Question

我熟悉Spark和Scala，目前的任务是“求和”这两个数据帧：

+---+--------+-------------------+
|cyl|avg(mpg)|      var_samp(mpg)|
+---+--------+-------------------+
|  8|    15.8| 1.0200000000000014|
|  6|    20.9|0.48999999999999966|
|  4|    33.9|                0.0|
+---+--------+-------------------+

+---+------------------+------------------+
|cyl|          avg(mpg)|     var_samp(mpg)|
+---+------------------+------------------+
|  8|             13.75| 6.746999999999998|
|  6|              21.4|               NaN|
+---+------------------+------------------+

在这种情况下，“键”为cyl，“值”为avg(mpg)和var_samp(mpg)。

这两个的（近似）结果为：

+---+--------+-------------------+
|cyl|avg(mpg)|      var_samp(mpg)|
+---+--------+-------------------+
|  8|   29.55|            7.76712|
|  6|    42.3|0.48999999999999966|
|  4|    33.9|                0.0|
+---+--------+-------------------+

注意如何将NaN视为零，以及某些DataFrame中可能缺少“键”（第二个键中缺少4个键）。

我怀疑reduceByKey是到达这里的方法，但无法正常工作。

到目前为止，这是我的代码：

case class Cars(car: String, mpg: String, cyl: String, disp: String, hp: String,
                drat: String, wt: String, qsec: String, vs: String, am: String, gear: String, carb: String)

object Bootstrapping extends App {
  override def main(args: Array[String]) {
    val conf = new SparkConf().setAppName("Spark and SparkSql").setMaster("local")
    val sc = new SparkContext(conf)

    sc.setLogLevel("WARN")

    // Exploring SparkSQL
    // Initialize an SQLContext
    val sqlContext = new org.apache.spark.sql.SQLContext(sc)
    import sqlContext._
    import sqlContext.implicits._

    // Load a cvs file
    val csv = sc.textFile("mtcars.csv")
    // Create a Spark DataFrame
    val headerAndRows = csv.map(line => line.split(",").map(_.trim))
    val header = headerAndRows.first
    val mtcdata = headerAndRows.filter(_(0) != header(0))
    val mtcars = mtcdata
      .map(p => Cars(p(0), p(1), p(2), p(3), p(4), p(5), p(6), p(7), p(8), p(9), p(10), p(11)))
      .toDF

    // Aggregate data after grouping by columns
    import org.apache.spark.sql.functions._

    mtcars.sort($"cyl").show()
    mtcars.groupBy("cyl").agg(avg("mpg"), var_samp("mpg")).sort($"cyl").show()

    //sample 25% of the population without replacement
    val sampledData = mtcars.sample(false, 0.25)

    //bootstrapping loop
    for (a <- 1 to 5) {

      //get bootstrap sample
      val bootstrapSample = sampledData.sample(true, 1)

      //HERE!!! I WANT TO SAVE THE AGGREGATED SUM OF THE FOLLOWING:
      bootstrapSample.groupBy("cyl").agg(avg("mpg"), var_samp("mpg"))

    }

  }
}

这是我正在使用的数据：Motor Trend Car Road Tests

Answer 1

一种方法是union两个数据框，使用when/otherwise转换NaN，并执行groupBy以汇总列的sum ，如下所示：

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

val df1 = Seq(
  (8, 15.8, 1.0200000000000014),
  (6, 20.9, 0.48999999999999966),
  (4, 33.9, 0.0)
).toDF("cyl", "avg_mpg", "var_samp_mpg")

val df2 = Seq(
  (8, 13.75, 6.746999999999998),
  (6, 21.4, Double.NaN)
).toDF("cyl", "avg_mpg", "var_samp_mpg")

(df1 union df2).
  withColumn("var_samp_mpg", when($"var_samp_mpg".isNaN, 0.0).otherwise($"var_samp_mpg")).
  groupBy("cyl").agg(sum("avg_mpg"), sum("var_samp_mpg")).
  show
// +---+------------+-------------------+
// |cyl|sum(avg_mpg)|  sum(var_samp_mpg)|
// +---+------------+-------------------+
// |  6|        42.3|0.48999999999999966|
// |  4|        33.9|                0.0|
// |  8|       29.55| 7.7669999999999995|
// +---+------------+-------------------+

如何在Scala中汇总这两个Spark数据帧？

1 个答案: