如何在PySpark中创建一个返回字符串数组的udf?

时间:2017-12-06 20:27:32

标签: python apache-spark pyspark apache-spark-sql user-defined-functions

我有一个udf,它返回一个字符串列表。这不应该太难。我在执行udf时传入数据类型,因为它返回一个字符串数组:ArrayType(StringType)

现在,某种方式不起作用:

我正在操作的数据帧是df_subsets_concat,如下所示:

df_subsets_concat.show(3,False)
+----------------------+
|col1                  |
+----------------------+
|oculunt               |
|predistposed          |
|incredulous           |
+----------------------+
only showing top 3 rows

,代码是

from pyspark.sql.types import ArrayType, FloatType, StringType

my_udf = lambda domain: ['s','n']
label_udf = udf(my_udf, ArrayType(StringType))
df_subsets_concat_with_md = df_subsets_concat.withColumn('subset', label_udf(df_subsets_concat.col1))

,结果是

/usr/lib/spark/python/pyspark/sql/types.py in __init__(self, elementType, containsNull)
    288         False
    289         """
--> 290         assert isinstance(elementType, DataType), "elementType should be DataType"
    291         self.elementType = elementType
    292         self.containsNull = containsNull

AssertionError: elementType should be DataType

我的理解是,这是正确的方法。以下是一些资源: pySpark Data Frames "assert isinstance(dataType, DataType), "dataType should be DataType" How to return a "Tuple type" in a UDF in PySpark?

但这些都没有帮助我解决为什么这不起作用。我正在使用pyspark 1.6.1。

如何在pyspark中创建一个返回字符串数组的udf?

1 个答案:

答案 0 :(得分:18)

您需要初始化StringType实例:

label_udf = udf(my_udf, ArrayType(StringType()))
#                                           ^^ 
df.withColumn('subset', label_udf(df.col1)).show()
+------------+------+
|        col1|subset|
+------------+------+
|     oculunt|[s, n]|
|predistposed|[s, n]|
| incredulous|[s, n]|
+------------+------+