Python Engineer

Free Python and Machine Learning Tutorials

Become A Patron and get exclusive content! Get access to ML From Scratch notebooks, join a private Slack channel, get priority response, and more! I really appreciate the support!

back to course overview

JSON - Advanced Python 11

02 Jul 2019

JSON (JavaScript Object Notation) is a leightweight data format for data exchange. In Python you have the built-in json module for encoding and decoding JSON data. Simply import it and you are ready to work with JSON data:

import json

Some advantages of JSON: - JSON exists as a "sequence of bytes" which is very useful in the case we need to transmit (stream) data over a network. - Compared to XML, JSON is much smaller, translating into faster data transfers, and better experiences. - JSON is extremely human-friendly since it is textual, and simultaneously machine-friendly.

JSON format

{ "firstName": "Jane", "lastName": "Doe", "hobbies": ["running", "swimming", "singing"], "age": 28, "children": [ { "firstName": "Alex", "age": 5 }, { "firstName": "Bob", "age": 7 } ] }

JSON supports primitive types (strings, numbers, boolean), as well as nested arrays and objects. Simple Python objects are translated to JSON according to the following conversion:

| Python | JSON | | :--- | ---: | | dict | object | | list, tuple | array | | str | string | | int, long, float | number | | True | true | | False | false | | None | null |

From Python to JSON (Serialization, Encode)

Convert Python objects into a JSON string with the json.dumps() method.

import json person = {"name": "John", "age": 30, "city": "New York", "hasChildren": False, "titles": ["engineer", "programmer"]} # convert into JSON: person_json = json.dumps(person) # use different formatting style person_json2 = json.dumps(person, indent=4, separators=("; ", "= "), sort_keys=True) # the result is a JSON string: print(person_json) print(person_json2)
{"name": "John", "age": 30, "city": "New York", "hasChildren": false, "titles":["engineer", "programmer"]} { "age"= 30; "city"= "New York"; "hasChildren"= false; "name"= "John"; "titles"= [ "engineer"; "programmer" ] }

Or convert Python objects into JSON objects and save them into a file with the json.dump() method.

import json person = {"name": "John", "age": 30, "city": "New York", "hasChildren": False, "titles": ["engineer", "programmer"]} with open('person.json', 'w') as f: json.dump(person, f) # you can also specify indent etc...

FROM JSON to Python (Deserialization, Decode)

Convert a JSON string into a Python object with the json.loads() method. The result will be a Python dictionary.

import json person_json = """ { "age": 30, "city": "New York", "hasChildren": false, "name": "John", "titles": [ "engineer", "programmer" ] } """ person = json.loads(person_json) print(person)
{'age': 30, 'city': 'New York', 'hasChildren': False, 'name': 'John', 'titles': ['engineer', 'programmer']}

Or load data from a file and convert it to a Python object with the json.load() method.

import json with open('person.json', 'r') as f: person = json.load(f) print(person)
{'name': 'John', 'age': 30, 'city': 'New York', 'hasChildren': False, 'titles': ['engineer', 'programmer']}

Working with Custom Objects


Encoding a custom object with the default JSONEncoder will raise a TypeError. We can specify a custom encoding function that will store the class name and all object variables in a dictionary. Use this function for the default argument in the json.dump() method.

import json def encode_complex(z): if isinstance(z, complex): # just the key of the class name is important, the value can be arbitrary. return {z.__class__.__name__: True, "real":z.real, "imag":z.imag} else: raise TypeError(f"Object of type '{z.__class__.__name__}' is not JSON serializable") z = 5 + 9j zJSON = json.dumps(z, default=encode_complex) print(zJSON)
{"complex": true, "real": 5.0, "imag": 9.0}

You can also create a custom Encoder class, and overwrite the default() method. Use this for the cls argument in the json.dump() method, or use the encoder directly.

from json import JSONEncoder class ComplexEncoder(JSONEncoder): def default(self, o): if isinstance(z, complex): return {z.__class__.__name__: True, "real":z.real, "imag":z.imag} # Let the base class default method handle other objects or raise a TypeError return JSONEncoder.default(self, o) z = 5 + 9j zJSON = json.dumps(z, cls=ComplexEncoder) print(zJSON) # or use encoder directly: zJson = ComplexEncoder().encode(z) print(zJSON)
{"complex": true, "real": 5.0, "imag": 9.0} {"complex": true, "real": 5.0, "imag": 9.0}


Decoding a custom object with the defaut JSONDecoder is possible, but it will be decoded into a dictionary. Write a custom decode function that will take a dictionary as input, and creates your custom object if it can find the object class name in the dictionary. Use this function for the object_hook argument in the json.load() method.

# Possible but decoded as a dictionary z = json.loads(zJSON) print(type(z)) print(z) def decode_complex(dct): if complex.__name__ in dct: return complex(dct["real"], dct["imag"]) return dct # Now the object is of type complex after decoding z = json.loads(zJSON, object_hook=decode_complex) print(type(z)) print(z)
<class 'dict'> {'complex': True, 'real': 5.0, 'imag': 9.0} <class 'complex'> (5+9j)

Template encode and decode functions

This works for all custom classes if all class variables are given in the __init__ method.

class User: # Custom class with all class variables given in the __init__() def __init__(self, name, age, active, balance, friends): = name self.age = age = active self.balance = balance self.friends = friends class Player: # Other custom class def __init__(self, name, nickname, level): = name self.nickname = nickname self.level = level def encode_obj(obj): """ Takes in a custom object and returns a dictionary representation of the object. This dict representation also includes the object's module and class names. """ # Populate the dictionary with object meta data obj_dict = { "__class__": obj.__class__.__name__, "__module__": obj.__module__ } # Populate the dictionary with object properties obj_dict.update(obj.__dict__) return obj_dict def decode_dct(dct): """ Takes in a dict and returns a custom object associated with the dict. It makes use of the "__module__" and "__class__" metadata in the dictionary to know which object type to create. """ if "__class__" in dct: # Pop ensures we remove metadata from the dict to leave only the instance arguments class_name = dct.pop("__class__") # Get the module name from the dict and import it module_name = dct.pop("__module__") # We use the built in __import__ function since the module name is not yet known at runtime module = __import__(module_name) # Get the class from the module class_ = getattr(module,class_name) # Use dictionary unpacking to initialize the object # Note: This only works if all __init__() arguments of the class are exactly the dict keys obj = class_(**dct) else: obj = dct return obj # User class works with our encoding and decoding methods user = User(name = "John",age = 28, friends = ["Jane", "Tom"], balance = 20.70, active = True) userJSON = json.dumps(user,default=encode_obj,indent=4, sort_keys=True) print(userJSON) user_decoded = json.loads(userJSON, object_hook=decode_dct) print(type(user_decoded)) # Player class also works with our custom encoding and decoding player = Player('Max', 'max1234', 5) playerJSON = json.dumps(player,default=encode_obj,indent=4, sort_keys=True) print(playerJSON) player_decoded = json.loads(playerJSON, object_hook=decode_dct) print(type(player_decoded))
{ "__class__": "User", "__module__": "__main__", "active": true, "age": 28, "balance": 20.7, "friends": [ "Jane", "Tom" ], "name": "John" } <class '__main__.User'> { "__class__": "Player", "__module__": "__main__", "level": 5, "name": "Max", "nickname": "max1234" } <class '__main__.Player'>