【Spark】（十）詳解 Spark DataSet

一、前言

Spark的發展史可以簡單概括爲三個階段，分別爲：RDD、DataFrame 和DataSet。在Spark 2.0之前，使用Spark必須先創建SparkConf和SparkContext，不過在Spark 2.0中只要創建一個SparkSession就可以了，SparkConf、SparkContext和SQLContext都已經被封裝在SparkSession當中，它是Spark的一個全新切入點，大大降低了Spark的學習難度。

二、創建SparkSession

創建SparkSession的方式非常簡單，如下：

//創建SparkSession
val spark = SparkSession.builder()
  .master("local[*]")
  .appName("dataset")
  .enableHiveSupport()  //支持hive，如果代碼中用不到hive的話，可以省略這一條
  .getOrCreate()

三、DataSet/DataFrame的創建

1、序列創建 DataSet

//1、產生序列dataset
val numDS = spark.range(5, 100, 5)
numDS.orderBy(desc("id")).show(5)  //降序排序，顯示5個
numDS.describe().show()  //打印numDS的摘要

結果如下所示：

+---+
| id|
+---+
| 95|
| 90|
| 85|
| 80|
| 75|
+---+
only showing top 5 rows
 
+-------+------------------+
|summary|                id|
+-------+------------------+
|  count|                19|
|   mean|              50.0|
| stddev|28.136571693556885|
|    min|                 5|
|    max|                95|
+-------+------------------+

2、集合創建 DataSet

首先創建幾個可能用到的樣例類：

//樣例類
case class Person(name: String, age: Int, height: Int)
case class People(age: Int, names: String)
case class Score(name: String, grade: Int)

然後定義隱式轉換：

import spark.implicits._

最後，定義集合，創建 DataSet

//2、集合轉成dataset
val seq1 = Seq(Person("xzw", 24, 183), Person("yxy", 24, 178), Person("lzq", 25, 168))
val ds1 = spark.createDataset(seq1)
ds1.show()

結果如下所示：

+----+---+------+
|name|age|height|
+----+---+------+
| xzw| 24|   183|
| yxy| 24|   178|
| lzq| 25|   168|
+----+---+------+

3、RDD 轉成 DataFrame

//3、RDD轉成DataFrame
val array1 = Array((33, 24, 183), (33, 24, 178), (33, 25, 168))
val rdd1 = spark.sparkContext.parallelize(array1, 3).map(f => Row(f._1, f._2, f._3))
val schema = StructType(
  StructField("a", IntegerType, false) ::
    StructField("b", IntegerType, true) :: Nil
)
val rddToDataFrame = spark.createDataFrame(rdd1, schema)
rddToDataFrame.show(false)

結果如下所示：

+---+---+
|a  |b  |
+---+---+
|33 |24 |
|33 |24 |
|33 |25 |
+---+---+

4、讀取文件

//4、讀取文件，這裏以csv文件爲例
val ds2 = spark.read.csv("C://Users//Machenike//Desktop//xzw//test.csv")
ds2.show()

結果如下所示：

+---+---+----+
|_c0|_c1| _c2|
+---+---+----+
|xzw| 24| 183|
|yxy| 24| 178|
|lzq| 25| 168|
+---+---+----+

5、讀取文件，並配置詳細參數

//5、讀取文件，並配置詳細參數
val ds3 = spark.read.options(Map(("delimiter", ","), ("header", "false")))
  .csv("C://Users//Machenike//Desktop//xzw//test.csv")
ds3.show()

結果如下圖所示：

+---+---+----+
|_c0|_c1| _c2|
+---+---+----+
|xzw| 24| 183|
|yxy| 24| 178|
|lzq| 25| 168|
+---+---+----+

四、DataSet 基礎函數

//1、DataSet存儲類型
val seq1 = Seq(Person("xzw", 24, 183), Person("yxy", 24, 178), Person("lzq", 25, 168))
val ds1 = spark.createDataset(seq1)
ds1.show()
ds1.checkpoint()
ds1.cache()
ds1.persist()
ds1.count()
ds1.unpersist(true)
 
//2、DataSet結構屬性
ds1.columns
ds1.dtypes
ds1.explain()
 
//3、DataSet rdd數據互換
val rdd1 = ds1.rdd
val ds2 = rdd1.toDS()
ds2.show()
val df2 = rdd1.toDF()
df2.show()
 
//4、保存文件
df2.select("name", "age", "height").write.format("csv").save("./save")

五、DataSet 的 Actions 操作

六、DataSet 的轉化操作

package sparkml
 
import org.apache.log4j.{Level, Logger}
import org.apache.spark.sql.SparkSession
import org.apache.spark.sql.functions._
 
