如何在Scala中的Apache Spark中将数据帧转换为数据集？

Question

我需要将我的数据帧转换为数据集,并使用以下代码:

    val final_df = Dataframe.withColumn(
      "features",
      toVec4(
        // casting into Timestamp to parse the string, and then into Int
        $"time_stamp_0".cast(TimestampType).cast(IntegerType),
        $"count",
        $"sender_ip_1",
        $"receiver_ip_2"
      )
    ).withColumn("label", (Dataframe("count"))).select("features", "label")

    final_df.show()

    val trainingTest = final_df.randomSplit(Array(0.3, 0.7))
    val TrainingDF = trainingTest(0)
    val TestingDF=trainingTest(1)
    TrainingDF.show()
    TestingDF.show()

    ///lets create our liner regression
    val lir= new LinearRegression()
    .setRegParam(0.3)
    .setElasticNetParam(0.8)
    .setMaxIter(100)
    .setTol(1E-6)

    case class df_ds(features:Vector, label:Integer)
    org.apache.spark.sql.catalyst.encoders.OuterScopes.addOuterScope(this)

    val Training_ds = TrainingDF.as[df_ds]

我的问题是,我收到以下错误:

Error:(96, 36) Unable to find encoder for type stored in a Dataset.  Primitive types (Int, String, etc) and Product types (case classes) are supported by importing spark.implicits._  Support for serializing other types will be added in future releases.
    val Training_ds = TrainingDF.as[df_ds]

似乎数据框中的值的数量与我的类中的值的数量不同.但是我使用的case class df_ds(features:Vector, label:Integer)是TrainingDF数据帧,因为它有一个特征向量和一个整数标签.这是TrainingDF数据帧:

+--------------------+-----+
|            features|label|
+--------------------+-----+
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,19...|   19|
|[1.497325796E9,10...|   10|
+--------------------+-----+

这是我原来的final_df数据帧:

+------------+-----------+-------------+-----+
|time_stamp_0|sender_ip_1|receiver_ip_2|count|
+------------+-----------+-------------+-----+
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.2|     10.0.0.3|   19|
|    05:49:56|   10.0.0.3|     10.0.0.2|   10|
+------------+-----------+-------------+-----+

但是我得到了上面提到的错误!有谁能够帮我？提前致谢.

Answer 1

您正在阅读的错误消息是一个非常好的指针.

当你将a转换DataFrame为a时,Dataset你必须拥有适合Encoder存储DataFrame在行中的任何东西.

类似原始类型(Ints,Strings等)的编码器,case classes只需导入SparkSession类似的implicits即可:

case class MyData(intField: Int, boolField: Boolean) // e.g.

val spark: SparkSession = ???
val df: DataFrame = ???

import spark.implicits._

val ds: Dataset[MyData] = df.as[MyData]

如果不工作,要么是因为你试图将类型强制转换的DataFrame,以不被支持.在这种情况下,你会写自己Encoder:你可能会发现更多关于它的信息在这里看到一个例子(Encoder为java.time.LocalDateTime)在这里.

Answer 2

火花 1.6.0

case class MyCase(id: Int, name: String)

val encoder = org.apache.spark.sql.catalyst.encoders.ExpressionEncoder[MyCase]

val dataframe = …

val dataset = dataframe.as(encoder)

Spark 2.0 或以上

case class MyCase(id: Int, name: String)

val encoder = org.apache.spark.sql.Encoders.product[MyCase]

val dataframe = …

val dataset = dataframe.as(encoder)