26. Building abstractions with recursive functions and higher-order functions

In programming, we control the intellectual complexity of our programs by building abstractions that hide details when appropriate. You already know of one such means to build abstractions in Python: the ability to define your own functions. This allows you to chunk compound operations as conceptual units, give them a name and manipulate them. In this module, you will explore two other means to build abstractions with functions: recursive functions and higher-order functions.

26.1. Recursive functions

1. Watch the introduction video and read the following summary paragraph.

Summary. Recursive definitions allow us to describe many complex concepts in a simple and elegant way. Likewise, recursive functions in Python allow us to define methods for solving problems in a simple and elegant way. When doing so, we think in terms of base cases, for which there is an immediate solution, and recursive cases, for which we can define the solution to our problem in terms of solutions to smaller versions of the same problem. Once the solution has been defined in such a way, translating it into Python code is a snap. This frees us from without having to worry about a whole series of questions that would arise if we were to solve the problem iteratively (how to traverse sub-problems, how to keep track of intermediate results, what the iteration variable is, what the stop conditions are…). In other words, we have abstracted the solution from the nitty-gritty details of control flow, memory, etc.

2. Read section 16.1 to 16.6 of the interactive book thinkcspy at: https://runestone.academy/runestone/books/published/thinkcspy/IntroRecursion/toctree.html

Sections 16.5. and 16.6 use the turtle graphics module which you may not be familiar with, but you should still be able to grasp the code easily. You can skip them if you want, but visualization is a powerful way to learn about recursion.

3. Do some of the following practice exercises.

You can move to the next section whenever you feel ready to learn more, but make sure to do at least two exercises (like any other programming concepts/techniques, recursive functions cannot be understood without practice).

Time to complete. [*]: short, [**]: medium, [***]: long.

Exercise 1. [*] Write a recursive function to reverse a list. (Hint: to concatenate two lists, use the ‘+’ operator).

Exercise 2. [*] Write a recursive function to reverse a list.

Exercise 3. [**] Write a script that displays the Koch snowflake (see on wikipedia) using a recursive function. You can use the turtle graphics module to draw on screen (as shown in section 16.5 of thinkcspy).

A solution: link to code.

Exercise 4. [**] Write a script that returns the pathnames of all the files ending in .txt contained inside a directory (at any depth of the hierarchy). You will need to use os.listdir() and os.path.isdir().

Exercise 5. [**] Write a recursive function to generate all permutations of a list of values. (This could be useful, e.g., to do a permutation test for statistical analyses.)

Exercise 6. [***] Write a recursive function to evaluate an arithmetic expression given in Polish notation (i.e. prefix notation, see on wikipedia)

Exercise 7. [***] Produce strings from the MIU formal grammar and investigate the MU puzzle. See detailed explanations on the MU Puzzle.

Side remark: in some cases, recursive functions can take a prohibitive amount of time and/or memory. This happens when the same computations are performed many times. This issue can be addressed using memoization. If you are curious, you can read the article Memoization in Python

26.2. Higher-order functions

1. Watch the introduction video.

2. Read section 1.6 Higher-Order Functions of the online textbook at: https://wizardforcel.gitbooks.io/sicp-in-python/content/6.html

This textbook is an adaptation for Python of the textbook cited at the bottom of the page. (Author’s note: I haven’t checked out the rest of this adaptation. I recommend to read the original book if you want to learn more.)

Stop and think: In the section of the book you have read, the functions summation and iter_improve were implemented iteratively, using a while statement. Can you think of how to implement them recursively?

3. Do some of the following practice exercises.

Time to complete. [*]: short, [**]: medium, [***]: long.

Exercise 1. [*] Write a product function to compute the product of the values of a function at points over a given range (the function should be an argument of product). Then, define a function to compute the factorial in terms of product. Finally, use product to compute an approximation to pi using the Wallis product formula (see on wikipedia).

Exercise 2. [**] The summation function (from the textbook section you read) and the product function (defined above) can be defined as special cases of an even more general reduce function which has two more arguments: binary_operator (which specifies what operation should be applied to reduce all the values to a single one) and neutral_element (which specifies what base value should be used when the range is empty). Write such a reduce function, and then define summation and product functions in terms of reduce. Make sure to test your functions.

Exercise 3. [**] Write a function that finds a fixed point of a function (see on wikipedia) starting from an initial guess value, which could be defined as: find_fixed_point(f, initial_value, error_margin). The returned value, x, should be such that abs(f(x) - x) < error_margin. Then, define a function to calculate the square root of a positive number, using the property that the square root of x is a fixed point for the function y -> 1/2 (y + x/y). Make sure to test your square root function.

Exercise 4. [**] Write a function to numerically compute any statistic of an arbitrary random variable. The statistic and the random variable should both be functions which are given as argument.

(Hints. A statistic can be defined as a function of a collection of samples, e.g. sample mean, sample variance. A random variable can be defined as a function that, when it is called, generates one sample.)

26.3. Reference

This module was inspired by: Abelson, Harold, and Gerald Jay Sussman. Structure and interpretation of computer programs. The MIT Press, 1996.

It is an excellent computer science textbook. If you are curious, go check it out, it is freely available online as pdf and as a web document.