使用作为字符串数组的行字段过滤火花数据帧

Question

使用 Spark 1.5 和 Scala 2.10.6

我正在尝试通过作为字符串数组的字段“标签”过滤数据框。查找所有带有“private”标签的行。

val report = df.select("*")
  .where(df("tags").contains("private"))

得到：

线程“主”org.apache.spark.sql.AnalysisException 中的异常： 由于数据类型不匹配，无法解析“包含（标签，私有）”： 参数 1 需要字符串类型，但是，'tags' 是数组 类型。;

过滤方法是否更适合？

更新：

数据来自 cassandra 适配器，但显示我正在尝试做的事情并得到上述错误的最小示例是：

  def testData (sc: SparkContext): DataFrame = {
    val stringRDD = sc.parallelize(Seq("""
      { "name": "ed",
        "tags": ["red", "private"]
      }""",
      """{ "name": "fred",
        "tags": ["public", "blue"]
      }""")
    )
    val sqlContext = new org.apache.spark.sql.SQLContext(sc)
    import sqlContext.implicits._
    sqlContext.read.json(stringRDD)
  }
  def run(sc: SparkContext) {
    val df1 = testData(sc)
    df1.show()
    val report = df1.select("*")
      .where(df1("tags").contains("private"))
    report.show()
  }

更新：标签数组可以是任意长度，“私有”标签可以在任意位置

更新：一种有效的解决方案：UDF

val filterPriv = udf {(tags: mutable.WrappedArray[String]) => tags.contains("private")}
val report = df1.filter(filterPriv(df1("tags")))

Answer 1

我认为如果您使用where(array_contains(...))，它会起作用。这是我的结果：

scala> import org.apache.spark.SparkContext
import org.apache.spark.SparkContext

scala> import org.apache.spark.sql.DataFrame
import org.apache.spark.sql.DataFrame

scala> def testData (sc: SparkContext): DataFrame = {
     |     val stringRDD = sc.parallelize(Seq
     |      ("""{ "name": "ned", "tags": ["blue", "big", "private"] }""",
     |       """{ "name": "albert", "tags": ["private", "lumpy"] }""",
     |       """{ "name": "zed", "tags": ["big", "private", "square"] }""",
     |       """{ "name": "jed", "tags": ["green", "small", "round"] }""",
     |       """{ "name": "ed", "tags": ["red", "private"] }""",
     |       """{ "name": "fred", "tags": ["public", "blue"] }"""))
     |     val sqlContext = new org.apache.spark.sql.SQLContext(sc)
     |     import sqlContext.implicits._
     |     sqlContext.read.json(stringRDD)
     |   }
testData: (sc: org.apache.spark.SparkContext)org.apache.spark.sql.DataFrame

scala>   
     | val df = testData (sc)
df: org.apache.spark.sql.DataFrame = [name: string, tags: array<string>]

scala> val report = df.select ("*").where (array_contains (df("tags"), "private"))
report: org.apache.spark.sql.DataFrame = [name: string, tags: array<string>]

scala> report.show
+------+--------------------+
|  name|                tags|
+------+--------------------+
|   ned|[blue, big, private]|
|albert|    [private, lumpy]|
|   zed|[big, private, sq...|
|    ed|      [red, private]|
+------+--------------------+

请注意，如果你写where(array_contains(df("tags"), "private"))，它可以工作，但如果你写where(df("tags").array_contains("private"))（更直接地类似于你最初写的）它会失败，array_contains is not a member of org.apache.spark.sql.Column。查看Column 的源代码，我看到有一些东西要处理contains（为此构造一个Contains 实例）但不是array_contains。也许这是一个疏忽。

Answer 2

您可以使用 ordinal 来引用 json 数组，例如在你的情况下df("tags")(0)。这是一个工作示例

scala> val stringRDD = sc.parallelize(Seq("""
     |       { "name": "ed",
     |         "tags": ["private"]
     |       }""",
     |       """{ "name": "fred",
     |         "tags": ["public"]
     |       }""")
     |     )
stringRDD: org.apache.spark.rdd.RDD[String] = ParallelCollectionRDD[87] at parallelize at <console>:22

scala> import sqlContext.implicits._
import sqlContext.implicits._

scala> sqlContext.read.json(stringRDD)
res28: org.apache.spark.sql.DataFrame = [name: string, tags: array<string>]

scala> val df=sqlContext.read.json(stringRDD)
df: org.apache.spark.sql.DataFrame = [name: string, tags: array<string>]

scala> df.columns
res29: Array[String] = Array(name, tags)

scala> df.dtypes
res30: Array[(String, String)] = Array((name,StringType), (tags,ArrayType(StringType,true)))

scala> val report = df.select("*").where(df("tags")(0).contains("private"))
report: org.apache.spark.sql.DataFrame = [name: string, tags: array<string>]

scala> report.show
+----+-------------+
|name|         tags|
+----+-------------+
|  ed|List(private)|
+----+-------------+