How to make a Python class serializable?

A simple class:

class FileItem:
    def __init__(self, fname):
        self.fname = fname

What should I do to be able to get output of:

>>> import json

>>> my_file = FileItem('/foo/bar')
>>> json.dumps(my_file)
TypeError: Object of type 'FileItem' is not JSON serializable

Without the error

Do you have an idea about the expected output? For e.g. will this do?

>>> f  = FileItem("/foo/bar")
>>> magic(f)
'{"fname": "/foo/bar"}'

In that case you can merely call json.dumps(f.__dict__).

If you want more customized output then you will have to subclass JSONEncoder and implement your own custom serialization.

For a trivial example, see below.

>>> from json import JSONEncoder
>>> class MyEncoder(JSONEncoder):
        def default(self, o):
            return o.__dict__    

>>> MyEncoder().encode(f)
'{"fname": "/foo/bar"}'

Then you pass this class into the json.dumps() method as cls kwarg:

json.dumps(cls=MyEncoder)

If you also want to decode then you’ll have to supply a custom object_hook to the JSONDecoder class. For e.g.

>>> def from_json(json_object):
        if 'fname' in json_object:
            return FileItem(json_object['fname'])
>>> f = JSONDecoder(object_hook = from_json).decode('{"fname": "/foo/bar"}')
>>> f
<__main__.FileItem object at 0x9337fac>
>>> 

Here is a simple solution for a simple feature:

.toJSON() Method

Instead of a JSON serializable class, implement a serializer method:

import json

class Object:
    def toJSON(self):
        return json.dumps(self, default=lambda o: o.__dict__, 
            sort_keys=True, indent=4)

So you just call it to serialize:

me = Object()
me.name = "Onur"
me.age = 35
me.dog = Object()
me.dog.name = "Apollo"

print(me.toJSON())

will output:

{
    "age": 35,
    "dog": {
        "name": "Apollo"
    },
    "name": "Onur"
}

For more complex classes you could consider the tool jsonpickle:

jsonpickle is a Python library for serialization and deserialization of complex Python objects to and from JSON.

The standard Python libraries for encoding Python into JSON, such as the stdlib’s json, simplejson, and demjson, can only handle Python primitives that have a direct JSON equivalent (e.g. dicts, lists, strings, ints, etc.). jsonpickle builds on top of these libraries and allows more complex data structures to be serialized to JSON. jsonpickle is highly configurable and extendable–allowing the user to choose the JSON backend and add additional backends.

(link to jsonpickle on PyPi)

Most of the answers involve changing the call to json.dumps(), which is not always possible or desirable (it may happen inside a framework component for example).

If you want to be able to call json.dumps(obj) as is, then a simple solution is inheriting from dict:

class FileItem(dict):
    def __init__(self, fname):
        dict.__init__(self, fname=fname)

f = FileItem('tasks.txt')
json.dumps(f)  #No need to change anything here

This works if your class is just basic data representation, for trickier things you can always set keys explicitly.

I like Onur’s answer but would expand to include an optional toJSON() method for objects to serialize themselves:

def dumper(obj):
    try:
        return obj.toJSON()
    except:
        return obj.__dict__
print json.dumps(some_big_object, default=dumper, indent=2)

Another option is to wrap JSON dumping in its own class:

import json

class FileItem:
    def __init__(self, fname):
        self.fname = fname

    def __repr__(self):
        return json.dumps(self.__dict__)

Or, even better, subclassing FileItem class from a JsonSerializable class:

import json

class JsonSerializable(object):
    def toJson(self):
        return json.dumps(self.__dict__)

    def __repr__(self):
        return self.toJson()


class FileItem(JsonSerializable):
    def __init__(self, fname):
        self.fname = fname

Testing:

>>> f = FileItem('/foo/bar')
>>> f.toJson()
'{"fname": "/foo/bar"}'
>>> f
'{"fname": "/foo/bar"}'
>>> str(f) # string coercion
'{"fname": "/foo/bar"}'

Just add to_json method to your class like this:

def to_json(self):
  return self.message # or how you want it to be serialized

And add this code (from this answer), to somewhere at the top of everything:

from json import JSONEncoder

def _default(self, obj):
    return getattr(obj.__class__, "to_json", _default.default)(obj)

_default.default = JSONEncoder().default
JSONEncoder.default = _default

This will monkey-patch json module when it’s imported so JSONEncoder.default() automatically checks for a special “to_json()” method and uses it to encode the object if found.

