如何使用 Java 在 Apache spark 中将一行数组平面映射为多行?

How do I flatMap a row of arrays into multiple rows in Apache spark using Java?

我有一个 json 数据文件,其中包含一个 属性,它是 "tags" 的字符串数组。 Apache Spark DataFrame 架构如下所示:

root
 |-- acceptedAnswerId: long (nullable = true)
 |-- answerCount: long (nullable = true)
 |-- body: string (nullable = true)
 |-- score: long (nullable = true)
 |-- <b>tags</b>: array (nullable = true)
 |    |-- element: string (containsNull = true)
 |-- title: string (nullable = true)
 |-- viewCount: long (nullable = true)

我想在 Java 中将每一行分解成几行。我可以使用 找到类似的答案,但无法在 Java 中转换解决方案。有什么建议吗?

JSON 中的 "tags" 属性 看起来像:

"tags":["c#",".net","compression","decompression"]

您可以简单地使用 explode 功能。

DataFrame df = ...
DataFrame expanded = df.withColumn(
  "tag", org.apache.spark.sql.functions.explode(df.col("tags"))).drop("tags");

为了使解决方案更直观,示例 json 数据如下所示:

{"id":4,"score":358,"viewCount":24247,"answerCount":13,"commentCount":1,"favoriteCount":28,"tags":["c#","winforms","type-conversion","opacity"]}

下面是将 json 数据读取为“DataFrame”对象的 Java 代码片段:

JavaSparkContext sc = new JavaSparkContext(conf);
SQLContext sqlContext = new SQLContext(sc);

String jsonData = "{\"id\":4,\"score\":358,\"viewCount\":24247,\"tags\":[\"c#\",\"winforms\",\"type-conversion\",\"opacity\"]}";    
    
List dataSet = Arrays.asList(jsonData);

JavaRDD distData = sc.parallelize(dataSet);

DataFrame Whosebug_Posts = sqlContext.read().json(distData);

Whosebug_Posts.printSchema(); //let's print out the DataFrame schema (Output#1)

Whosebug_Posts.show(); //let's show the DataFrame content (Ouput#2)

架构:Output#1 如下所示:

root
 |-- id: long (nullable = true)
 |-- score: long (nullable = true)
 |-- tags: array (nullable = true)
 |    |-- element: string (containsNull = true)
 |-- viewCount: long (nullable = true)

数据:Output#2 如下所示:

+---+-----+--------------------+---------+
| id|score|                tags|viewCount|
+---+-----+--------------------+---------+
|  4|  358|[c#, winforms, ty...|    24247|
+---+-----+--------------------+---------+

根据zero323的信息,我继续处理:

DataFrame expanded = Whosebug_Posts.withColumn("tag", org.apache.spark.sql.functions.explode(Whosebug_Posts.col("tags")));
        
expanded.printSchema(); //let's print out the DataFrame schema again (Output#3)
        
expanded.show(); //let's show the DataFrame content (Output#4)

架构:Output#3 如下所示:

root
 |-- id: long (nullable = true)
 |-- score: long (nullable = true)
 |-- tags: array (nullable = true)
 |    |-- element: string (containsNull = true)
 |-- viewCount: long (nullable = true)
 |-- tag: string (nullable = true)

以及爆炸后的数据:Output#4

+---+-----+--------------------+---------+---------------+
| id|score|                tags|viewCount|            tag|
+---+-----+--------------------+---------+---------------+
|  4|  358|[c#, winforms, ty...|    24247|             c#|
|  4|  358|[c#, winforms, ty...|    24247|       winforms|
|  4|  358|[c#, winforms, ty...|    24247|type-conversion|
|  4|  358|[c#, winforms, ty...|    24247|        opacity|
+---+-----+--------------------+---------+---------------+

结果看起来与使用 SQL 连接两个 table 非常相似。