Functional programming

Functional programming refers to a programming paradigm which avoids changes in program state as much as possible. Variables are generally avoided. Instead, chains of function calls form the backbone of the program.

Lambda expressions and different types of comprehensions are common techniques in the functional programming style, as they let you process data without storing it in variables, so that the state of the program does not change. For example, a lambda expression is for all intents and purposes a function, but we do not need to store a named reference to it anywhere.

As mentioned above, functional programming is a programming paradigm, or a style of programming. There are many different programming paradigms, and we've already come across some of them:

• imperative programming, where the program consists of a sequence of commands which is executed in order
• procedural programming, where the program is grouped into procedures or sub-programs
• object-oriented programming, where the program and its state is stored in objects defined in classes.

There are differing opinions on the divisions between the different paradigms; for example, some maintain that imperative and procedural programming mean the same thing, while others place imperative programming as an umbrella term which covers both procedural and object-oriented programming. Th terminology and divisions are not that important, and neither is strictly sticking to one or the other paradigm, but it is important to understand that such different approaches exist, as they affect the choices programmers make.

Many programming languages are designed with one or the other programming paradigm in mind, but Python is a rather versatile programming language, and allows for following several different programming paradigms, even within a single program. This lets us choose the most efficient and clear method for solving each problem.

Let's have a look at some functional programming tools provided by Python.

map

The map function executes some operation on each item in an iterable series. This sounds a lot like the effect a comprehension has, but the syntax is different.

Let's assume we have list of strings which we want to convert into a list of integers:

str_list = ["123","-10", "23", "98", "0", "-110"]

integers = map(lambda x : int(x), str_list)

print(integers) # this tells us the type of object we're dealing with

for number in integers:
    print(number)

The general syntax for the map function is

map(<function>, <series>)

where function is the operation we want to execute on each item in the series.

The map function returns an object of type map, which is iterable, and can be converted into a list:

def capitalize(my_string: str):
    first = my_string[0]
    first = first.upper()
    return first + my_string[1:]

test_list = ["first", "second", "third", "fourth"]

capitalized = map(capitalize, test_list)

capitalized_list = list(capitalized)
print(capitalized_list)

As you can see from the examples above, the map function accepts both an anonymous lambda function and a named function defined with the def keyword.

We could achieve the same result with a list comprehension:

def capitalize(my_string: str):
    first = my_string[0]
    first = first.upper()
    return first + my_string[1:]

test_list = ["first", "second", "third", "fourth"]

capitalized_list = [capitalize(item) for item in test_list]
print(capitalized_list)

...or we could go through the original list with a for loop and save the processed items in a new list with the append method. Typically, in programming there are many different solutions to each problem. There are rarely any absolutely right or wrong answers. Knowing many different approaches helps you choose the most appropriate one for each situation, or one that best suits your own tastes.

It is worth pointing out that the map function does not return a list, but an iterator object of type map. An iterator behaves in many ways like a list, but there are exceptions, as can be seen in the following example:

def capitalize(my_string: str):
    first = my_string[0]
    first = first.upper()
    return first + my_string[1:]

test_list = ["first", "second", "third", "fourth"]

# store the return value from the map function
capitalized = map(capitalize, test_list)

for word in capitalized:
  print(word)

print("print the same again:")
for word in capitalized:
  print(word)

This would print out the following:

Above we tried to print out the contents of the map iterator twice, but the second attempt produced no printout. The reason is that map is an iterator; passing through it with a for loop "depletes" it, much like a generator is depleted once its maximum value is reached. Once the items in the iterator have been traversed with a for loop, there is nothing left to go through.

