问题描述
case class Foo1(codes:Seq[String], name:String)
case class Foo2(code:String, description:String)
val ds1 = Seq(
Foo1(Seq("A"), "foo1"),
Foo1(Seq("A", "B"), "foo2"),
Foo1(Seq("B", "C", "D"), "foo3"),
Foo1(Seq("C"), "foo4"),
Foo1(Seq("C", "D"), "foo5")
).toDS
val ds2 = Seq(
Foo2("A", "product A"),
Foo2("B", "product B"),
Foo2("C", "product C"),
Foo2("D", "product D"),
Foo2("E", "product E")
).toDS
val j = ds1.join(ds2, ds2("code") isin (ds1("codes")))
希望这个 Scala 代码片段清楚地说明了我要完成的任务，我们的数据是结构化的，因此一个数据集有一个包含值数组的列，我希望将该集合中的值连接到另一个数据集.例如，ds1 中的 Seq("A", "B") 将与 "A" 和 "B" 在 ds2 中.
Hopefully this Scala code fragment makes it clear what I'm trying to accomplish, our data is structured so that one data set has a column which contains an array of values, and I wish to join the values within that collection to another data set. So for example Seq("A", "B") in ds1 would join with "A" and "B" in ds2.
Column 上的isin"操作符似乎正是我想要的，它构建并运行，但是当我运行它时，我收到以下错误:
The "isin" operator on Column seems to be exactly what I want, and this builds and runs, but when I run it I get the following error:
org.apache.spark.sql.AnalysisException: 由于数据类型不匹配，无法解析 '(code IN (codes))':参数必须是相同类型；;
进一步阅读我看到 isin() 想要采用可变参数(splatted args")并且似乎更适合 filter().所以我的问题是，这是这个运算符的预期用途，还是有其他方法来执行这种类型的连接?
Reading further I see that isin() wants to take a varargs ("splatted args") and seems more suitable for a filter(). So my question is, is this the intended use of this operator, or is there some other way to perform this type of join?
推荐答案
请使用array_contains:
ds1.crossJoin(ds2).where("array_contains(codes, code)").show
+---------+----+----+-----------+
| codes|name|code|description|
+---------+----+----+-----------+
| [A]|foo1| A| product A|
| [A, B]|foo2| A| product A|
| [A, B]|foo2| B| product B|
|[B, C, D]|foo3| B| product B|
|[B, C, D]|foo3| C| product C|
|[B, C, D]|foo3| D| product D|
| [C]|foo4| C| product C|
| [C, D]|foo5| C| product C|
| [C, D]|foo5| D| product D|
+---------+----+----+-----------+
如果您使用 Spark 1.x 或 2.0，请将 crossJoin 替换为标准连接，并启用交叉连接必要时进行配置.
If you use Spark 1.x or 2.0 replace crossJoin with standard join, and enable cross joins in configuration, if necessary.
使用 explode 可以避免笛卡尔积:
It might by possible to avoid Cartesian product with explode:
ds1.withColumn("code", explode($"codes")).join(ds2, Seq("code")).show
+----+---------+----+-----------+
|code| codes|name|description|
+----+---------+----+-----------+
| B| [A, B]|foo2| product B|
| B|[B, C, D]|foo3| product B|
| D|[B, C, D]|foo3| product D|
| D| [C, D]|foo5| product D|
| C|[B, C, D]|foo3| product C|
| C| [C]|foo4| product C|
| C| [C, D]|foo5| product C|
| A| [A]|foo1| product A|
| A| [A, B]|foo2| product A|
+----+---------+----+-----------+
这篇关于如何在连接中将 Column.isin 与数组列一起使用?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持！