Adventures in Machine Learning

Optimizing Python Serialization with Pickle and Compression

In Python, serialization is a process that enables us to save objects’ state and data structures in a file or any other form of data storage, so that they can be passed between different environments. Serialization is used for many purposes, including data persistence, messaging, and caching.

In this article, we will explore the basics of serialization in Python, focusing on the modules and methods involved in the operation. We will also take a closer look at the Python Pickle module and how it can be used for serialization, including an example of pickling a custom-defined class.

Serialization in Python

Serialization is a process of converting an object into a format that can be stored, transmitted or reproduced. In Python, serialization is particularly useful for storing complex data structures, preserving objects’ states or transferring data between different instances and environments.

Python provides several standard modules for serialization. The most commonly used ones include pickle, json, and xml.

These modules allow us to convert objects, including lists, tuples, dictionaries, and custom-defined classes, into a stream of bytes that can be stored or transmitted.

Serialization Modules in Python

Python offers different modules for serialization that are catered to different data types and performance requirements.

The JSON (JavaScript Object Notation) module is designed for working with simple data types such as strings, numbers, and lists.

It provides an easily readable, lightweight format that is often used in web applications since it is based on JavaScript. The XML module is efficient in parsing and processing data with hierarchial relationships, such as nested or linked structures.

It is often used in web applications and data exchange formats. The Pickle module is one of the most widely used modules for Python serialization.

It can handle complex data types such as sets, classes, and even functions. It is also known for being efficient and secure when serializing data.

Recommended Modules to Use for Serialization

The module that we choose to use for serialization depends on the nature of the data we want to serialize and the requirements of the project. However, as mentioned earlier, the Pickle module is a highly recommended choice for handling serialization in Python applications.

The pickle module allows us to work with any Python object type, including complex custom classes. It also provides a built-in mechanism for error detection and resolution to ensure serializing and deserializing data is successful.

Inside the Python pickle Module

The Pickle module is a robust feature in Python that is used for serialization. It provides different methods and functions to serialize objects, including the ‘dump’ and ‘load’ methods, both of which can be used to persist object data.

The ‘dump’ method allows us to save data into a file, whereas the ‘load’ method allows us to load or retrieve data from an existing file. To serialize data, we pass the data as input to the dump() function, and it returns a stream of bytes that can be stored in a file or transmitted to another environment.

Methods Included in the Pickle Module

The Pickle module provides several functions that can be used for serialization. Some of the main functions include:


dump() – Serializes an object into a file. 2.

dumps() – Serializes an object into a bytes object. 3.

load() – Deserializes data from a file. 4.

loads() – Deserializes data from a bytes object.

Pickling a Custom-Defined Class

Let’s see an example of pickling a custom-defined class. “` python

import pickle

class Student:

def __init__(self, name, age, major): = name

self.age = age

self.major = major

def display(self):

print(f”{} is {self.age} years old and is majoring in {self.major}.”)

# Create an object of the student class

student_data = Student(‘John’, 25, ‘Computer Science’)

# Serialize the object

serialized_data = pickle.dumps(student_data)

# Save it to a file

with open(“student_data.pickle”, “wb”) as f:


# Load it back from the file

with open(“student_data.pickle”, “rb”) as f:

data = pickle.load(f)

# Display the contents



In this example, we create a Student class and define an object of the class. We then serialize the object using the dumps() function and then save it into a binary file using the “wb” mode.

We then load the pickled data from the file using the open() function, and finally, we deserialize it with the load() function. We can then call the display() function to show the object’s details.


Serialization is a crucial aspect of Python programming. This process allows us to save and transport different data types, including complex custom classes, between different environments.

We have learned about three of the most widely used serialization modules in Python, their recommended use cases, and how to use the Pickle module for serialization and deserialization. By following best practices on serialization in Python, we can ensure that our data is secure and properly managed, allowing us to build robust software applications.

3) Protocol Formats of the Python Pickle Module

The Pickle module in Python provides a default protocol for serializing objects to a stream of bytes. The protocol version determines the way pickling is done, with each protocol version having specific differences, trade-offs, and optimizations.

Overview of Protocol Versions for Pickling

Python provides different versions of protocols that determine how the Pickle module serializes objects into a byte stream. The protocol version has evolved over time, with the latest protocol as of version 3.10 being protocol version 5.

The Pickle module automatically selects the highest protocol version available for serialization during the runtime. In cases where we have to select a specific protocol version or choose an older version for backward compatibility, we can do so using the protocol argument in the dump() and dumps() functions.

Differences Between Protocol Versions

The Pickle module in Python provides five protocol versions, each with its unique features and optimizations. The differences between the versions include the following:

– Protocol Version 0: This version maximizes compatibility, providing an alphanumerical byte stream with the least optimization.

It does not use codes for references, making it slower for caches and less space-efficient. It is not recommended for modern use cases.

– Protocol Version 1: This version is also an alphanumerical format but finds repeating objects and saves references to them as codes, increasing efficiency and speed. It is still slower for caches like lists with repeated values.

– Protocol Version 2: This version is a binary format that addresses the issues the earlier versions have by replacing numerical codes with named references. It optimizes strings, and it’s more space-efficient.

– Protocol Version 3: This version uses a binary format similar to Version 2, but it optimizes storing small integers and optimizing the objects. It should be used in Python 3.0 – 3.3.

– Protocol Version 4: This version was introduced in Python 3.4 and includes new optimizations such as using a more compact binary representation, supporting more built-in types, and inline memoirs.

Protocol Version 5, introduced in Python 3.8, builds on this optimization.

How to Identify the Highest Protocol Version for an Interpreter

To find out the highest protocol version supported by your Python interpreter, you can run the following command:

“` python

import pickle



This will print the highest protocol version supported by your interpreter, which can be used in the dumps() and dump() functions.

4) Picklable and Unpicklable Types

The Pickle module in Python has limitations on what objects it can pickle. Some objects are considered unpicklable since they cannot be serialized using the Pickle module.

These objects could include database connections, sockets, running threads, and more. When pickling such objects, the interpreter raises a PicklingError.

Definition and Examples of Unpicklable Objects

Unpicklable objects are those that cannot be serialized by the Pickle module due to their underlying architecture, behavior, or implementation. Such objects may include mutable data structures, modules with running threads, shared memory, and system resources.

Trying to serialize these objects with the Pickle module will raise exceptions or lead to unexpected outcomes. For example, a file object in its open state cannot be serialized because it’s still open and may change outside of the serialization the Dill Module as an Alternative to Pickle

The dill module is an alternative to the Pickle module.

It is known to serialize a broader range of Python objects, including classes, functions with closures, and more. The dill module builds on the Pickle module, providing an extension for some of its limitations.

Dill adds more features to pickle, such as serialization of decorated functions, lambda functions, and object membership (circular reference). It is compatible with Python 2.6 through 3.9.

Example of How to Exclude Unpicklable Objects from the Pickling Process

When trying to serialize an object that has unpicklable attributes, we can choose to exclude them from the serialization process. One way to do this is by defining the un-picklable objects as non-essential attributes.

“` python

import pickle

class User:

def __init__(self, name, age, db_conn=None): = name

self.age = age

self._db_conn = db_conn

def __getstate__(self):

state = self.__dict__.copy()

del state[‘_db_conn’]

return state

def __setstate__(self, state):


# Define object to serialize

user = User(‘John Doe’, 32, db_conn=None)

# Serialize the object

serialized_user = pickle.dumps(user)

# De-serialize the object

deserialized_user = pickle.loads(serialized_user)

# Display Result

print(deserialized_user.__dict__) # {‘name’: ‘John Doe’, ‘age’: 32}


In this example, the emp object has an unpicklable attribute, db_conn, which is a database connection. We exclude it from the serialization process by defining the `__getstate__()` method that gets called when the object is being pickled.

Within the `__getstate__()` method, we make a copy of the `__dict__` attribute and remove the unpicklable attribute. We then define the `__setstate__()` method that gets called when the object is being unpickled.

Within this method, we update the object’s `__dict__` attribute with the `state` parameter.


The Pickle module is a powerful and efficient tool used in Python to serialize objects by converting them into a byte stream. We have learned about the different protocol versions, their differences, and how to find the highest protocol version supported by your interpreter.

We also learned about unpicklable objects and how to exclude them from the serialization process. Finally, we introduced the dill module, an alternative to Pickle, which supports a more comprehensive range of Python objects.

5) Compression of Pickled Objects

The Pickle module is a powerful tool used in Python for serializing and deserializing objects. Sometimes the serialized objects can be quite large, especially when dealing with complex objects, which can take up a considerable amount of memory.

One way to reduce the memory usage of such objects is to apply compression to the pickled data. In this article, we will discuss the process of compressing pickled strings using the bzip2 or gzip compression Compressing Pickled Strings with bzip2 or gzip

Most serialization libraries, including Pickle, generate data in a human-readable structure.

However, compressed data is not typically human-readable but still contains the same information, only in a smaller size. By compressing the pickled data, it will take up less space and can be more efficient to store or transfer.

The compression algorithms used to compress pickled data include bzip2 or gzip. These algorithms can be applied to the serialized data to reduce the memory usage of the data.

Bzip2 is a compression algorithm that provides better compression than gzip but is slower than gzip at both compression and decompression. Gzip, on the other hand, is faster at both compression and decompression, although the compression rate is lower than bzip2.

Benefits of Compressed Pickled Strings

Compressing pickled data has several advantages:

1. Reduced Memory Usage: Compressed data takes up less space in memory, allowing for greater efficiency in storage and transfer.

2. Faster Data Transfer: Compressed data can be sent over the network more quickly, reducing transfer times.

3. Better File Management: Compressed data can be more easily managed, as it takes up less space on the hard drive.

4. Improved Security: Compressed data can be encrypted more easily, increasing the security of the data.

How to Compress Pickled Strings with bzip2 or gzip

To compress pickled data using bzip2 or gzip, we first need to import the appropriate module. “` python

import bz2

import gzip

import pickle


Once we have imported the required modules, we can serialize the object and compress it using either module.

“` python

# Sample Object

data = {

‘name’: ‘John’,

‘age’: 32,

’email’: ‘[email protected]


# Serializing Data

serialized_data = pickle.dumps(data)

# Compressing Pickled String with bzip2

compressed_data = bz2.compress(serialized_data)

# Compressing Pickled String with gzip

compressed_data_gzip = gzip.compress(serialized_data)


In this example, we use the Pickle module to serialize the data into a byte stream.

We then compress the pickled data using either bzip2 or gzip compression. To decompress the compressed data, we can use the `decompress()` method of the compression module, as shown in the example below.

“` python

# Decompressing Pickled String with bzip2

decompressed_data = bz2.decompress(compressed_data)

# Decompressing Pickled String with gzip

decompressed_data_gzip = gzip.decompress(compressed_data_gzip)

# Deserializing data from decompressed string

data = pickle.loads(decompressed_data)

data_gzip = pickle.loads(decompressed_data_gzip)


In this example, we use the `decompress()` method of each compression module to decompress the pickled data. We then deserialize the data by calling the `loads()` method of the Pickle module on the decompressed data.


By compressing pickled data with bzip2 or gzip compression, we can optimize memory usage and improve the efficiency of data transfer and storage. This article has provided an overview of the benefits of compressing pickled data, and we have shown how to compress and decompress pickled data using bzip2 or gzip compression.

By leveraging these techniques, we can make our Python applications more robust, efficient, and optimized. In conclusion, this article has provided an in-depth exploration of the Python Pickle module, covering topics such as serialization, protocol versioning, and unpicklable objects.

We have also explored the benefits of compressing pickled strings using bzip2 and gzip compression algorithms. Compressing pickled data can lead to reduced memory usage, faster data transfer, and improved security.

The main takeaway is that understanding the limitations of the Pickle module and how to overcome them is crucial to building efficient and robust Python applications. Compressing pickled data is just one way to optimize the serialization process and improve application performance.