//樣例類
case class Person(name: String, age: Int, height: Int)
case class People(age: Int, names: String)
case class Score(name: String, grade: Int)
 
object WordCount2 {
  def main(args: Array[String]): Unit = {
    //設置日誌輸出格式
    Logger.getLogger("org").setLevel(Level.WARN)
 
    //創建SparkSession
    val spark = SparkSession.builder()
      .master("local[*]")
      .appName("dataset")
      .getOrCreate()
 
    import spark.implicits._
 
    //seq創建dataset
    val seq1 = Seq(Person("leo", 29, 170), Person("jack", 21, 170), Person("xzw", 21, 183))
    val ds1 = spark.createDataset(seq1)
 
    //1、map操作，flatmap操作
    ds1.map{x => (x.age + 1, x.name)}.show()
    ds1.flatMap{x =>
      val a = x.age
      val s = x.name.split("").map{x => (a, x)}
      s
    }.show()
 
    //2、filter操作，where操作
    ds1.filter("age >= 25 and height >= 170").show()
    ds1.filter($"age" >= 25 && $"height" >= 170).show()
    ds1.filter{x => x.age >= 25 && x.height >= 170}.show()
    ds1.where("age >= 25 and height >= 170").show()
    ds1.where($"age" >= 25 && $"height" >= 170).show()
 
    //3、去重操作
    ds1.distinct().show()
    ds1.dropDuplicates("age").show()
    ds1.dropDuplicates("age", "height").show()
    ds1.dropDuplicates(Seq("age", "height")).show()
    ds1.dropDuplicates(Array("age", "height")).show()
 
    //4、加法減法操作
    val seq2 = Seq(Person("leo", 18, 183), Person("jack", 18, 175), Person("xzw", 22, 183), Person("lzq", 23, 175))
    val ds2 = spark.createDataset(seq2)
    val seq3 = Seq(Person("leo", 19, 183), Person("jack", 18, 175), Person("xzw", 22, 170), Person("lzq", 23, 175))
    val ds3 = spark.createDataset(seq3)
    ds3.union(ds2).show()  //並集
    ds3.except(ds2).show()  // 差集
    ds3.intersect(ds2).show()  //交集
 
    //5、select操作
    ds2.select("name", "age").show()
    ds2.select(expr("height + 1").as[Int].as("height")).show()
 
    //6、排序操作
    ds2.sort("age").show()  //默認升序排序
    ds2.sort($"age".desc, $"height".desc).show()
    ds2.orderBy("age").show()  //默認升序排序
    ds2.orderBy($"age".desc, $"height".desc).show()
 
    //7、分割抽樣操作
    val ds4 = ds3.union(ds2)
    val rands = ds4.randomSplit(Array(0.3, 0.7))
    println(rands(0).count())
    println(rands(1).count())
    rands(0).show()
    rands(1).show()
    val ds5 = ds4.sample(true, 0.5)
    println(ds5.count())
    ds5.show()
 
    //8、列操作
    val ds6 = ds4.drop("height")
    println(ds6.columns)
    ds6.show()
    val ds7 = ds4.withColumn("add", $"age" + 2)
    println(ds7.columns)
    ds7.show()
    val ds8 = ds7.withColumnRenamed("add", "age_new")
    println(ds8.columns)
    ds8.show()
    ds4.withColumn("add_col", lit(1)).show()
 
    //9、join操作
    val seq4 = Seq(Score("leo", 85), Score("jack", 63), Score("wjl", 70), Score("zyn", 90))
    val ds9 = spark.createDataset(seq4)
    val ds10 = ds2.join(ds9, Seq("name"), "inner")
    ds10.show()
    val ds11 = ds2.join(ds9, Seq("name"), "left")
    ds11.show()
 
    //10、分組聚合操作
    val ds12 = ds4.groupBy("height").agg(avg("age").as("avg_age"))
    ds12.show()
  }
}

七、DataSet 的內置函數

八、例子：WordCount

package sparkml
 
import org.apache.spark.sql.SparkSession
 
object WordCount {
  def main(args: Array[String]): Unit = {
    //創建SparkSession
    val spark = SparkSession.builder()
      .appName("Dataset")
      .master("local[*]")
      .getOrCreate()
 
    import spark.implicits._
    val data = spark.read.textFile("C://xzw//wordcount")
      .flatMap(_.split(" "))
      .map(_.toLowerCase())
      .filter($"value"=!="," && $"value"=!="." && $"value"=!="not")
    data.groupBy($"value").count().sort($"count".desc).show(50)
  }
 
}