count - Spark:如何转换Dataframe API的count(distinct(value))

Question

我正在尝试比较汇总数据的不同方式。

这是我的输入数据，包含2个元素(页面，访问者):

(PAG1,V1)
(PAG1,V1)
(PAG2,V1)
(PAG2,V2)
(PAG2,V1)
(PAG1,V1)
(PAG1,V2)
(PAG1,V1)
(PAG1,V2)
(PAG1,V1)
(PAG2,V2)
(PAG1,V3)

使用以下代码在Spark SQL中使用SQL命令:

import sqlContext.implicits._
case class Log(page: String, visitor: String)
val logs = data.map(p => Log(p._1,p._2)).toDF()
logs.registerTempTable("logs")
val sqlResult= sqlContext.sql(
                              """select page
                                       ,count(distinct visitor) as visitor
                                   from logs
                               group by page
                              """)
val result = sqlResult.map(x=>(x(0).toString,x(1).toString))
result.foreach(println)

我得到以下输出:

(PAG1,3) // PAG1 has been visited by 3 different visitors
(PAG2,2) // PAG2 has been visited by 2 different visitors

现在，我想使用Dataframes和thiers API获得相同的结果，但无法获得相同的输出:

import sqlContext.implicits._
case class Log(page: String, visitor: String)
val logs = data.map(p => Coppia(p._1,p._2)).toDF()
val result = log.select("page","visitor").groupBy("page").count().distinct
result.foreach(println)

实际上，这就是我得到的输出:

[PAG1,8]  // just the simple page count for every page
[PAG2,4]

这可能有点愚蠢，但我现在看不到。

提前致谢!

FF

Answer 1

您需要的是DataFrame聚合函数countDistinct:

import sqlContext.implicits._
import org.apache.spark.sql.functions._

case class Log(page: String, visitor: String)

val logs = data.map(p => Log(p._1,p._2))
            .toDF()

val result = logs.select("page","visitor")
            .groupBy('page)
            .agg('page, countDistinct('visitor))

result.foreach(println)