解决这个问题的简单方法是:

创建一个新的Python文件，如:json_to_csv.py

添加以下代码:

import csv, json, sys
#if you are not using utf-8 files, remove the next line
sys.setdefaultencoding("UTF-8")
#check if you pass the input file and output file
if sys.argv[1] is not None and sys.argv[2] is not None:

    fileInput = sys.argv[1]
    fileOutput = sys.argv[2]

    inputFile = open(fileInput)
    outputFile = open(fileOutput, 'w')
    data = json.load(inputFile)
    inputFile.close()

    output = csv.writer(outputFile)

    output.writerow(data[0].keys())  # header row

    for row in data:
        output.writerow(row.values())

添加代码后，保存文件并在终端上运行:

Python json_to_csv.py input.txt output.csv

我希望这对你有帮助。

韩国歌手组合!

这是@MikeRepass回答的修改。此版本将CSV写入文件，适用于Python 2和Python 3。

import csv,json
input_file="data.json"
output_file="data.csv"
with open(input_file) as f:
    content=json.load(f)
try:
    context=open(output_file,'w',newline='') # Python 3
except TypeError:
    context=open(output_file,'wb') # Python 2
with context as file:
    writer=csv.writer(file)
    writer.writerow(content[0].keys()) # header row
    for row in content:
        writer.writerow(row.values())

不幸的是，我没有足够的声誉来为@Alec McGail的惊人回答做出小小的贡献。 我正在使用Python3，我需要将映射转换为@Alexis R注释后面的列表。

另外，我发现csv作者添加了一个额外的CR文件(我有一个空行每一行与数据在csv文件)。根据@Jason R. Coombs对这个帖子的回答，解决方法非常简单: CSV在Python中添加了一个额外的回车

您只需将lineterminator='\n'参数添加到csv.writer。它将是:csv_w = csv。Writer (out_file, lineterminator='\n')

使用pandas中的json_normalize:

在名为test.json的文件中使用来自OP的示例数据。 这里使用了Encoding ='utf-8'，但在其他情况下可能不需要。 下面的代码利用了pathlib库。 .open是pathlib的一个方法。 也适用于非windows路径。 使用pandas.to_csv(…)将数据保存为csv文件。

import pandas as pd
# As of Pandas 1.01, json_normalize as pandas.io.json.json_normalize is deprecated and is now exposed in the top-level namespace.
# from pandas.io.json import json_normalize
from pathlib import Path
import json

# set path to file
p = Path(r'c:\some_path_to_file\test.json')

# read json
with p.open('r', encoding='utf-8') as f:
    data = json.loads(f.read())

# create dataframe
df = pd.json_normalize(data)

# dataframe view
 pk            model  fields.codename           fields.name  fields.content_type
 22  auth.permission     add_logentry     Can add log entry                    8
 23  auth.permission  change_logentry  Can change log entry                    8
 24  auth.permission  delete_logentry  Can delete log entry                    8
  4  auth.permission        add_group         Can add group                    2
 10  auth.permission      add_message       Can add message                    4

# save to csv
df.to_csv('test.csv', index=False, encoding='utf-8')

CSV输出:

pk,model,fields.codename,fields.name,fields.content_type
22,auth.permission,add_logentry,Can add log entry,8
23,auth.permission,change_logentry,Can change log entry,8
24,auth.permission,delete_logentry,Can delete log entry,8
4,auth.permission,add_group,Can add group,2
10,auth.permission,add_message,Can add message,4

嵌套更重的JSON对象的资源:

所以答案: 用python平化JSON数组 如何平嵌套的JSON递归，与平坦JSON 如何json_normalize一个列与nan 使用pandas将一列字典拆分为单独的列 有关其他相关问题，请参阅json_normalize标记。

您可以使用此代码将json文件转换为csv文件 读取文件后，我将对象转换为熊猫数据框架，然后将其保存为CSV文件

import os
import pandas as pd
import json
import numpy as np

data = []
os.chdir('D:\\Your_directory\\folder')
with open('file_name.json', encoding="utf8") as data_file:    
     for line in data_file:
        data.append(json.loads(line))

dataframe = pd.DataFrame(data)        
## Saving the dataframe to a csv file
dataframe.to_csv("filename.csv", encoding='utf-8',index= False)

JSON可以表示各种各样的数据结构——JS的“对象”大致类似于Python的dict(带有字符串键)，JS的“数组”大致类似于Python列表，只要最后的“叶子”元素是数字或字符串，你就可以嵌套它们。

CSV本质上只能表示一个2-D表——可选的第一行是“标题”，即“列名”，这可以使表可解释为字典列表，而不是正常的解释，一个列表的列表(同样，“叶子”元素可以是数字或字符串)。

So, in the general case, you can't translate an arbitrary JSON structure to a CSV. In a few special cases you can (array of arrays with no further nesting; arrays of objects which all have exactly the same keys). Which special case, if any, applies to your problem? The details of the solution depend on which special case you do have. Given the astonishing fact that you don't even mention which one applies, I suspect you may not have considered the constraint, neither usable case in fact applies, and your problem is impossible to solve. But please do clarify!