Question

如何在Pyspark中更改嵌套列的数据类型？对于示例，如何将值的数据类型从字符串更改为int？

参考：how to change a Dataframe column from String type to Double type in pyspark

{
    "x": "12",
    "y": {
        "p": {
            "name": "abc",
            "value": "10"
        },
        "q": {
            "name": "pqr",
            "value": "20"
        }
    }
}

Answer 1

您可以使用

读取json数据

from pyspark import SQLContext

sqlContext = SQLContext(sc)
data_df = sqlContext.read.json("data.json", multiLine = True)

data_df.printSchema()

输出

root
 |-- x: long (nullable = true)
 |-- y: struct (nullable = true)
 |    |-- p: struct (nullable = true)
 |    |    |-- name: string (nullable = true)
 |    |    |-- value: long (nullable = true)
 |    |-- q: struct (nullable = true)
 |    |    |-- name: string (nullable = true)
 |    |    |-- value: long (nullable = true)

现在您可以从y列访问数据

data_df.select("y.p.name")
data_df.select("y.p.value")

输出

abc, 10

好的，解决方案是添加一个具有正确架构的新嵌套列，并删除具有错误架构的列

from pyspark.sql.functions import *
from pyspark.sql import Row

df3 = spark.read.json("data.json", multiLine = True)

# create correct schema from old 
c = df3.schema['y'].jsonValue()
c['name'] = 'z'
c['type']['fields'][0]['type']['fields'][1]['type'] = 'long'
c['type']['fields'][1]['type']['fields'][1]['type'] = 'long'

y_schema = StructType.fromJson(c['type'])

# define a udf to populate the new column. Row are immuatable so you 
# have to build it from start.

def foo(row):
    d = Row.asDict(row)
    y = {}
    y["p"] = {}
    y["p"]["name"] = d["p"]["name"]
    y["p"]["value"] = int(d["p"]["value"])
    y["q"] = {}
    y["q"]["name"] = d["q"]["name"]
    y["q"]["value"] = int(d["p"]["value"])

    return(y)
map_foo = udf(foo, y_schema)

# add the column
df3_new  = df3.withColumn("z", map_foo("y"))

# delete the column
df4 = df3_new.drop("y")


df4.printSchema()

输出

root
 |-- x: long (nullable = true)
 |-- z: struct (nullable = true)
 |    |-- p: struct (nullable = true)
 |    |    |-- name: string (nullable = true)
 |    |    |-- value: long (nullable = true)
 |    |-- q: struct (nullable = true)
 |    |    |-- name: string (nullable = true)
 |    |    |-- value: long (nullable = true)


df4.show()

输出

+---+-------------------+
|  x|                  z|
+---+-------------------+
| 12|[[abc,10],[pqr,10]]|
+---+-------------------+

Answer 2

使用任意变量名称似乎很简单，但这是有问题的，与PEP8相反。在处理数字时，我建议避免使用迭代这些结构的常用名称...即值。

import json

with open('random.json') as json_file:
    data = json.load(json_file)

for k, v in data.items():
    if k == 'y':
        for key, item in v.items():
            item['value'] = float(item['value'])


print(type(data['y']['p']['value']))
print(type(data['y']['q']['value']))
# mac → python3 make_float.py
# <class 'float'>
# <class 'float'>
json_data = json.dumps(data, indent=4, sort_keys=True)
with open('random.json', 'w') as json_file:
    json_file.write(json_data)

Pyspark：更改嵌套列数据类型

2 个答案: