Normalize JSON using Python

Question

I am relatively new to JSON and Python, and I am struggling to flatten JSON since last two days. I read the example at http://pandas.pydata.org/pandas-docs/version/0.19/generated/pandas.io.json.json_normalize.html, but I didn't understand how to unlist some nested elements. I also read a few threads Flatten JSON based on an attribute - python How to normalize complex nested json in python? and https://towardsdatascience.com/flattening-json-objects-in-python-f5343c794b10. I tried all without any luck.

Here's the first record of my JSON file:

d = 
{'city': {'url': 'link',
  'name': ['San Francisco']},
 'rank': 1,
 'resident': [
  {'link': ['bit.ly/0842/'], 'name': ['John A']},
  {'link': ['bit.ly/5835/'], 'name': ['Tedd B']},
  {'link': ['bit.ly/2011/'], 'name': ['Cobb C']},
  {'link': ['bit.ly/0855/'], 'name': ['Jack N']},
  {'link': ['bit.ly/1430/'], 'name': ['Jack K']},
  {'link': ['bit.ly/3081/'], 'name': ['Edward']},
  {'link': ['bit.ly/2001/'], 'name': ['Jack W']},
  {'link': ['bit.ly/0020/'], 'name': ['Henry F']},
  {'link': ['bit.ly/2137/'], 'name': ['Joseph S']},
  {'link': ['bit.ly/3225/'], 'name': ['Ed B']},
  {'link': ['bit.ly/3667/'], 'name': ['George Vvec']},
  {'link': ['bit.ly/6434/'], 'name': ['Robert W']},
  {'link': ['bit.ly/4036/'], 'name': ['Rudy B']},
  {'link': ['bit.ly/6450/'], 'name': ['James K']},
  {'link': ['bit.ly/5180/'], 'name': ['Billy N']},
  {'link': ['bit.ly/7847/'], 'name': ['John S']}]
}

Here's the expected output:

city_url  city_name      rank    resident_link   resident_name  
link      San Francisco   1     'bit.ly/0842/'   'John A'
link      San Francisco   1     'bit.ly/5835/'   'Tedd B'
link      San Francisco   1     'bit.ly/2011/'   'Cobb C'
link      San Francisco   1     'bit.ly/0855/'   'Jack N'
link      San Francisco   1     'bit.ly/1430/'   'Jack K'
link      San Francisco   1     'bit.ly/3081/'   'Edward'
link      San Francisco   1     'bit.ly/2001/'   'Jack W'
link      San Francisco   1     'bit.ly/0020/'   'Henry F'
link      San Francisco   1     'bit.ly/2137/'   'Joseph S'
link      San Francisco   1     'bit.ly/3225/'   'Ed B'
link      San Francisco   1     'bit.ly/3667/'   'George Vvec'
link      San Francisco   1     'bit.ly/6434/'   'Robert W'
link      San Francisco   1     'bit.ly/4036/'   'Rudy B'
link      San Francisco   1     'bit.ly/6450/'   'James K'
link      San Francisco   1     'bit.ly/5180/'   'Billy N'
link      San Francisco   1     'bit.ly/7847/'   'John S'

The flatten_json() function (from Medium.com above) destroys the hierarchy. Here are first few rows:

{'city_url': 'link',
 'city_name_0': 'San Francisco',
 'rank': 1,
 'resident_0_link_0': 'bit.ly/0842/',
 'resident_0_name_0': 'John A', ...

Can someone please help me how to think about converting these datasets? Unfortunately, pandas documentation provides no guidance for beginners. HEre's what I was playing with. Nothing worked.

from pandas.io.json import json_normalize
json_normalize(d,['city',['name','rank']])
json_normalize(d,['city','name','rank'])
json_normalize(d,['city','name'])

I'd appreciate if someone guide how to do these type of conversion and the thought process.

Also, I'm looking for a vectorized operation or O(N) operation rather than O(N2) because of the amount of data in the original dataset. Hence, anything slower than O(N) won't work.

caverac · Accepted Answer · 2018-12-06 13:18:22Z

1

If you know the structure of the json blob this will do it

resident_link = [k['link'][0] for k in d['resident']]
resident_name = [k['name'][0] for k in d['resident']]
n = len(d['resident'])
city_url = n * [d['city']['url']]
city_name = n * [d['city']['name'][0]]
rank = n * [d['rank']]

df = pandas.DataFrame({
    'resident_name' : resident_name,
    'resident_link' : resident_link,
    'city_url' : city_url,
    'city_name' : city_name,
    'rank' : rank
})

Which produces

        city_name city_url  rank resident_link resident_name
0   San Francisco     link     1  bit.ly/0842/        John A
1   San Francisco     link     1  bit.ly/5835/        Tedd B
2   San Francisco     link     1  bit.ly/2011/        Cobb C
3   San Francisco     link     1  bit.ly/0855/        Jack N
4   San Francisco     link     1  bit.ly/1430/        Jack K
5   San Francisco     link     1  bit.ly/3081/        Edward
6   San Francisco     link     1  bit.ly/2001/        Jack W
7   San Francisco     link     1  bit.ly/0020/       Henry F
8   San Francisco     link     1  bit.ly/2137/      Joseph S
9   San Francisco     link     1  bit.ly/3225/          Ed B
10  San Francisco     link     1  bit.ly/3667/   George Vvec
11  San Francisco     link     1  bit.ly/6434/      Robert W
12  San Francisco     link     1  bit.ly/4036/        Rudy B
13  San Francisco     link     1  bit.ly/6450/       James K
14  San Francisco     link     1  bit.ly/5180/       Billy N
15  San Francisco     link     1  bit.ly/7847/        John S

EDIT

As the OP say in the comments, imagine there's many records like this, each with the same structure

nrecords = 10
dd = {k : d for k in range(nrecords)}

dd now has 10 copies of the original json blob. And this is how the code should be updated

ff = pandas.DataFrame()

for record in range(nrecords):

    n = len(dd[record]['resident'])

    df = {
        'resident_link' : [k['link'][0] for k in dd[record]['resident']],
        'resident_name' : [k['name'][0] for k in dd[record]['resident']],
        'city_url' : n * [dd[record]['city']['url']],
        'city_name' : n * [dd[record]['city']['name'][0]],
        'rank' : n * [dd[record]['rank']]
        }

    df = pandas.DataFrame(df)
    ff = ff.append(df).reset_index(drop = True)

Below there's an estimation of running time as a function of number of records. Based on this it will take around 1 h to complete 1.5 million records

edited Dec 6, 2018 at 13:18

answered Dec 6, 2018 at 2:05

caverac

1,7072 gold badges16 silver badges23 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

watchtower Over a year ago

thanks for your help. How can I do this for, say, 10 records? We could assume that all 10 records are the same. Will I have to loop through the 10 records? The original dataset that I have has ~1.5M records. I couldn't post sample 10 records because of space issues. Hope you would be able to help me. rank signifies nth record.

caverac Over a year ago

@watchtower Yes, repeat the same and use concat to string the dataframes together

watchtower Over a year ago

Is there any faster method? I tried concat, and this is taking a lot of time.

Collectives™ on Stack Overflow

Normalize JSON using Python

1 Answer 1

3 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related