Just like Onur said, but this time you don’t have to update every json.dumps() in your project.

I came across this problem the other day and implemented a more general version of an Encoder for Python objects that can handle nested objects and inherited fields:

import json
import inspect

class ObjectEncoder(json.JSONEncoder):
    def default(self, obj):
        if hasattr(obj, "to_json"):
            return self.default(obj.to_json())
        elif hasattr(obj, "__dict__"):
            d = dict(
                (key, value)
                for key, value in inspect.getmembers(obj)
                if not key.startswith("__")
                and not inspect.isabstract(value)
                and not inspect.isbuiltin(value)
                and not inspect.isfunction(value)
                and not inspect.isgenerator(value)
                and not inspect.isgeneratorfunction(value)
                and not inspect.ismethod(value)
                and not inspect.ismethoddescriptor(value)
                and not inspect.isroutine(value)
            )
            return self.default(d)
        return obj

Example:

class C(object):
    c = "NO"
    def to_json(self):
        return {"c": "YES"}

class B(object):
    b = "B"
    i = "I"
    def __init__(self, y):
        self.y = y

    def f(self):
        print "f"

class A(B):
    a = "A"
    def __init__(self):
        self.b = [{"ab": B("y")}]
        self.c = C()

print json.dumps(A(), cls=ObjectEncoder, indent=2, sort_keys=True)

Result:

{
  "a": "A", 
  "b": [
    {
      "ab": {
        "b": "B", 
        "i": "I", 
        "y": "y"
      }
    }
  ], 
  "c": {
    "c": "YES"
  }, 
  "i": "I"
}

If you’re using Python3.5+, you could use jsons. It will convert your object (and all its attributes recursively) to a dict.

import jsons

a_dict = jsons.dump(your_object)

Or if you wanted a string:

a_str = jsons.dumps(your_object)

Or if your class implemented jsons.JsonSerializable:

a_dict = your_object.json

import simplejson

class User(object):
    def __init__(self, name, mail):
        self.name = name
        self.mail = mail

    def _asdict(self):
        return self.__dict__

print(simplejson.dumps(User('alice', 'alice@mail.com')))

if use standard json, u need to define a default function

import json
def default(o):
    return o._asdict()

print(json.dumps(User('alice', 'alice@mail.com'), default=default))

json is limited in terms of objects it can print, and jsonpickle (you may need a pip install jsonpickle) is limited in terms it can’t indent text. If you would like to inspect the contents of an object whose class you can’t change, I still couldn’t find a straighter way than:

 import json
 import jsonpickle
 ...
 print  json.dumps(json.loads(jsonpickle.encode(object)), indent=2)

Note: that still they can’t print the object methods.

This class can do the trick, it converts object to standard json .

import json


class Serializer(object):
    @staticmethod
    def serialize(object):
        return json.dumps(object, default=lambda o: o.__dict__.values()[0])

usage:

Serializer.serialize(my_object)

working in python2.7 and python3.

import json

class Foo(object):
    def __init__(self):
        self.bar = 'baz'
        self._qux = 'flub'

    def somemethod(self):
        pass

def default(instance):
    return {k: v
            for k, v in vars(instance).items()
            if not str(k).startswith('_')}

json_foo = json.dumps(Foo(), default=default)
assert '{"bar": "baz"}' == json_foo

print(json_foo)

jaraco gave a pretty neat answer. I needed to fix some minor things, but this works:

Code

# Your custom class
class MyCustom(object):
    def __json__(self):
        return {
            'a': self.a,
            'b': self.b,
            '__python__': 'mymodule.submodule:MyCustom.from_json',
        }

    to_json = __json__  # supported by simplejson

    @classmethod
    def from_json(cls, json):
        obj = cls()
        obj.a = json['a']
        obj.b = json['b']
        return obj

# Dumping and loading
import simplejson

obj = MyCustom()
obj.a = 3
obj.b = 4

json = simplejson.dumps(obj, for_json=True)

# Two-step loading
obj2_dict = simplejson.loads(json)
obj2 = MyCustom.from_json(obj2_dict)

# Make sure we have the correct thing
assert isinstance(obj2, MyCustom)
assert obj2.__dict__ == obj.__dict__

Note that we need two steps for loading. For now, the __python__ property is not used.

How common is this?

Using the method of AlJohri, I check popularity of approaches:

Serialization (Python -> JSON):

• to_json: 266,595 on 2018-06-27
• toJSON: 96,307 on 2018-06-27
• __json__: 8,504 on 2018-06-27
• for_json: 6,937 on 2018-06-27

Deserialization (JSON -> Python):

• from_json: 226,101 on 2018-06-27

This has worked well for me:

class JsonSerializable(object):

    def serialize(self):
        return json.dumps(self.__dict__)

    def __repr__(self):
        return self.serialize()

    @staticmethod
    def dumper(obj):
        if "serialize" in dir(obj):
            return obj.serialize()

        return obj.__dict__

and then

class FileItem(JsonSerializable):
    ...

and

log.debug(json.dumps(<my object>, default=JsonSerializable.dumper, indent=2))

If you don’t mind installing a package for it, you can use json-tricks:

pip install json-tricks

After that you just need to import dump(s) from json_tricks instead of json, and it’ll usually work:

from json_tricks import dumps
json_str = dumps(cls_instance, indent=4)

which’ll give

{
        "__instance_type__": [
                "module_name.test_class",
                "MyTestCls"
        ],
        "attributes": {
                "attr": "val",
                "dct_attr": {
                        "hello": 42
                }
        }
}

And that’s basically it!

This will work great in general. There are some exceptions, e.g. if special things happen in __new__, or more metaclass magic is going on.

Obviously loading also works (otherwise what’s the point):

from json_tricks import loads
json_str = loads(json_str)

This does assume that module_name.test_class.MyTestCls can be imported and hasn’t changed in non-compatible ways. You’ll get back an instance, not some dictionary or something, and it should be an identical copy to the one you dumped.

If you want to customize how something gets (de)serialized, you can add special methods to your class, like so:

class CustomEncodeCls:
        def __init__(self):
                self.relevant = 42
                self.irrelevant = 37

        def __json_encode__(self):
                # should return primitive, serializable types like dict, list, int, string, float...
                return {'relevant': self.relevant}

        def __json_decode__(self, **attrs):
                # should initialize all properties; note that __init__ is not called implicitly
                self.relevant = attrs['relevant']
                self.irrelevant = 12

which serializes only part of the attributes parameters, as an example.

And as a free bonus, you get (de)serialization of numpy arrays, date & times, ordered maps, as well as the ability to include comments in json.

Disclaimer: I created json_tricks, because I had the same problem as you.

jsonweb seems to be the best solution for me. See http://www.jsonweb.info/en/latest/

from jsonweb.encode import to_object, dumper

@to_object()
class DataModel(object):
  def __init__(self, id, value):
   self.id = id
   self.value = value

>>> data = DataModel(5, "foo")
>>> dumper(data)
'{"__type__": "DataModel", "id": 5, "value": "foo"}'

Here is my 3 cents …
This demonstrates explicit json serialization for a tree-like python object.
Note: If you actually wanted some code like this you could use the twisted FilePath class.

import json, sys, os

class File:
    def __init__(self, path):
        self.path = path

    def isdir(self):
        return os.path.isdir(self.path)

    def isfile(self):
        return os.path.isfile(self.path)

    def children(self):        
        return [File(os.path.join(self.path, f)) 
                for f in os.listdir(self.path)]

    def getsize(self):        
        return os.path.getsize(self.path)

    def getModificationTime(self):
        return os.path.getmtime(self.path)

def _default(o):
    d = {}
    d['path'] = o.path
    d['isFile'] = o.isfile()
    d['isDir'] = o.isdir()
    d['mtime'] = int(o.getModificationTime())
    d['size'] = o.getsize() if o.isfile() else 0
    if o.isdir(): d['children'] = o.children()
    return d

folder = os.path.abspath('.')
json.dump(File(folder), sys.stdout, default=_default)

I ran into this problem when I tried to store Peewee’s model into PostgreSQL JSONField.

