How to read this lecture...

Code should execute sequentially if run in a Jupyter notebook

Object Oriented Programming


OOP is one of the major paradigms in programming, and nicely supported in Python

OOP has become an important concept in modern software engineering because

  • It can help facilitate clean, efficient code (if used well)
  • The OOP design pattern fits well with many computing problems

OOP is about producing well organized code — an important determinant of productivity

Moreover, OOP is a part of Python, and to progress further it’s necessary to understand the basics

About OOP

OOP is supported in many languages:

  • JAVA and Ruby are relatively pure OOP
  • Python supports both procedural and object-oriented programming
  • Fortran and MATLAB are mainly procedural, some OOP recently tacked on
  • C is a procedural language, while C++ is C with OOP added on top

Let’s look at general concepts before we specialize to Python

Key Concepts

The traditional (non-OOP) paradigm is called procedural, and works as follows

  • The program has a state that contains the values of its variables
  • Functions are called to act on these data according to the task
  • Data are passed back and forth via function calls

In contrast, in the OOP paradigm, data and functions are bundled together into “objects”

An example is a Python list, which not only stores data, but also knows how to sort itself, etc.

x = [1, 5, 4]
[1, 4, 5]

Here sort is a function that is “part of” the list object

In the OOP setting, functions are usually called methods (e.g., sort is a list method)

Standard Terminology

A class definition is a blueprint for a particular class of objects (e.g., lists, strings or complex numbers)

It describes

  • What kind of data the class stores
  • What methods it has for acting on these data

An object or instance is a realization of the class, created from the blueprint

  • Each instance has its own unique data
  • Methods set out in the class definition act on this (and other) data

In Python, the data and methods of an object are collectively referred to as attributes

Attributes are accessed via “dotted attribute notation”

  • object_name.method_name()

In the example

x = [1, 5, 4]
  • x is an object or instance, created from the definition for Python lists, but with its own particular data
  • x.sort() and x.__class__ are two attributes of x
  • dir(x) can be used to view all the attributes of x

Why is OOP Useful?

OOP is useful for the same reason that abstraction is useful: for recognizing and exploiting common structure

  • E.g., a general equilibrium theory consists of a commodity space, preferences, technologies, and an equilibrium definition
  • E.g., a game consists of a list of players, lists of actions available to each player, player payoffs as functions of all players’ actions, and a timing protocol

One concrete setting where OOP is almost always used is computer desktop environments with windows

Windows have common functionality and individual data, which makes them suitable for implementing with OOP

  • individual data: contents of specific windows
  • common functionality: closing, maximizing, etc.

Individual windows are created as objects from a class definition, with their own “instance” data

Common functionality is implemented as set of methods, which all of these objects share

Data Encapsulation

Another, more prosaic, use of OOP is data encapsulation

Data encapsulation means “hiding” variables rather than making them directly accessible

The alternative is filling the global namespace with variable names, which frequently leads to conflicts

  • Think of the global namespace as any name you can refer to without a dot in front of it

Example. The modules os and sys both define a different attribute called path

The following code leads immediately to a conflict

from os import path
from sys import path

At this point, both variables have been brought into the global namespace, and the second will shadow the first

A better idea is to replace the above with

import os
import sys

and then reference the path you want with either os.path or sys.path

This example shows that modules provide one means of data encapsulation

As will now become clear, OOP provides another

Defining Your Own Classes

Let’s build a super simple class as an exercise

class Consumer:

c1 = Consumer()  # Create an instance
c1.wealth = 10

Comments on notation:

  • The class keyword indicates that we are building a class
  • The pass keyword is used in Python to stand in for an empty code block
  • “Calling” the class with syntax ClassName() creates an instance of the class

Notice the flexibility of Python:

  • We don’t actually need to specify what attributes a class will have
  • We can attach new attributes to instances of the class on the fly

However, most classes have more structure than our Consumer class

In fact the main point of classes is to provide a blueprint containing useful functionality for a given set of tasks

  • For example, the sort method in x.sort() is specified in the blueprint for the list data type because it is useful for working with lists

Let’s try to build something a bit closer to this standard conception of OOP

Example: Another Consumer Class

Let’s build a Consumer class with more structure:

