大数据学习day29-----spark09-------1. 练习： 统计店铺按月份的销售额和累计到该月的总销售额（SQL, DSL,RDD） 2. 分组topN的实现（row_number(), rank(), dense_rank()方法的区别）3. spark自定义函数-UDF

1. 练习

数据：

 （1）需求1：统计有过连续3天以上销售的店铺有哪些,并且计算出连续三天以上的销售额

第一步：将每天的金额求和（同一天可能会有多个订单）

SELECT
  sid,dt,SUM(money) day_money
FROM
  v_orders
GROUP BY sid,dt
View Code

第二步：给每个商家中每日的订单按时间排序并打上编号

SELECT 
  sid,dt,day_money,
  ROW_NUMBER() OVER(PARTITION BY sid ORDER BY dt) rn
FROM
(
  SELECT
    sid,dt,SUM(money) day_money
  FROM
    v_orders
  GROUP BY sid,dt
) t1
View Code

 第三步：获取date与rn的差值的字段

SELECT
  sid ,dt,day_money,date_sub(dt,rn) diff
FROM
(
  SELECT 
    sid,dt,day_money,
    ROW_NUMBER() OVER(PARTITION BY sid ORDER BY dt) rn
  FROM
  (
    SELECT
      sid,dt,SUM(money) day_money
    FROM
      v_orders
    GROUP BY sid,dt
  ) t1
) t2
View Code

第四步： 最终结果

SELECT
  sid,MIN(dt),MAX(dt),SUM(day_money) cmoney,COUNT(*) cc
FROM
(
  SELECT
    sid ,dt,day_money,date_sub(dt,rn) diff
  FROM
  (
    SELECT 
      sid,dt,day_money,
      ROW_NUMBER() OVER(PARTITION BY sid ORDER BY dt) rn
    FROM
    (
      SELECT
        sid,dt,SUM(money) day_money
      FROM
        v_orders
      GROUP BY sid,dt
    ) t1
  ) t2
)
GROUP BY sid,diff
HAVING cc >=3
View Code

 （2）需求2：统计店铺按月份的销售额和累计到该月的总销售额

  • SQL风格（只写sq语句，省略代码部分）
SELECT 
  sid,month,month_sales,
  SUM(month_sales) OVER(PARTITION BY sid ORDER BY month) total_sales  // 默认是其实位置到当前位置的累加
  --PARTITION BY sid ORDER BY mth ASC ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW  完整的写法
FROM
(
  SELECT
    sid,
    DATE_FORMAT(dt,'yyyy-MM') month,
    --substr(dt,1,7) month,  用此函数来取月份也行
    SUM(money) month_sales
  FROM
    v_orders
  GROUP BY sid, month
)

结果

  • DSL风格
object RollupMthIncomeDSL {
  def main(args: Array[String]): Unit = {
    // 创建SparkSession
    val spark: SparkSession = SparkSession.builder()
      .appName(this.getClass.getSimpleName)
      .master("local[*]")
      .getOrCreate()
    // 读取文件创建DataSet
    val orders: DataFrame = spark.read
      .option("header", "true")
      .option("inferSchema", "true")  // inferSchema为true可以自动推测数据的类型，默认false，则所有的数据都是String类型的
      .csv("F:\大数据第三阶段\spark\spark-day09\资料\order.csv")
    import spark.implicits._
    import org.apache.spark.sql.functions._
    // 获取月份，并按照sid和月份进行分组，聚合
    val result: DataFrame = orders.groupBy($"sid", date_format($"dt", "yyyy-MM") as "month")
      .agg(sum($"money") as "month_sales")
      // withColumn相当于在原有基础上再增加一列，此处使用select重新获取表也行
      //.select('sid, 'month, 'month_sales, sum('month_sales) over(Window.partitionBy('sid)
      // .orderBy('month).rowsBetween(Window.unboundedPreceding, Window.currentRow)) as "rollup_sales")
      .withColumn("rollup_sales", sum('month_sales) over (Window.partitionBy('sid) // 'sid相当于$"sid"
        .orderBy('month).rowsBetween(Window.unboundedPreceding, Window.currentRow)))
    result.show()
    spark.stop()
  }
}
  • RDD风格
object RollupMthIncomeRDD {
  def main(args: Array[String]): Unit = {
    // 创建SparkContext
    val conf = new SparkConf()
      .setAppName(this.getClass.getName)
      .setMaster("local[*]")
    val sc: SparkContext = new SparkContext(conf)
    val lines: RDD[String] = sc.textFile("F:\大数据第三阶段\spark\spark-day09\资料\order.csv")
    val reduced: RDD[((String, String), Double)] = lines.map(line => {
      val fields: Array[String] = line.split(",")
      val sid: String = fields(0)
      val dateStr: String = fields(1)
      val month: String = dateStr.substring(0, 7)
      val money: Double = fields(2).toDouble
      ((sid, month), money)
    }).reduceByKey(_ + _)
    // 按照shop id分组
    val result: RDD[(String, String, String, Double)] = reduced.groupBy(_._1._1).mapValues(it => {
      //将迭代器中的数据toList放入到内存
      //并且按照月份排序【字典顺序】
      val sorted: List[((String, String), Double)] = it.toList.sortBy(_._1._2)
      var rollup = 0.0
      sorted.map(t => {
        val sid = t._1._1
        val month = t._1._2
        val month_sales = t._2
        rollup += month_sales
        (sid, month, rollup)
      })
    }).flatMapValues(lst => lst).map(t => (t._1, t._2._1, t._2._2, t._2._3))
    result.foreach(println)
    sc.stop()
    }
}
View Code