If you need to go through the contents of a map iterator more than once, you could, for example, convert the map into a list:

test_list = ["first", "second", "third", "fourth"]

# convert the return value of the map function into a list
capitalized = list(map(capitalize, test_list))

for word in capitalized:
  print(word)

print("print the same again:")
for word in capitalized:
  print(word)

The map function and your own classes

You can naturally also process instances of your own classes with the map function. There are no special gimmicks involved, as you can see in the example below:

class BankAccount:
    def __init__(self, account_number: str, name: str, balance: float):
        self.__account_number = account_number
        self.name = name
        self.__balance = balance

    def deposit(self, amount: float):
        if amount > 0:
            self.__balance += amount

    def get_balance(self):
        return self.__balance

a1 = BankAccount("123456", "Randy Riches", 5000)
a2 = BankAccount("12321", "Paul Pauper", 1)
a3 = BankAccount("223344", "Mary Millionaire ", 1000000)

accounts = [a1, a2, a3]

clients = map(lambda t: t.name, accounts)
for name in clients:
  print(name)

balances = map(lambda t: t.get_balance(), accounts)
for balance in balances:
  print(balance)

Here we first collect the names of the account holders with the map function. An anonymous lambda function is used to retrieve the value of the name attribute from each BankAccount object:

clients = map(lambda t: t.name, accounts)

Similarly, the balance of each BankAccount is collected. The lambda function looks a bit different, because the balance is retrieved with a method call, not from the attribute directly:

balances = map(lambda t: t.get_balance(), accounts)

filter

The built-in Python function filter is similar to the map function, but, as the name implies, it doesn't take all the items from the source. Instead, it filters them with a criterion function, which is passed as an argument. If the criterion function returns True, the item is selected.

Let's look at an example using filter:

integers = [1, 2, 3, 5, 6, 4, 9, 10, 14, 15]

even_numbers = filter(lambda number: number % 2 == 0, integers)

for number in even_numbers:
    print(number)

It might make the above example a bit clearer if we used a named function instead:

def is_it_even(number: int):
    if number % 2 == 0:
        return True
    return False

integers = [1, 2, 3, 5, 6, 4, 9, 10, 14, 15]

even_numbers = filter(is_it_even, integers)

for number in even_numbers:
    print(number)

These two programs are functionally completely identical. It is mostly a matter of opinion which you consider the better approach.

Let's have a look at another filtering example. This program models fishes, and selects only those which weigh at least 1000 grams:

class Fish:
    """ The class models a fish of a certain species and weight """
    def __init__(self, species: str, weight: int):
        self.species = species
        self.weight = weight

    def __repr__(self):
        return f"{self.species} ({self.weight} g.)"

if __name__ == "__main__":
    f1 = Fish("Pike", 1870)
    f2 = Fish("Perch", 763)
    f3 = Fish("Pike", 3410)
    f4 = Fish("Cod", 2449)
    f5 = Fish("Roach", 210)

    fishes = [f1, f2, f3, f4, f5]

    over_a_kilo = filter(lambda fish : fish.weight >= 1000, fishes)

    for fish in over_a_kilo:
        print(fish)

We could just as well use a list comprehension and achieve the same result:

over_a_kilo = [fish for fish in fishes if fish.weight >= 1000]

The return value of filter is an iterator

The filter function resembles the map function in also that it returns an iterator. There are situations where you should be especially careful with filter as iterators can only be traversed once. So, trying to print out the collection of large fishes twice will not work quite as straightforwardly as you might think:

f1 = Fish("Pike", 1870)
f2 = Fish("Perch", 763)
f3 = Fish("Pike", 3410)
f4 = Fish("Cod", 2449)
f5 = Fish("Roach", 210)

fishes = [f1, f2, f3, f4, f5]

over_a_kilo = filter(lambda fish : fish.weight >= 1000, fishes)

for fish in over_a_kilo:
    print(fish)

print("print the same again:")

for Fish in over_a_kilo:
    print(Fish)

This would print out the following:

If you need to go through the contents of a filter iterator more than once, you could convert the result into a list:

fishes = [f1, f2, f3, f4, f5]

# convert the return value of the filter function into a list
over_a_kilo = list(filter(lambda fish : fish.weight >= 1000, fishes))

reduce

A third cornerstone function in this introduction to functional programming principles is reduce, from the functools module. As the name implies, its purpose is to reduce the items in a series into a single value.

The reduce function starts with an operation and an initial value. It performs the given operation on each item in the series in turn, so that the value changes at each step. Once all items have been processed, the resulting value is returned.

