如何过滤一个spark数据框与另一个数据框

Question

我试图过滤一个数据帧对另一个:

scala> val df1 = sc.parallelize((1 to 100).map(a=>(s"user $a", a*0.123, a))).toDF("name", "score", "user_id")
scala> val df2 = sc.parallelize(List(2,3,4,5,6)).toDF("valid_id")

现在我想过滤df1，并得到一个数据框，其中包含df1中user_id在df2("valid_id")中的所有行。换句话说，我需要df1中user_id为2 3 4 5或6的所有行

scala> df1.select("user_id").filter($"user_id" in df2("valid_id"))
warning: there were 1 deprecation warning(s); re-run with -deprecation for details
org.apache.spark.sql.AnalysisException: resolved attribute(s) valid_id#20 missing from user_id#18 in operator !Filter user_id#18 IN (valid_id#20);  

另一方面，当我尝试对一个函数做一个过滤器时，一切看起来都很好:

scala> df1.select("user_id").filter(($"user_id" % 2) === 0)
res1: org.apache.spark.sql.DataFrame = [user_id: int]

为什么我得到这个错误?我的语法有问题吗?

下面的注释我已经尝试做一个左外连接:

scala> df1.show
+-------+------------------+-------+
|   name|             score|user_id|
+-------+------------------+-------+
| user 1|             0.123|      1|
| user 2|             0.246|      2|
| user 3|             0.369|      3|
| user 4|             0.492|      4|
| user 5|             0.615|      5|
| user 6|             0.738|      6|
| user 7|             0.861|      7|
| user 8|             0.984|      8|
| user 9|             1.107|      9|
|user 10|              1.23|     10|
|user 11|             1.353|     11|
|user 12|             1.476|     12|
|user 13|             1.599|     13|
|user 14|             1.722|     14|
|user 15|             1.845|     15|
|user 16|             1.968|     16|
|user 17|             2.091|     17|
|user 18|             2.214|     18|
|user 19|2.3369999999999997|     19|
|user 20|              2.46|     20|
+-------+------------------+-------+
only showing top 20 rows
scala> df2.show
+--------+
|valid_id|
+--------+
|       2|
|       3|
|       4|
|       5|
|       6|
+--------+
scala> df1.join(df2, df1("user_id") === df2("valid_id"))
res6: org.apache.spark.sql.DataFrame = [name: string, score: double, user_id: int, valid_id: int]
scala> res6.collect
res7: Array[org.apache.spark.sql.Row] = Array()
scala> df1.join(df2, df1("user_id") === df2("valid_id"), "left_outer")
res8: org.apache.spark.sql.DataFrame = [name: string, score: double, user_id: int, valid_id: int]
scala> res8.count
res9: Long = 0

我正在运行spark 1.5.0和scala 2.10.5

Answer 1

你想要一个(常规的)内部连接，而不是外部连接:)

df1.join(df2, df1("user_id") === df2("valid_id"))

Answer 2

您也可以这样编写代码
连接类型如INNER, LEFT_OUTER, RIGHT_OUTER等

df1.join(df2, col("user_id") === col("valid_id"), "${type_of_join}")