After struggling for a while, here’s the general solution.

The key to my solution is going through Python’s source code and realizing that the code documentation (described here) already explains how to extend the existing json.dumps to support other data types.

Suppose you current have a model that contains some fields that are not serializable to JSON and the model that contains the JSON field originally looks like this:

class SomeClass(Model):
    json_field = JSONField()

Just define a custom JSONEncoder like this:

class CustomJsonEncoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, SomeTypeUnsupportedByJsonDumps):
            return < whatever value you want >
        return json.JSONEncoder.default(self, obj)

    @staticmethod
    def json_dumper(obj):
        return json.dumps(obj, cls=CustomJsonEncoder)

And then just use it in your JSONField like below:

class SomeClass(Model):
    json_field = JSONField(dumps=CustomJsonEncoder.json_dumper)

The key is the default(self, obj) method above. For every single ... is not JSON serializable complaint you receive from Python, just add code to handle the unserializable-to-JSON type (such as Enum or datetime)

For example, here’s how I support a class inheriting from Enum:

class TransactionType(Enum):
   CURRENT = 1
   STACKED = 2

   def default(self, obj):
       if isinstance(obj, TransactionType):
           return obj.value
       return json.JSONEncoder.default(self, obj)

Finally, with the code implemented like above, you can just convert any Peewee models to be a JSON-seriazable object like below:

peewee_model = WhateverPeeweeModel()
new_model = SomeClass()
new_model.json_field = model_to_dict(peewee_model)

Though the code above was (somewhat) specific to Peewee, but I think:

1. It’s applicable to other ORMs (Django, etc) in general
2. Also, if you understood how json.dumps works, this solution also works with Python (sans ORM) in general too

Any questions, please post in the comments section. Thanks!

This function uses recursion to iterate over every part of the dictionary and then calls the repr() methods of classes that are not build-in types.

def sterilize(obj):
    object_type = type(obj)
    if isinstance(obj, dict):
        return {k: sterilize(v) for k, v in obj.items()}
    elif object_type in (list, tuple):
        return [sterilize(v) for v in obj]
    elif object_type in (str, int, bool):
        return obj
    else:
        return obj.__repr__()

This is a small library that serializes an object with all its children to JSON and also parses it back:

https://github.com/Toubs/PyJSONSerialization/

I came up with my own solution. Use this method, pass any document (dict,list, ObjectId etc) to serialize.

def getSerializable(doc):
    # check if it's a list
    if isinstance(doc, list):
        for i, val in enumerate(doc):
            doc[i] = getSerializable(doc[i])
        return doc

    # check if it's a dict
    if isinstance(doc, dict):
        for key in doc.keys():
            doc[key] = getSerializable(doc[key])
        return doc

    # Process ObjectId
    if isinstance(doc, ObjectId):
        doc = str(doc)
        return doc

    # Use any other custom serializting stuff here...

    # For the rest of stuff
    return doc

I chose to use decorators to solve the datetime object serialization problem. Here is my code:

#myjson.py
#Author: jmooremcc 7/16/2017

import json
from datetime import datetime, date, time, timedelta
"""
This module uses decorators to serialize date objects using json
The filename is myjson.py
In another module you simply add the following import statement:
    from myjson import json

json.dumps and json.dump will then correctly serialize datetime and date 
objects
"""

def json_serial(obj):
    """JSON serializer for objects not serializable by default json code"""

    if isinstance(obj, (datetime, date)):
        serial = str(obj)
        return serial
    raise TypeError ("Type %s not serializable" % type(obj))


def FixDumps(fn):
    def hook(obj):
        return fn(obj, default=json_serial)

    return hook

def FixDump(fn):
    def hook(obj, fp):
        return fn(obj,fp, default=json_serial)

    return hook


json.dumps=FixDumps(json.dumps)
json.dump=FixDump(json.dump)


if __name__=="__main__":
    today=datetime.now()
    data={'atime':today, 'greet':'Hello'}
    str=json.dumps(data)
    print str

By importing the above module, my other modules use json in a normal way (without specifying the default keyword) to serialize data that contains date time objects. The datetime serializer code is automatically called for json.dumps and json.dump.

