如何将 UDF 中的结构或类数组返回到数据帧列值中？

Question

d = [{'ID': '1', 'pID': 1000, 'startTime':'2018.07.02T03:34:20', 'endTime':'2018.07.03T02:40:20'}, {'ID': '1', 'pID': 1000, 'startTime':'2018.07.02T03:45:20', 'endTime':'2018.07.03T02:50:20'}, {'ID': '2', 'pID': 2000, 'startTime':'2018.07.02T03:34:20', 'endTime':'2018.07.03T02:40:20'}, {'ID': '2', 'pID': 2000, 'startTime':'2018.07.02T03:45:20', 'endTime':'2018.07.03T02:50:20'}]

df = spark.createDataFrame(d)

Dates = namedtuple("Dates", "startTime endTime")


def MergeAdjacentUsage(timeSets):
  DatesArray = []
  for times in timeSets:
    DatesArray.append(Dates(startTime=times.startTime, endTime=times.endTime))
  return DatesArray


MergeAdjacentUsages = udf(MergeAdjacentUsage,ArrayType(Dates()))

df1=df.groupBy(['ID','pID']).agg(MergeAdjacentUsages(F.collect_list(struct('startTime','endTime'))).alias("Times"))

display(df1)

我想要的只是将列值设置为 UDF 返回的结构数组。它给我的错误是：

类型错误：new () 恰好需要 3 个参数（给定 1 个）

() 22 return DatesArray 23 ---> 24 MergeAdjacentUsages = udf(MergeAdjacentUsage,ArrayType(Dates())) 25 26 df1=df.groupBy(['ID','pID'] 中的 TypeError Traceback (最近一次调用最后一次) ).agg(MergeAdjacentUsages(F.collect_list(struct('startTime','endTime'))).alias("Times"))

任何帮助、想法或提示将不胜感激。

Answer 1

pyspark 不允许用户定义类对象作为数据框列类型。相反，我们需要创建StructType类似于 python 中的类/命名元组的使用方式。

例如：

from pyspark.sql.types import *
from pyspark.sql.functions import udf
from pyspark.sql import functions as F
# from pyspark.sql.functions import *

d = [{'ID': '1', 'pID': 1000, 'startTime': '2018.07.02T03:34:20', 'endTime': '2018.07.03T02:40:20'},
     {'ID': '1', 'pID': 1000, 'startTime': '2018.07.02T03:45:20', 'endTime': '2018.07.03T02:50:20'},
     {'ID': '2', 'pID': 2000, 'startTime': '2018.07.02T03:34:20', 'endTime': '2018.07.03T02:40:20'},
     {'ID': '2', 'pID': 2000, 'startTime': '2018.07.02T03:45:20', 'endTime': '2018.07.03T02:50:20'}]

df = spark.createDataFrame(d)

# Dates = namedtuple("Dates", "startTime endTime")

schema = ArrayType(StructType([
    StructField("startTime", StringType(), False),
    StructField("endTime", StringType(), False)
]))


MergeAdjacentUsages = udf(lambda xs: xs, schema)

df1 = df.groupBy(['ID', 'pID']).agg(MergeAdjacentUsages(
    F.collect_list(F.struct('startTime', 'endTime'))).alias("Times"))
df1.show(truncate=False)

+---+----+----------------------------------------------------------------------------------------+
|ID |pID |Times                                                                                   |
+---+----+----------------------------------------------------------------------------------------+
|2  |2000|[[2018.07.02T03:34:20, 2018.07.03T02:40:20], [2018.07.02T03:45:20, 2018.07.03T02:50:20]]|
|1  |1000|[[2018.07.02T03:34:20, 2018.07.03T02:40:20], [2018.07.02T03:45:20, 2018.07.03T02:50:20]]|
+---+----+----------------------------------------------------------------------------------------+

希望这可以帮助！