我有一个如下所示的Spark数据框:
+------+-----+-----+
|acctId|vehId|count|
+------+-----+-----+
| 1| 666| 1|
| 1| 777| 3|
| 1| 888| 2|
| 1| 999| 3|
| 2| 777| 1|
| 2| 888| 3|
| 2| 999| 1|
| 3| 777| 4|
| 3| 888| 2|
+------+-----+-----+
我想将每个acctId的vehId映射到其计数,并将其存储回数据框中,因此最终结果如下所示:
+------+---------------------------------------------+
|acctId| map |
+------+---------------------------------------------+
| 1| Map(666 -> 1, 777 -> 3, 888 -> 2, 999 -> 3) |
| 2| Map(777 -> 1, 888 -> 3, 999 -> 1) |
| 3| Map(777 -> 4, 888 -> 2) |
+------+---------------------------------------------+
最好的方法是什么?
我曾尝试将数据帧转换为RDD并在行上执行映射,但是我不确定如何将每个映射聚合回单个acctId。我一般不熟悉Spark和数据帧,但是已经尽力尝试查找类似的问题,如果这是一个非常常见的问题,我们深表歉意。
供我参考/使用,这是我如何生成测试数据的方法:
val testData = Seq(
(1, 999),
(1, 999),
(2, 999),
(1, 888),
(2, 888),
(3, 888),
(2, 888),
(2, 888),
(1, 888),
(1, 777),
(1, 666),
(3, 888),
(1, 777),
(3, 777),
(2, 777),
(3, 777),
(3, 777),
(1, 999),
(3, 777),
(1, 777)
).toDF("acctId", "vehId")
val grouped = testData.groupBy("acctId", "vehId").count
答案 0 :(得分:2)
我认为您必须如下使用双get_ipython().profile_dir.startup_dir
groupBy
输出:
val testData = Seq(
(1, 999),
(1, 999),
(2, 999),
(1, 888),
(2, 888),
(3, 888),
(2, 888),
(2, 888),
(1, 888),
(1, 777),
(1, 666),
(3, 888),
(1, 777),
(3, 777),
(2, 777),
(3, 777),
(3, 777),
(1, 999),
(3, 777),
(1, 777)
).toDF("acctId", "vehId")
//udf to convert list to map
val listToMap = udf((input: Seq[Row]) => input.map(row => (row.getAs[Int](0), row.getAs[Long](1))).toMap)
val resultDF = testData.groupBy("acctId", "vehId")
.agg(count("acctId").cast("long").as("count"))
.groupBy("acctId")
.agg(collect_list(struct("vehId", "count")) as ("map"))
.withColumn("map", listToMap($"map"))
模式:
resultDF.show(false)
+------+----------------------------------------+
|acctId|map |
+------+----------------------------------------+
|1 |[777 -> 3, 666 -> 1, 999 -> 3, 888 -> 2]|
|3 |[777 -> 4, 888 -> 2] |
|2 |[777 -> 1, 999 -> 1, 888 -> 3] |
+------+----------------------------------------+