Spark-将包含JSON字符串的coulmn从StringType转换为ArrayType（StringType（））

Question

我有一个包含json字符串的数据框架df，如下所示，

'''[{"@id":"Party_1","@ObjectID":"Policy_1"},{"@id":"Party_2","@ObjectID":"Policy_2"},{"@id":"Party_3","@ObjectID":"Policy_3"}]'''

df 架构：

root
 |-- col1: string (nullable = true)

如何将其转换为字符串数组（数组类型（字符串类型（））？

结果应该是这样的，

['{"@id":"Party_1","@OriginatingObjectID":"Policy_1"}',
 '{"@id":"Party_2","@OriginatingObjectID":"Policy_2"}',
 '{"@id":"Party_3","@OriginatingObjectID":"Policy_3"}']

结果模式：

root
 |-- arr_col: array (nullable = true)
 |          |-- element: string (containsNull = true)

任何帮助都将不胜感激。谢谢你！

Answer 1

您可以使用from_json函数获取json字段，只需对值进行轻微修改，如下所示

data = [
    ('[{"@id":"Party_1","@ObjectID":"Policy_1"},{"@id":"Party_2","@ObjectID":"Policy_2"},{"@id":"Party_3","@ObjectID":"Policy_3"}]', 2767),
    ('[{"@id":"Party_1","@ObjectID":"Policy_1"},{"@id":"Party_2","@ObjectID":"Policy_2"},{"@id":"Party_3","@ObjectID":"Policy_3"}]', 4235)
]

df = spark.createDataFrame(data).toDF(*["value", "count"])\
    .withColumn("value", f.regexp_replace(f.col("value"), "\\[\\{", "{\"arr\": [{"))\
    .withColumn("value", f.regexp_replace(f.col("value"), "\\}\\]", "}]}"))


json_schema = spark.read.json(df.rdd.map(lambda row: row.value)).schema
resultDF = df.select(f.from_json("value", 
schema=json_schema).alias("array_col"))\
    .select("array_col.*")

resultDF.printSchema()
resultDF.show(truncate=False)

或者，如果您希望嵌套json作为字符串，您可以使用自定义模式。

输出架构：

root
 |-- arr: array (nullable = true)
 |    |-- element: struct (containsNull = true)
 |    |    |-- @ObjectID: string (nullable = true)
 |    |    |-- @id: string (nullable = true)

输出:

+---------------------------------------------------------------+
|arr                                                            |
+---------------------------------------------------------------+
|[{Policy_1, Party_1}, {Policy_2, Party_2}, {Policy_3, Party_3}]|
|[{Policy_1, Party_1}, {Policy_2, Party_2}, {Policy_3, Party_3}]|
+---------------------------------------------------------------+