PySpark查找另一列中是否存在一列中的模式

Posted 2021-05-07

我有两个pyspark数据帧。一个包含FullAddress字段（比如col1），另一个数据框包含其中一个列中的city / town / suburb的名称（比如col2）。我想比较col2和col1，如果有匹配则返回col2。

此外，郊区名称可以是郊区名称列表。

包含完整地址的Dataframe1

+--------+--------+----------------------------------------------------------+
|Postcode|District|City/ Town/ Suburb                                        |
+--------+--------+----------------------------------------------------------+
|2000    |Sydney  |Dawes Point, Haymarket, Millers Point, Sydney, The Rocks  |
|2001    |Sydney  |Sydney                                                    |
|2113    |Sydney  |North Ryde                                                |
+--------+--------+----------------------------------------------------------+



+-----------------------------------------------------------+
|FullAddress                                                |
+-----------------------------------------------------------+
|BADAJOZ ROAD NORTH RYDE 2113, NSW, Australia               |
| HAY STREET HAYMARKET 2000, NSW, Australia                 |
| SMART STREET FAIRFIELD 2165, NSW, Australia               |
|CLARENCE STREET SYDNEY 2000, NSW, Australia                |
+-----------------------------------------------------------+

我想要这样的东西

+-----------------------------------------------------------++-----------+
|FullAddress                                                |suburb      |
+-----------------------------------------------------------++-----------+
|BADAJOZ ROAD NORTH RYDE 2113, NSW, Australia               |NORTH RYDE  |
| HAY STREET HAYMARKET 2000, NSW, Australia                 |HAYMARKET   |
| SMART STREET FAIRFIELD 2165, NSW, Australia               |NULL        |
|CLARENCE STREET SYDNEY 2000, NSW, Australia                |SYDNEY      |
+-----------------------------------------------------------++-----------+

答案

有两个DataFrames -

DataFrame 1：包含完整地址的DataFrame。

DataFrame 2：包含基础数据的DataFrame - Postcode，District和City / Town / Suburb。

问题的目的是从suburb中为DataFrame 1提取适当的DataFrame 2。尽管OP尚未明确指定我们可以加入两个DataFrame的key，但Postcode似乎只是合理的选择。

# Importing requisite functions
from pyspark.sql.functions import col,regexp_extract,split,udf
from pyspark.sql.types import StringType

让我们创建DataFrame 1作为df。在这个DataFrame我们需要提取Postcode。在澳大利亚，所有邮政编码都是4 digit long，因此我们使用regexp_extract()从string列中提取4位数字。

df = sqlContext.createDataFrame([('BADAJOZ ROAD NORTH RYDE 2113, NSW, Australia ',),
                                 ('HAY STREET HAYMARKET 2000, NSW, Australia',),
                                 ('SMART STREET FAIRFIELD 2165, NSW, Australia',),
                                 ('CLARENCE STREET SYDNEY 2000, NSW, Australia',)],
                                 ('FullAddress',))
df = df.withColumn('Postcode', regexp_extract('FullAddress', "(\d{4})" , 1 ))
df.show(truncate=False)
+---------------------------------------------+--------+
|FullAddress                                  |Postcode|
+---------------------------------------------+--------+
|BADAJOZ ROAD NORTH RYDE 2113, NSW, Australia |2113    |
|HAY STREET HAYMARKET 2000, NSW, Australia    |2000    |
|SMART STREET FAIRFIELD 2165, NSW, Australia  |2165    |
|CLARENCE STREET SYDNEY 2000, NSW, Australia  |2000    |
+---------------------------------------------+--------+

现在，我们已经提取了Postcode，我们创造了key加入两个DataFrames。让我们创建DataFrame 2，我们需要从中提取各自的suburb。

df_City_Town_Suburb = sqlContext.createDataFrame([(2000,'Sydney','Dawes Point, Haymarket, Millers Point, Sydney, The Rocks'),
                                             (2001,'Sydney','Sydney'),(2113,'Sydney','North Ryde')],
                                             ('Postcode','District','City_Town_Suburb'))
df_City_Town_Suburb.show(truncate=False)

