Python 性能调优：JSON 到 CSV，大文件

Question

一位同事要求我将“Yelp 数据集挑战”中的 6 个大文件从有点“扁平”的常规 JSON 转换为 CSV（他认为它们看起来很有趣的教学数据） .

我想我可以用：

# With thanks to http://www.diveintopython3.net/files.html and https://www.reddit.com/r/MachineLearning/comments/33eglq/python_help_jsoncsv_pandas/cqkwyu8/

import os
import pandas

jsondir = 'c:\\example\\bigfiles\\'
csvdir = 'c:\\example\\bigcsvfiles\\'
if not os.path.exists(csvdir): os.makedirs(csvdir)

for file in os.listdir(jsondir):
    with open(jsondir+file, 'r', encoding='utf-8') as f: data = f.readlines()
    df = pandas.read_json('[' + ','.join(map(lambda x: x.rstrip(), data)) + ']')
    df.to_csv(csvdir+os.path.splitext(file)[0]+'.csv',index=0,quoting=1)

很遗憾，我的计算机内存无法胜任这种大小的文件的任务。 （即使我摆脱了循环，虽然它会在不到一分钟的时间内敲出一个 50MB 的文件，但它很难避免冻结我的计算机或在 100MB 以上的文件上崩溃，最大的文件是 3.25GB。）

我可以运行其他简单但高效的东西吗？

在循环中会很棒，但如果对内存有影响，我也可以使用单独的文件名运行 6 次（只有 6 个文件）。

这里是一个“.json”文件内容的例子——注意每个文件实际上有很多 JSON 对象，每行 1 个。

{"business_id":"xyzzy","name":"Business A","neighborhood":"","address":"XX YY ZZ","city":"Tempe","state":"AZ","postal_code":"85283","latitude":33.32823894longitude":-111.28948,"stars":3,"review_count":3,"is_open":0,"attributes":["BikeParking: True","BusinessAcceptsBitcoin: False","BusinessAcceptsCreditCards: True","BusinessParking: {'garage': False, 'street': False, 'validated': False, 'lot': True, 'valet': False}","DogsAllowed: False","RestaurantsPriceRange2: 2","WheelchairAccessible: True"],"categories":["Tobacco Shops","Nightlife","Vape Shops","Shopping"],"hours":["Monday 11:0-21:0","Tuesday 11:0-21:0","Wednesday 11:0-21:0","Thursday 11:0-21:0","Friday 11:0-22:0","Saturday 10:0-22:0","Sunday 11:0-18:0"],"type":"business"}
{"business_id":"dsfiuweio2f","name":"Some Place","neighborhood":"","address":"Strip or something","city":"Las Vegas","state":"NV","postal_code":"89106","latitude":36.189134,"longitude":-115.92094,"stars":1.5,"review_count":2,"is_open":1,"attributes":["BusinessAcceptsBitcoin: False","BusinessAcceptsCreditCards: True"],"categories":["Caterers","Grocery","Food","Event Planning & Services","Party & Event Planning","Specialty Food"],"hours":["Monday 0:0-0:0","Tuesday 0:0-0:0","Wednesday 0:0-0:0","Thursday 0:0-0:0","Friday 0:0-0:0","Saturday 0:0-0:0","Sunday 0:0-0:0"],"type":"business"}
{"business_id":"abccb","name":"La la la","neighborhood":"Blah blah","address":"Yay that","city":"Toronto","state":"ON","postal_code":"M6H 1L5","latitude":43.283984,"longitude":-79.28284,"stars":2,"review_count":6,"is_open":1,"attributes":["Alcohol: none","Ambience: {'romantic': False, 'intimate': False, 'classy': False, 'hipster': False, 'touristy': False, 'trendy': False, 'upscale': False, 'casual': False}","BikeParking: True","BusinessAcceptsCreditCards: True","BusinessParking: {'garage': False, 'street': False, 'validated': False, 'lot': False, 'valet': False}","Caters: True","GoodForKids: True","GoodForMeal: {'dessert': False, 'latenight': False, 'lunch': False, 'dinner': False, 'breakfast': False, 'brunch': False}","HasTV: True","NoiseLevel: quiet","OutdoorSeating: False","RestaurantsAttire: casual","RestaurantsDelivery: True","RestaurantsGoodForGroups: True","RestaurantsPriceRange2: 1","RestaurantsReservations: False","RestaurantsTableService: False","RestaurantsTakeOut: True","WiFi: free"],"categories":["Restaurants","Pizza","Chicken Wings","Italian"],"hours":["Monday 11:0-2:0","Tuesday 11:0-2:0","Wednesday 11:0-2:0","Thursday 11:0-3:0","Friday 11:0-3:0","Saturday 11:0-3:0","Sunday 11:0-2:0"],"type":"business"}

嵌套的 JSON 数据可以简单地保留为表示它的字符串文字——我只是希望将顶级键转换为 CSV 文件标题。

Answer 1

问题在于您的代码将整个文件读入内存，然后在内存中创建它的近似副本。我怀疑它还会创建第三个副本，但尚未验证。 Neo X 建议的解决方案是逐行读取文件并进行相应处理。这是 for 循环的替换：

for file in os.listdir(jsondir):
    csv_file = csvdir + os.path.splitext(file)[0] + '.csv'
    with open(jsondir+file, 'r', encoding='utf-8') as f, open(csv_file, 'w', encoding='utf-8') as csv:
        header = True
        for line in f:
            df = pandas.read_json(''.join(('[', line.rstrip(), ']')))
            df.to_csv(csv, header=header, index=0, quoting=1)
            header = False

我在 Mac 上使用 python 3.5 对此进行了测试；它应该可以在 Windows 上运行，但我还没有在那里测试过。

注意事项：

1. 我已经调整了你的 json 数据；第一行的纬度/经度似乎有错误。

2. 这只是用一个小文件测试过；我不确定从哪里获取 3.5 GB 文件。

3. 我假设这是您朋友的一次性使用。如果这是生产代码，您需要验证“with”语句的异常处理是否正确。详情请见How can I open multiple files using "with open" in Python?。

4. 这应该是相当高效的，但同样，我不确定从哪里获取大文件。