The pickle module implements an algorithm for turning an arbitrary Python object into a series of bytes. This process is also called serializing the object. The byte stream representing the object can then be transmitted or stored, and later reconstructed to create a new object with the same characteristics.

For the simplest code, we use the dump() and load() functions.

To serialize the object

import pickle

# An arbitrary collection of objects supported by pickle.
data = {
    'a': [1, 2.0, 3, 4+6j],
    'b': ("character string", b"byte string"),
    'c': {None, True, False}
}

with open('data.pickle', 'wb') as f:
    # Pickle the 'data' dictionary using the highest protocol available.
    pickle.dump(data, f, pickle.HIGHEST_PROTOCOL)

---

To deserialize the object

import pickle

with open('data.pickle', 'rb') as f:
    # The protocol version used is detected automatically, so we do not
    # have to specify it.
    data = pickle.load(f)

---

Using pickle and byte objects

It is also possible to serialize into and deserialize out of byte objects, using the dumps and loads function, which are equivalent to dump and load.

serialized_data = pickle.dumps(data, pickle.HIGHEST_PROTOCOL)
# type(serialized_data) is bytes

deserialized_data = pickle.loads(serialized_data)
# deserialized_data == data

Some data cannot be pickled. Other data should not be pickled for other reasons.

What will be pickled can be defined in __getstate__ method. This method must return something that is picklable.

On the oposite side is __setstate__: it will receive what __getstate__ created and has to initialize the object.

class A(object):
    def __init__(self, important_data):
        self.important_data = important_data
        
        # Add data which cannot be pickled:
        self.func = lambda: 7
        
        # Add data which should never be pickled, because it expires quickly:
        self.is_up_to_date = False
    
    def __getstate__(self):
        return [self.important_data] # only this is needed
    
    def __setstate__(self, state):
        self.important_data = state[0]
        
        self.func = lambda: 7  # just some hard-coded unpicklable function
        
        self.is_up_to_date = False  # even if it was before pickling

Now, this can be done:

>>> a1 = A('very important')
>>>
>>> s = pickle.dumps(a1)  # calls a1.__getstate__()
>>>
>>> a2 = pickle.loads(s)  # calls a1.__setstate__(['very important'])
>>> a2
<__main__.A object at 0x0000000002742470>
>>> a2.important_data
'very important'
>>> a2.func()
7

The implementation here pikles a list with one value: [self.important_data]. That was just an example, __getstate__ could have returned anything that is picklable, as long as __setstate__ knows how to do the oppoisite. A good alternative is a dictionary of all values: {'important_data': self.important_data}.

Constructor is not called! Note that in the previous example instance a2 was created in pickle.loads without ever calling A.__init__, so A.__setstate__ had to initialize everything that __init__ would have initialized if it were called.