+--------+--------+--------------------------------------------------------+
|Postcode|District|City_Town_Suburb                                        |
+--------+--------+--------------------------------------------------------+
|2000    |Sydney  |Dawes Point, Haymarket, Millers Point, Sydney, The Rocks|
|2001    |Sydney  |Sydney                                                  |
|2113    |Sydney  |North Ryde                                              |
+--------+--------+--------------------------------------------------------+

加入两个DataFrames与left加入 -

df = df.join(df_City_Town_Suburb.select('Postcode','City_Town_Suburb'), ['Postcode'],how='left')
df.show(truncate=False)
+--------+---------------------------------------------+--------------------------------------------------------+
|Postcode|FullAddress                                  |City_Town_Suburb                                        |
+--------+---------------------------------------------+--------------------------------------------------------+
|2113    |BADAJOZ ROAD NORTH RYDE 2113, NSW, Australia |North Ryde                                              |
|2165    |SMART STREET FAIRFIELD 2165, NSW, Australia  |null                                                    |
|2000    |HAY STREET HAYMARKET 2000, NSW, Australia    |Dawes Point, Haymarket, Millers Point, Sydney, The Rocks|
|2000    |CLARENCE STREET SYDNEY 2000, NSW, Australia  |Dawes Point, Haymarket, Millers Point, Sydney, The Rocks|
+--------+---------------------------------------------+--------------------------------------------------------+

使用City_Town_Suburb函数将列split()拆分为数组 -

df = df.select('Postcode','FullAddress',split(col("City_Town_Suburb"), ",s*").alias("City_Town_Suburb"))
df.show(truncate=False)
+--------+---------------------------------------------+----------------------------------------------------------+
|Postcode|FullAddress                                  |City_Town_Suburb                                          |
+--------+---------------------------------------------+----------------------------------------------------------+
|2113    |BADAJOZ ROAD NORTH RYDE 2113, NSW, Australia |[North Ryde]                                              |
|2165    |SMART STREET FAIRFIELD 2165, NSW, Australia  |null                                                      |
|2000    |HAY STREET HAYMARKET 2000, NSW, Australia    |[Dawes Point, Haymarket, Millers Point, Sydney, The Rocks]|
|2000    |CLARENCE STREET SYDNEY 2000, NSW, Australia  |[Dawes Point, Haymarket, Millers Point, Sydney, The Rocks]|
+--------+---------------------------------------------+----------------------------------------------------------+

最后创建一个UDF来检查数组City_Town_Suburb中的每个元素（如果它存在于FullAddress列中）。如果存在一个，我们立即返回，否则返回None。

def suburb(FullAddress,City_Town_Suburb):
   # Check for the case where there is no Array, otherwise we will get an Error
   if City_Town_Suburb == None:
      return None
   # Checking each and every Array element if it exists in 'FullAddress',
   # and if a match is found, it's immediately returned.
   for sub in City_Town_Suburb:
      if sub.strip().upper() in FullAddress:
         return sub.upper()
   return None
suburb_udf = udf(suburb,StringType())

应用此UDF -

df = df.withColumn('suburb', suburb_udf(col('FullAddress'),col('City_Town_Suburb'))).drop('City_Town_Suburb')
df.show(truncate=False)
+--------+---------------------------------------------+----------+
|Postcode|FullAddress                                  |suburb    |
+--------+---------------------------------------------+----------+
|2113    |BADAJOZ ROAD NORTH RYDE 2113, NSW, Australia |NORTH RYDE|
|2165    |SMART STREET FAIRFIELD 2165, NSW, Australia  |null      |
|2000    |HAY STREET HAYMARKET 2000, NSW, Australia    |HAYMARKET |
|2000    |CLARENCE STREET SYDNEY 2000, NSW, Australia  |SYDNEY    |
+--------+---------------------------------------------+----------+