  • A wealth attribute that stores the consumer’s wealth (data)
  • An earn method, where earn(y) increments the consumer’s wealth by y
  • A spend method, where spend(x) either decreases wealth by x or returns an error if insufficient funds exist

Admittedly a little contrived, this example of a class helps us internalize some new syntax

Here’s one implementation, from file

class Consumer:
    def __init__(self, w):
        "Initialize consumer with w dollars of wealth"
        self.wealth = w
    def earn(self, y):
        "The consumer earns y dollars" 
        self.wealth += y
    def spend(self, x):
        "The consumer spends x dollars if feasible"
        new_wealth = self.wealth - x
        if new_wealth < 0:
            print("Insufficent funds")
            self.wealth = new_wealth

There’s some special syntax here so let’s step through carefully

This class defines instance data wealth and three methods: __init__, earn and spend

  • wealth is instance data because each consumer we create (each instance of the Consumer class) will have its own separate wealth data

The ideas behind the earn and spend methods were discussed above

Both of these act on the instance data wealth

The __init__ method is a constructor method

Whenever we create an instance of the class, this method will be called automatically

Calling __init__ sets up a “namespace” to hold the instance data — more on this soon

We’ll also discuss the role of self just below


Here’s an example of usage

c1 = Consumer(10)  # Create instance with initial wealth 10
Insufficent funds

We can of course create multiple instances each with its own data

c1 = Consumer(10)
c2 = Consumer(12)

In fact each instance stores its data in a separate namespace dictionary

{'wealth': 10}
{'wealth': 8}

When we access or set attributes we’re actually just modifying the dictionary maintained by the instance


If you look at the Consumer class definition again you’ll see the word self throughout the code

The rules with self are that

  • Any instance data should be prepended with self

    • e.g., the earn method references self.wealth rather than just wealth
  • Any method defined within the class should have self as its first argument

    • e.g., def earn(self, y) rather than just def earn(y)
  • Any method referenced within the class should be called as self.method_name

There are no examples of the last rule in the preceding code but we will see some shortly


In this section we look at some more formal details related to classes and self

  • You might wish to skip to the next section on first pass of this lecture
  • You can return to these details after you’ve familiarized yourself with more examples

Methods actually live inside a class object formed when the interpreter reads the class definition

print(Consumer.__dict__)  # Show __dict__ attribute of class object
{'earn': <function Consumer.earn at 0x7f2590054d90>,
'spend': <function Consumer.spend at 0x7f2590054e18>,
'__doc__': None,
'__weakref__': <attribute '__weakref__' of 'Consumer' objects>,
'__init__': <function Consumer.__init__ at 0x7f2590054d08>,
'__module__': '__main__',
'__dict__': <attribute '__dict__' of 'Consumer' objects>}

Note how the three methods __init__, earn and spend are stored in the class object

Consider the following code

c1 = Consumer(10)

When you call earn via c1.earn(10) the interpreter passes the instance c1 and the argument 10 to Consumer.earn

In fact the following are equivalent

  • c1.earn(10)
  • Consumer.earn(c1, 10)

In the function call Consumer.earn(c1, 10) note that c1 is the first argument

Recall that in the definition of the earn method, self is the first parameter

def earn(self, y):
     "The consumer earns y dollars"
     self.wealth += y

The end result is that self is bound to the instance c1 inside the function call

That’s why the statement self.wealth += y inside earn ends up modifying c1.wealth

Example: The Solow Growth Model

For our next example, let’s write a simple class to implement the Solow growth model

The Solow growth model is a neoclassical growth model where the amount of capital stock per capita \(k_t\) evolves according to the rule

(1)\[k_{t+1} = \frac{s z k_t^{\alpha} + (1 - d) k_t}{1 + n}\]


  • \(s\) is an exogenously given savings rate
  • \(z\) is a productivity parameter
  • \(\alpha\) is capital’s share of income
  • \(n\) is the population growth rate
  • \(d\) is the depreciation rate

The steady state of the model is the \(k\) that solves (1) when \(k_{t+1} = k_t = k\)

