在spark数据框结构中键入更改

时间:2018-05-04 10:13:28

标签: scala apache-spark spark-dataframe

我有以下架构:

root
 |-- Id: long (nullable = true)    
 |-- element: struct (containsNull = true)
 |    |-- Amount: double (nullable = true)
 |    |-- Currency: string (nullable = true)

我想将Amount的类型更改为整数。它不适用于withColumn,因为类型保持不变:

df.withColumn("element.Amount", $"element.Amount".cast(sql.types.IntegerType))

如何更改struct中的列类型?

1 个答案:

答案 0 :(得分:2)

如果您无法在源代码中解决问题,可以投射:

case class Amount(amount: Double, currency: String)
case class Row(id: Long, element: Amount)

val df = Seq(Row(1L, Amount(0.96, "EUR"))).toDF

val dfCasted = df.withColumn(
  "element", $"element".cast("struct<amount: integer, currency: string>")
)

dfCasted.show
// +---+--------+
// | id| element|
// +---+--------+
// |  1|[0, EUR]|
// +---+--------+


dfCasted.printSchema
// root
//  |-- id: long (nullable = false)
//  |-- element: struct (nullable = true)
//  |    |-- amount: integer (nullable = true)
//  |    |-- currency: string (nullable = true)



dfCasted.printSchema

在简单的情况下,您可以尝试重建树:

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

dfCasted.withColumn(
  "element",
  struct($"element.amount".cast("integer"), $"element.currency")
)
// org.apache.spark.sql.DataFrame = [id: bigint, element: struct<col1: int, currency: string>]

但它不适用于复杂的树木。