I liked Lost Koder’s method the most. I ran into issues when trying to serialize more complex objects whos members/methods aren’t serializable. Here’s my implementation that works on more objects:

class Serializer(object):
    @staticmethod
    def serialize(obj):
        def check(o):
            for k, v in o.__dict__.items():
                try:
                    _ = json.dumps(v)
                    o.__dict__[k] = v
                except TypeError:
                    o.__dict__[k] = str(v)
            return o
        return json.dumps(check(obj).__dict__, indent=2)

If you are able to install a package, I’d recommend trying dill, which worked just fine for my project. A nice thing about this package is that it has the same interface as pickle, so if you have already been using pickle in your project you can simply substitute in dill and see if the script runs, without changing any code. So it is a very cheap solution to try!

(Full anti-disclosure: I am in no way affiliated with and have never contributed to the dill project.)

Install the package:

pip install dill

Then edit your code to import dill instead of pickle:

# import pickle
import dill as pickle

Run your script and see if it works. (If it does you may want to clean up your code so that you are no longer shadowing the pickle module name!)

Some specifics on datatypes that dill can and cannot serialize, from the project page:

dill can pickle the following standard types:

none, type, bool, int, long, float, complex, str, unicode, tuple, list, dict, file, buffer, builtin, both old and new style classes, instances of old and new style classes, set, frozenset, array, functions, exceptions

dill can also pickle more ‘exotic’ standard types:

functions with yields, nested functions, lambdas, cell, method, unboundmethod, module, code, methodwrapper, dictproxy, methoddescriptor, getsetdescriptor, memberdescriptor, wrapperdescriptor, xrange, slice, notimplemented, ellipsis, quit

dill cannot yet pickle these standard types:

frame, generator, traceback

I see no mention here of serial versioning or backcompat, so I will post my solution which I’ve been using for a bit. I probably have a lot more to learn from, specifically Java and Javascript are probably more mature than me here but here goes

https://gist.github.com/andy-d/b7878d0044a4242c0498ed6d67fd50fe

To add another option: You can use the attrs package and the asdict method.

class ObjectEncoder(JSONEncoder):
    def default(self, o):
        return attr.asdict(o)

json.dumps(objects, cls=ObjectEncoder)

and to convert back

def from_json(o):
    if '_obj_name' in o:
        type_ = o['_obj_name']
        del o['_obj_name']
        return globals()[type_](**o)
    else:
        return o

data = JSONDecoder(object_hook=from_json).decode(data)

class looks like this

@attr.s
class Foo(object):
    x = attr.ib()
    _obj_name = attr.ib(init=False, default='Foo')

In addition to the Onur’s answer, You possibly want to deal with datetime type like below.
(in order to handle: ‘datetime.datetime’ object has no attribute ‘dict‘ exception.)

def datetime_option(value):
    if isinstance(value, datetime.date):
        return value.timestamp()
    else:
        return value.__dict__

Usage:

def toJSON(self):
    return json.dumps(self, default=datetime_option, sort_keys=True, indent=4)

First we need to make our object JSON-compliant, so we can dump it using the standard JSON module. I did it this way:

def serialize(o):
    if isinstance(o, dict):
        return {k:serialize(v) for k,v in o.items()}
    if isinstance(o, list):
        return [serialize(e) for e in o]
    if isinstance(o, bytes):
        return o.decode("utf-8")
    return o

Building on Quinten Cabo‘s answer:

def sterilize(obj):
    if type(obj) in (str, float, int, bool, type(None)):
        return obj
    elif isinstance(obj, dict):
        return {k: sterilize(v) for k, v in obj.items()}
    elif hasattr(obj, '__iter__') and callable(obj.__iter__):
        return [sterilize(v) for v in obj]
    elif hasattr(obj, '__dict__'):
        return {k: sterilize(v) for k, v in obj.__dict__.items() if k not in ['__module__', '__dict__', '__weakref__', '__doc__']}
    else:
        return repr(obj)

The differences are

1. Works for any iterable instead of just list and tuple (it works for NumPy arrays, etc.)
2. Works for dynamic types (ones that contain a __dict__).
3. Includes native types float and None so they don’t get converted to string.

Left as an exercise to the reader is to handle __slots__, classes that are both iterable and have members, classes that are dictionaries and also have members, etc.