The code shown below can be downloaded here

from __future__ import division  # Omit for Python 3.x
import numpy as np

class Solow:
    Implements the Solow growth model with update rule

    .. math::
        k_{t+1} = \frac{s z k^{\alpha}_t}{1 + n}  + k_t \frac{1 - d}{1 + n}

    def __init__(self, n, s, d, alpha, z, k):
        Solow growth model with Cobb Douglas production function.  All
        parameters are scalars.  See
        for interpretation.
        self.n, self.s, self.d, self.alpha, self.z = n, s, d, alpha, z
        self.k = k

    def h(self):
        "Evaluate the h function"
        temp = self.s * self.z * self.k**self.alpha + self.k * (1 - self.d)
        return temp / (1 + self.n)

    def update(self):
        "Update the current state (i.e., the capital stock)."
        self.k =  self.h()
    def steady_state(self):
         "Compute the steady state value of capital."
         return ((self.s * self.z) / (self.n + self.d))**(1 / (1 - self.alpha))
    def generate_sequence(self, t):
        "Generate and return a time series of length t"
        path = []
        for i in range(t):
        return path

Some points of interest in the code are

  • An instance maintains a record of its current capital stock in the variable self.k

  • The h method implements the right hand side of (1)

  • The update method uses h to update capital as per (1)

    • Notice how inside update the reference to the local method h is self.h

The methods steady_state and generate_sequence are fairly self explanatory

Here’s a little program that uses the class to compute time series from two different initial conditions

The common steady state is also plotted for comparison

import matplotlib.pyplot as plt

baseline_params = 0.05, 0.25, 0.1, 0.3, 2.0, 1.0
s1 = Solow(*baseline_params)  # The 'splat' operator * breaks up the tuple
s2 = Solow(*baseline_params)
s2.k = 8.0  # Reset s2.k to make high capital economy
T = 60
fig, ax = plt.subplots()
# Plot the common steady state value of capital
ax.plot([s1.steady_state()]*T, 'k-', label='steady state')
# Plot time series for each economy
for s in s1, s2:
    lb = 'capital series from initial state {}'.format(s.k)
    ax.plot(s.generate_sequence(T), 'o-', lw=2, alpha=0.6, label=lb)

ax.legend(loc='lower right')

Here’s the figure it produces


Example: A Market

Next let’s write a class for a simple one good market where agents are price takers

The market consists of the following objects:

  • A linear demand curve \(Q = a_d - b_d p\)
  • A linear supply curve \(Q = a_z + b_z (p - t)\)


  • \(p\) is price paid by the consumer, \(Q\) is quantity, and \(t\) is a per unit tax
  • Other symbols are demand and supply parameters

The class provides methods to compute various values of interest, including competitive equlibrium price and quantity, tax revenue raised, consumer surplus and producer surplus

Here’s our implementation


from __future__ import division
from scipy.integrate import quad

class Market:

    def __init__(self, ad, bd, az, bz, tax):
        Set up market parameters.  All parameters are scalars.  See for interpretation.

        """,,,, = ad, bd, az, bz, tax
        if ad < az:
            raise ValueError('Insufficient demand.')
    def price(self):
        "Return equilibrium price"
        return  ( - +* + 
    def quantity(self):
        "Compute equilibrium quantity"
        return - * self.price()
    def consumer_surp(self):
        "Compute consumer surplus"
        # == Compute area under inverse demand function == #
        integrand = lambda x: ( - (1/* x
        area, error = quad(integrand, 0, self.quantity())
        return area - self.price() * self.quantity()
    def producer_surp(self):
        "Compute producer surplus"
        #  == Compute area above inverse supply curve, excluding tax == #
        integrand = lambda x: -( + (1/ * x
        area, error = quad(integrand, 0, self.quantity())  
        return (self.price() - * self.quantity() - area
    def taxrev(self):
        "Compute tax revenue"
        return * self.quantity()
    def inverse_demand(self,x):
        "Compute inverse demand"
        return - (1/* x
    def inverse_supply(self,x):
        "Compute inverse supply curve"
        return -( + (1/ * x +
    def inverse_supply_no_tax(self,x):
        "Compute inverse supply curve without tax"
        return -( + (1/ * x

Here’s a sample of usage

baseline_params = 15, .5, -2, .5, 3
m = Market(*baseline_params)
print("equilibrium price = ", m.price())
equilibrium price =  18.5
print("consumer surplus = ", m.consumer_surp())
consumer surplus =  33.0625

