRA问题.

同样的说明适用于

df.orderBy(...).dropDuplicates(...)

内部使用等效的执行计划.

Answer 2

对于Spark 2.0.2,按多列分组:

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

val w = Window.partitionBy($"col1", $"col2", $"col3").orderBy($"timestamp".desc)

val refined_df = df.withColumn("rn", row_number.over(w)).where($"rn" === 1).drop("rn")

Answer 3

这与zero323的答案完全相同,但是在SQL查询方式中.

假设数据框已创建并注册为

df.createOrReplaceTempView("table")
//+----+--------+----------+
//|Hour|Category|TotalValue|
//+----+--------+----------+
//|0   |cat26   |30.9      |
//|0   |cat13   |22.1      |
//|0   |cat95   |19.6      |
//|0   |cat105  |1.3       |
//|1   |cat67   |28.5      |
//|1   |cat4    |26.8      |
//|1   |cat13   |12.6      |
//|1   |cat23   |5.3       |
//|2   |cat56   |39.6      |
//|2   |cat40   |29.7      |
//|2   |cat187  |27.9      |
//|2   |cat68   |9.8       |
//|3   |cat8    |35.6      |
//+----+--------+----------+

窗口功能:

sqlContext.sql("select Hour, Category, TotalValue from (select *, row_number() OVER (PARTITION BY Hour ORDER BY TotalValue DESC) as rn  FROM table) tmp where rn = 1").show(false)
//+----+--------+----------+
//|Hour|Category|TotalValue|
//+----+--------+----------+
//|1   |cat67   |28.5      |
//|3   |cat8    |35.6      |
//|2   |cat56   |39.6      |
//|0   |cat26   |30.9      |
//+----+--------+----------+

简单的SQL聚合后跟连接:

sqlContext.sql("select Hour, first(Category) as Category, first(TotalValue) as TotalValue from " +
  "(select Hour, Category, TotalValue from table tmp1 " +
  "join " +
  "(select Hour as max_hour, max(TotalValue) as max_value from table group by Hour) tmp2 " +
  "on " +
  "tmp1.Hour = tmp2.max_hour and tmp1.TotalValue = tmp2.max_value) tmp3 " +
  "group by tmp3.Hour")
  .show(false)
//+----+--------+----------+
//|Hour|Category|TotalValue|
//+----+--------+----------+
//|1   |cat67   |28.5      |
//|3   |cat8    |35.6      |
//|2   |cat56   |39.6      |
//|0   |cat26   |30.9      |
//+----+--------+----------+

使用结构排序:

sqlContext.sql("select Hour, vs.Category, vs.TotalValue from (select Hour, max(struct(TotalValue, Category)) as vs from table group by Hour)").show(false)
//+----+--------+----------+
//|Hour|Category|TotalValue|
//+----+--------+----------+
//|1   |cat67   |28.5      |
//|3   |cat8    |35.6      |
//|2   |cat56   |39.6      |
//|0   |cat26   |30.9      |
//+----+--------+----------+

DataSets方式和不做的方式与原始答案相同