處理巢狀 JSON

使用 pandas 的精實資料導入

Amany Mahfouz

Instructor

巢狀 JSON

JSON 由屬性—值配對的物件組成
當值本身是物件時，JSON 就是巢狀的

來自 Yelp Business API 文件的範例 JSON 回應

Yelp 回應資料範例，標示出巢狀的座標與位置資料

Yelp 回應資料範例，標示出巢狀的分類資料

Yelp 回應資料範例，標示 businesses 下的巢狀紀錄

# Print columns containing nested data
print(bookstores[["categories", "coordinates", "location"]].head(3))

                                          categories  \
0   [{'alias': 'bookstores', 'title': 'Bookstores'}]   
1  [{'alias': 'bookstores', 'title': 'Bookstores'...   
2   [{'alias': 'bookstores', 'title': 'Bookstores'}]     

                                         coordinates  \
0  {'latitude': 37.7975997924805, 'longitude': -1...   
1  {'latitude': 37.7885846793652, 'longitude': -1...   
2  {'latitude': 37.7589836120605, 'longitude': -1...    

                                            location  
0  {'address1': '261 Columbus Ave', 'address2': '...  
1  {'address1': '50 2nd St', 'address2': '', 'add...  
2  {'address1': '866 Valencia St', 'address2': ''...

pandas.io.json

pandas.io.json 子模組提供讀寫 JSON 的工具
- 需要各自的 import 陳述式
json_normalize()
- 接受字典／字典清單（類似 pd.DataFrame()）
- 回傳攤平的 dataframe
- 預設欄名樣式：attribute.nestedattribute
- 以 sep 參數指定不同分隔符

載入巢狀 JSON 資料

import pandas as pd
import requests

from pandas.io.json import json_normalize



# Set up headers, parameters, and API endpoint
api_url = "https://api.yelp.com/v3/businesses/search"
headers = {"Authorization": "Bearer {}".format(api_key)}
params = {"term": "bookstore", 
          "location": "San Francisco"}


# Make the API call and extract the JSON data
response = requests.get(api_url, 
                        headers=headers,
                        params=params)
data = response.json()

# Flatten data and load to dataframe, with _ separators
bookstores = json_normalize(data["businesses"], sep="_")
print(list(bookstores))

['alias', 
 'categories',
 'coordinates_latitude',
 'coordinates_longitude',
 ...
 'location_address1',
 'location_address2',
 'location_address3',
 'location_city',
 'location_country',
 'location_display_address',
 'location_state',
 'location_zip_code',
 ...
 'url']

更深層巢狀資料

print(bookstores.categories.head())

0     [{'alias': 'bookstores', 'title': 'Bookstores'}]
1    [{'alias': 'bookstores', 'title': 'Bookstores'...
2     [{'alias': 'bookstores', 'title': 'Bookstores'}]
3     [{'alias': 'bookstores', 'title': 'Bookstores'}]
4    [{'alias': 'bookstores', 'title': 'Bookstores'...
Name: categories, dtype: object

更深層巢狀資料

json_normalize()
- record_path：巢狀資料的屬性字串／字串清單
- meta：一併載入到 dataframe 的其他屬性清單
- meta_prefix：加在 meta 欄名的字首字串

更深層巢狀資料

# Flatten categories data, bring in business details
df = json_normalize(data["businesses"],
                    sep="_",

                    record_path="categories",

                    meta=["name", 
                          "alias", 
                          "rating", 
                          ["coordinates", "latitude"], 
                          ["coordinates", "longitude"]],

                    meta_prefix="biz_")

print(df.head(4))

        alias               title                biz_name  \
0  bookstores          Bookstores   City Lights Bookstore   
1  bookstores          Bookstores  Alexander Book Company   
2  stationery  Cards & Stationery  Alexander Book Company   
3  bookstores          Bookstores       Borderlands Books   

                              biz_alias  biz_rating  biz_coordinates_latitude  \
0   city-lights-bookstore-san-francisco         4.5                 37.797600   
1  alexander-book-company-san-francisco         4.5                 37.788585   
2  alexander-book-company-san-francisco         4.5                 37.788585   
3       borderlands-books-san-francisco         5.0                 37.758984   

   biz_coordinates_longitude  
0                -122.406578  
1                -122.400631  
2                -122.400631  
3                -122.421638

一起來練習吧！

使用 pandas 的精實資料導入