Here’s a short program that uses this class to plot an inverse demand curve and curves supply with and without tax

import matplotlib.pyplot as plt
import numpy as np

# Baseline ad, bd, az, bz, tax
baseline_params = 15, .5, -2, .5, 3
m = Market(*baseline_params) 

q_max = m.quantity() * 2
q_grid = np.linspace(0.0, q_max, 100)
pd = m.inverse_demand(q_grid)
ps = m.inverse_supply(q_grid)
psno = m.inverse_supply_no_tax(q_grid)    

fig, ax = plt.subplots()
ax.plot(q_grid, pd, lw=2, alpha=0.6, label='demand')
ax.plot(q_grid, ps, lw=2, alpha=0.6, label='supply') 
ax.plot(q_grid, psno, '--k', lw=2, alpha=0.6, label='supply without tax')
ax.set_xlabel('quantity', fontsize=14)
ax.set_xlim(0, q_max)
ax.set_ylabel('price', fontsize=14)
ax.legend(loc='lower right', frameon=False, fontsize=14)

The figure produced looks as follows


The next program provides a function that

  • takes an instance of Market as a parameter
  • computes dead weight loss from the imposition of the tax

def deadw(m):
    "Computes deadweight loss for market m."
    # == Create analogous market with no tax == #
    m_no_tax = Market(,,,, 0)   
    # == Compare surplus, return difference == #
    surp1 = m_no_tax.consumer_surp() + m_no_tax.producer_surp()  
    surp2 = m.consumer_surp() + m.producer_surp() + m.taxrev()
    return surp1 - surp2

Here’s an example of usage

baseline_params = 15, .5, -2, .5, 3
m = Market(*baseline_params)
deadw(m)  # Show deadweight loss

Example: Chaos

Let’s look at one more example, related to chaotic dynamics in nonlinear systems

One simple transition rule that can generate complex dynamics is the logistic map

(2)\[x_{t+1} = r x_t(1 - x_t) , \quad x_0 \in [0, 1], \quad r \in [0, 4]\]

Let’s write a class for generating time series from this model

Here’s one implementation, in file

class Chaos:
    Models the dynamical system with :math:`x_{t+1} = r x_t (1 - x_t)`
    def __init__(self, x0, r):
        Initialize with state x0 and parameter r 
        self.x, self.r = x0, r
    def update(self):
        "Apply the map to update state."
        self.x =  self.r * self.x *(1 - self.x)
    def generate_sequence(self, n):
        "Generate and return a sequence of length n."
        path = []
        for i in range(n):
        return path

Here’s an example of usage

ch = Chaos(0.1, 4.0) # x0 = 0.1 and r = 0.4
ch.generate_sequence(5)  # First 5 iterates
[0.1, 0.36000000000000004, 0.9216, 0.28901376000000006, 0.8219392261226498]

This piece of code plots a longer trajectory

import matplotlib.pyplot as plt

ch = Chaos(0.1, 4.0) 
ts_length = 250

fig, ax = plt.subplots()
ax.set_xlabel(r'$t$', fontsize=14)
ax.set_ylabel(r'$x_t$', fontsize=14)
x = ch.generate_sequence(ts_length)
ax.plot(range(ts_length), x, 'bo-', alpha=0.5, lw=2, label=r'$x_t$')

The resulting figure looks as follows


The next piece of code provides a bifurcation diagram

import matplotlib.pyplot as plt

fig, ax = plt.subplots()
ch = Chaos(0.1, 4)
r = 2.5
while r < 4:
    ch.r = r
    t = ch.generate_sequence(1000)[950:]
    ax.plot([r] * len(t), t, 'b.', ms=0.6)
    r = r + 0.005

ax.set_xlabel(r'$r$', fontsize=16)

Here is the figure it generates


On the horizontal axis is the parameter \(r\) in (2)

The vertical axis is the state space \([0, 1]\)

For each \(r\) we compute a long time series and then plot the tail (the last 50 points)

The tail of the sequence shows us where the trajectory concentrates after settling down to some kind of steady state, if a steady state exists

Whether it settles down, and the character of the steady state to which it does settle down, depend on the value of \(r\)

