Question

我有一个特殊的用例，其中我为同一位客户有多行，每行对象看起来像：

root
 -c1: BigInt
 -c2: String
 -c3: Double
 -c4: Double
 -c5: Map[String, Int]

现在我按列c1进行分组，并为同一客户收集所有行作为列表，例如：

c1, [Row1, Row3, Row4]
c2, [Row2, Row5]

我尝试过这种方式 dataset.withColumn("combined", array("c1","c2","c3","c4","c5")).groupBy("c1").agg(collect_list("combined"))，但出现异常：

Exception in thread "main" org.apache.spark.sql.AnalysisException: cannot resolve 'array(`c1`, `c2`, `c3`, `c4`, `c5`)' due to data type mismatch: input to function array should all be the same type, but it's [bigint, string, double, double, map<string,map<string,double>>];;

Answer 1

您可以使用array函数代替struct来合并列，并使用groupBy和collect_list聚合函数作为

import org.apache.spark.sql.functions._
df.withColumn("combined", struct("c1","c2","c3","c4","c5"))
    .groupBy("c1").agg(collect_list("combined").as("combined_list"))
    .show(false)

，这样您就可以将schema作为

的分组数据集

root
 |-- c1: integer (nullable = false)
 |-- combined_list: array (nullable = true)
 |    |-- element: struct (containsNull = true)
 |    |    |-- c1: integer (nullable = false)
 |    |    |-- c2: string (nullable = true)
 |    |    |-- c3: string (nullable = true)
 |    |    |-- c4: string (nullable = true)
 |    |    |-- c5: map (nullable = true)
 |    |    |    |-- key: string
 |    |    |    |-- value: integer (valueContainsNull = false)

我希望答案会有所帮助

Answer 2

如果您希望结果包含Row的集合，请考虑如下转换为RDD：

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

def df = Seq(
    (BigInt(10), "x", 1.0, 2.0, Map("a"->1, "b"->2)),
    (BigInt(10), "y", 3.0, 4.0, Map("c"->3)),
    (BigInt(20), "z", 5.0, 6.0, Map("d"->4, "e"->5))
  ).
  toDF("c1", "c2", "c3", "c4", "c5").
  // as[(BigInt, String, Double, Double, Map[String, Int])]

df.rdd.map(r => (r.getDecimal(0), r)).groupByKey.collect
// res1: Array[(java.math.BigDecimal, Iterable[org.apache.spark.sql.Row])] = Array(
//   (10,CompactBuffer([10,x,1.0,2.0,Map(a -> 1, b -> 2)], [10,y,3.0,4.0,Map(c -> 3)])),
//   (20,CompactBuffer([20,z,5.0,6.0,Map(d -> 4, e -> 5)]))
// )

或者，如果您对DataFrame中的struct型行的集合比较满意，则可以采用另一种方法：

val cols = ds.columns

df.groupBy("c1").agg(collect_list(struct(cols.head, cols.tail: _*)).as("row_list")).
  show(false)
// +---+----------------------------------------------------------------+
// |c1 |row_list                                                        |
// +---+----------------------------------------------------------------+
// |20 |[[20,z,5.0,6.0,Map(d -> 4, e -> 5)]]                            |
// |10 |[[10,x,1.0,2.0,Map(a -> 1, b -> 2)], [10,y,3.0,4.0,Map(c -> 3)]]|
// +---+----------------------------------------------------------------+

用apache spark按组收集行

2 个答案: