Introduction to Object-Oriented Programming

super(), __init__?

Specialist Analytics Team (SAT)

Learning Objectives

Python is an object-oriented language. Every entity is treated as an object; even single integers are objects of the “int” class.
An understanding of object-oriented programming will help give you a better understanding of how the packages you use function.
You can use this understanding to create your own programs that harness the strengths of objected-oriented programming:
- Convenience
- Flexibility
- Extensibility
- Simpler interfacing

In the early days of programming, variables could only be the “primitive” data types containing a single value
- Integer, Float, Boolean, Char
Later came Structures (“Structs”), which can contain multiple values of different types.
Structs were the precursor to objects, but they couldn’t yet contain associated functions within them.
Objects first appear in the Simula programming language in the 1960s for modelling physical phenomena.
Those objects influenced Alan Kay, who coined the term “object-oriented programming” to describe architecture where objects pass information to one another.

Classes act as templates for objects
Objects are referred to as “instances” of classes
- We talk of objects being “instantiated” from a class
- Think of each object as being a copy created using the class template
Objects represent entities with their own data (attributes) and behaviours (methods)
We can create lots of instances of an object with their own attribute values and call methods on them separately yet consistently
Objects are self-contained units that can interact with objects both of the same and of other classes

A diagram representing a class at the top and object instances created from the class

Child classes inherit attributes and methods from parent classes
Child classes can modify / override and add to what they have inherited
Reduces code duplication; increases re-usability
Improves extensibility: i.e. new classes with the same core behaviours, but new features, can be based on existing classes

A parent class (green) and two different child classes (blue and orange), each with objects created from them

Bundling data (attributes) with functions (methods)
Methods are tailor-made to work with the data contained in the object
Saves on having to pass data between multiple functions, which is particularly useful in machine learning models
Pandas DataFrames demonstrate encapsulation. They contain data, but also have methods associated with them
- df = pd.DataFrame(data) <– Instantiating a dataframe object
- df.head(), df.describe(), df.drop() <– calling methods

Objects of different types can be treated in the same way, even if the behaviour differs
- With Pandas DataFrames, .head() will work on both a DataFrame and a Series¹
“Duck typing”: If the behaviour of a thing matches that of another thing, they are considered the same. In OOP terms, the presence of certain methods is more important than which class an object comes from²
- The scikit-learn library’s allows the same code to work for different models

Separating the implementation code from the functionality that users (i.e. other programmers) interact with
Creates a simple interface for parts of a program pass information between each other
Examples:
- When working with machine learning models, users only need to apply simple methods to train the model and make a prediction.
- Entities interacting with each other within a simulation model.

When you want to easily re-use code, to avoid repetition and to extend functionality
Discrete Event Simulations for modelling queueing / capacity problems
Creating custom, branded visualisation packages, for example an NHS-branded SPC chart
- Creating a package that can be used to import the latest data from a website without users having to understand API calls or the website’s structure
When you want to model real-world entities
- Discrete Event Simulations for modelling queueing / capacity problems
When you want to make code modular and easy for others to work with
When you want to simplify end-users’ interaction with Python, fostering a self-service approach to analytics
- Creating custom, branded visualisation packages, for example an NHS-branded SPC chart
Less appropriate for: When you want to be certain of the state of your data at each step of a process, for example when cleansing data

Contact:

Notebook & Slides:

… And don’t forget to give us your feedback.