For \(r\) between about 2.5 and 3, the time series settles into a single fixed point plotted on the vertical axis

For \(r\) between about 3 and 3.45, the time series settles down to oscillating between the two values plotted on the vertical axis

For \(r\) a little bit higher than 3.45, the time series settles down to oscillating among the four values plotted on the vertical axis

Notice that there is no value of \(r\) that leads to a steady state oscillating among three values

Special Methods

Python provides special methods with which some neat tricks can be performed

For example, recall that lists and tuples have a notion of length, and that this length can be queried via the len function

x = (10, 20)

If you want to provide a return value for the len function when applied to your user-defined object, use the __len__ special method

class Foo:

    def __len__(self):
        return 42

Now we get

f = Foo()

A special method we will use regularly is the __call__ method

This method can be used to make your instances callable, just like functions

class Foo:

    def __call__(self, x):
        return x + 42

After running we get

f = Foo()
f(8)  # Exactly equivalent to f.__call__(8)

Exercise 1 provides a more useful example


Exercise 1

The empirical cumulative distribution function (ecdf) corresponding to a sample \(\{X_i\}_{i=1}^n\) is defined as

(3)\[F_n(x) := \frac{1}{n} \sum_{i=1}^n \mathbf{1}\{X_i \leq x\} \qquad (x \in \mathbb{R})\]

Here \(\mathbf{1}\{X_i \leq x\}\) is an indicator function (one if \(X_i \leq x\) and zero otherwise) and hence \(F_n(x)\) is the fraction of the sample that falls below \(x\)

The Glivenko–Cantelli Theorem states that, provided that the sample is iid, the ecdf \(F_n\) converges to the true distribution function \(F\)

Implement \(F_n\) as a class called ECDF, where

  • A given sample \(\{X_i\}_{i=1}^n\) are the instance data, stored as self.observations
  • The class implements a __call__ method that returns \(F_n(x)\) for any \(x\)

Your code should work as follows (modulo randomness)

from random import uniform

samples = [uniform(0, 1) for i in range(10)]
F = ECDF(samples)
F(0.5)  # Evaluate ecdf at x = 0.5
F.observations = [uniform(0, 1) for i in range(1000)]

Aim for clarity, not efficiency

Exercise 2

In an earlier exercise, you wrote a function for evaluating polynomials

This exercise is an extension, where the task is to build a simple class called Polynomial for representing and manipulating polynomial functions such as

(4)\[p(x) = a_0 + a_1 x + a_2 x^2 + \cdots a_N x^N = \sum_{n=0}^N a_n x^n \qquad (x \in \mathbb{R})\]

The instance data for the class Polynomial will be the coefficients (in the case of (4), the numbers \(a_0, \ldots, a_N\))

Provide methods that

  1. Evaluate the polynomial (4), returning \(p(x)\) for any \(x\)
  2. Differentiate the polynomial, replacing the original coefficients with those of its derivative \(p'\)

Avoid using any import statements


Exercise 1

class ECDF(object):

    def __init__(self, observations):
        self.observations = observations

    def __call__(self, x):
        counter = 0.0
        for obs in self.observations:
            if obs <= x:
                counter += 1
        return counter / len(self.observations)
# == test == #

from random import uniform
samples = [uniform(0, 1) for i in range(10)]
F = ECDF(samples)

print(F(0.5))  # Evaluate ecdf at x = 0.5

F.observations = [uniform(0, 1) for i in range(1000)]


Exercise 2

class Polynomial(object):

    def __init__(self, coefficients):
        Creates an instance of the Polynomial class representing

            p(x) = a_0 x^0 + ... + a_N x^N,

        where a_i = coefficients[i].
        self.coefficients = coefficients

    def __call__(self, x):
        "Evaluate the polynomial at x."
        y = 0
        for i, a in enumerate(self.coefficients):
            y += a * x**i
        return y

    def differentiate(self):
        "Reset self.coefficients to those of p' instead of p."
        new_coefficients = []
        for i, a in enumerate(self.coefficients):
            new_coefficients.append(i * a)
        # Remove the first element, which is zero
        del new_coefficients[0]
        # And reset coefficients data to new values
        self.coefficients = new_coefficients