We have done summation of lists of integers in different ways before, but here we have an example with the help of the reduce function. Notice the import statement; in Python versions 3 and higher it is necessary to access the reduce function. In older Python versions the import statement was not needed, so you may come across examples without it online.

from functools import reduce

my_list = [2, 3, 1, 5]

sum_of_numbers = reduce(lambda reduced_sum, item: reduced_sum + item, my_list, 0)

print(sum_of_numbers)

Let's take a closer look at what's happening here. The reduce function takes three arguments: a function, a series of items, and an initial value. In this case, the series is a list of integers, and as we are calculating a sum, a suitable initial value is zero.

The first argument is a function, which represents the operation we want to perform on each item. Here the function is an anonymous lambda function:

lambda reduced_sum, item: reduced_sum + item

This function takes two arguments: the current reduced value and the item whose turn it is to be processed. These are used to calculate a new value for the reduced value. In this case the new value is the sum of the old value and the current item.

It may be easier to comprehend what the reduce function actually does if we use a normal named function instead of a lambda function. That way we can also include helpful printouts:

from functools import reduce

my_list = [2, 3, 1, 5]

# a helper function for reduce, adds one value to the current reduced sum
def sum_helper(reduced_sum, item):
  print(f"the reduced sum is now {reduced_sum}, next item is {item}")
  # the new reduced sum is the old sum + the next item 
  return reduced_sum + item

sum_of_numbers = reduce(sum_helper, my_list, 0)

print(sum_of_numbers)

The program prints out:

First, the function takes care of the item with value 2. To begin with, the reduced sum is 0, which is the initial value passed to the reduce function. The function calculates and returns the sum of these two: 0 + 2 = 2.

This is the value stored in reduced_sum as the reduce function processes the next item on the list, with value 3. The function calculates and returns the sum of these two: 2 + 3 = 5. This result is then used when processing the next item, and so forth, and so forth.

Now, summation is simple, as there is even the built-in sum function for this purpose. But how about multiplication? Only minor changes are needed to create a reduced product:

from functools import reduce

my_list = [2, 2, 4, 3, 5, 2]

product_of_list = reduce(lambda product, item: product * item, my_list, 1)

print(product_of_list)

As we are dealing with multiplication the initial value is not zero. Instead, we use 1. What would happen if we used 0 as the initial value?

Above we have dealt largely with integers, but map, filter and reduce can all handle a collection of objects of any type.

As an example, let's generate a sum total of the balances of all accounts in a bank, with the help of reduce:

class BankAccount:
    def __init__(self, account_number: str, name: str, balance: float):
        self.__account_number = account_number
        self.name = name
        self.__balance = balance

    def deposit(self, amount: float):
        if amount > 0:
            self.__balance += amount

    def get_balance(self):
        return self.__balance

a1 = BankAccount("123456", "Randy Riches", 5000)
a2 = BankAccount("12321", "Paul Pauper", 1)
a3 = BankAccount("223344", "Mary Millionaire ", 1000000)

accounts = [a1, a2, a3]

from functools import reduce

def balance_sum_helper(balance_sum, account):
  return balance_sum + account.get_balance()

balances_total = reduce(balance_sum_helper, accounts, 0)

print("The total of the bank's balances:")
print(balances_total)

This program would print out:

The balance_sum_helper function grabs the balance of each bank account, with the method dedicated for the purpose in the BankAccount class definition:

def balance_sum_helper(balance_sum, account):
  return balance_sum + account.get_balance()

NB: if the items in the series are of a different type than the intended reduced result, the thrd argument is mandatory. The example with the bank accounts would not work without the initial value. That is, trying this

balances_total = reduce(balance_sum_helper, accounts)

would produce an error:

TypeError: unsupported operand type(s) for +: 'BankAccount' and 'int'

In the above case, when reduce tries to execute the balance_sum_helper function for the first time, the arguments it uses are the two first items in the list, which are both of type BankAccount. Specifically, the value assigned to the parameter balance_sum is the first item in the list. The balance_sum_helper function tries to add an integer value to it, but adding an integer directly to a BankAccount object is not a supported operation.

You can check your current points from the blue blob in the bottom-right corner of the page.