注意点：可以直接读取csv文件获取DataFram，再获取rdd，如下

 2. 分组topN的实现（大数据学习day21中的计算学科最受欢迎老师topN）

  • 　SQL　

注意点：此处的文件格式是text的，所以需要用SparkContext的textFile方法来读取数据，然后处理此数据，得到需要的字段（subject,teacher），再利用toDF("subject", "teacher")方法获取对应的DataFrame，从而创建相应的视图

object FavoriteTeacherSQL {
  def main(args: Array[String]): Unit = {
    val spark: SparkSession = SparkSession.builder()
      .appName(this.getClass.getSimpleName)
      .master("local[*]")
      .getOrCreate()
        import spark.implicits._
    val lines: RDD[String] = spark.sparkContext.textFile("E:\javafile\spark\teacher100.txt")
    // 处理数据，获取DataFrame，用于创建视图
    val df: DataFrame = lines.map(line => {
      val fields = line.split("/")
      val subject = fields(2).split("\.")(0)
      val teacher = fields(3)
      (subject, teacher)
    }).toDF("subject", "teacher")
    // 创建视图
    df.createTempView("v_teacher")

    var topN: Int = 2
    // SQL实现分组topN
    spark.sql(
      s"""
        |SELECT
        |  subject,teacher,counts
        |  rk
        |FROM
        |(
        |  SELECT
        |    subject,teacher,counts,
        |    RANK() OVER(PARTITION BY subject ORDER BY counts DESC) rk
        |  FROM
        |  (
        |    SELECT
        |      subject,teacher,
        |      count(*) counts
        |    FROM
        |      v_teacher
        |    GROUP BY subject, teacher
        |  ) t1
        |) t2 WHERE rk <= $topN
        |""".stripMargin).show()
  }
}
View Code

此处的小知识点：

row_number(), rank(), dense_rank()方法的区别

row_number() over() 打行号，行号从1开始
rank() over() 排序，有并列，如果有两个第1，就没有第2了,然后直接第3，跳号
dense_rank() over() 排序，有并列，不跳号
  • DSL
object FavoriteTeacherDSL {
  def main(args: Array[String]): Unit = {
    val spark: SparkSession = SparkSession.builder()
      .appName(this.getClass.getSimpleName)
      .master("local[*]")
      .getOrCreate()
    import spark.implicits._
    val lines: RDD[String] = spark.sparkContext.textFile("E:\javafile\spark\teacher100.txt")
    // 处理数据，获取DataFrame，用于创建视图
    val df: DataFrame = lines.map(line => {
      val fields = line.split("/")
      val subject = fields(2).split("\.")(0)
      val teacher = fields(3)
      (subject, teacher)
    }).toDF("subject", "teacher")
    import org.apache.spark.sql.functions._
    df.groupBy("subject","teacher")
      .agg(count("*") as "counts")
      .withColumn("rk",dense_rank().over(Window.partitionBy($"subject").orderBy($"counts" desc)) )
      .filter('rk <= 2)
      .show()

    spark.stop()
  }
}
View Code

3. spark自定义函数-UDF

　　UDF：一进一出（输入一行，返回一行）

　　UDTF: 一进多出

　　UDAF: 多进一出

object MyConcatWsUDF {
  def main(args: Array[String]): Unit = {
    val spark = SparkSession.builder().appName(this.getClass.getSimpleName)
      .master("local[*]")
      .getOrCreate()
    import spark.implicits._
    val tp: Dataset[(String, String)] = spark.createDataset(List(("aaa", "bbb"), ("aaa", "ccc"), ("aaa", "ddd")))
    val df: DataFrame = tp.toDF("f1", "f2")
    //注册自定义函数
    //MY_CONCAT_WS函数名称
    //后面传入的scala的函数就是具有的实现逻辑
    spark.udf.register("MY_CONCAT_WS", (s: String, first: String, second:String) => {
      first + s + second
    })
    
    import org.apache.spark.sql.functions._
    //df.selectExpr("CONCAT_WS('-', f1, f2) as f3")
    //df.select(concat_ws("-", $"f1", 'f2) as "f3").show()
    //df.selectExpr("MY_CONCAT_WS('_', f1, f2) as f3").show()
    df.createTempView("v_data")

    spark.sql(
      """
        |SELECT MY_CONCAT_WS('-', f1, f2) f3 FROM v_data
      """.stripMargin).show()
    spark.stop()
  }
}
View Code
【推广】 免费学中医，健康全家人
原文地址：https://www.cnblogs.com/jj1106/p/12143413.html