Are programming competitions a good use of time?

10 minutes remaining in the contest, but you’re still a few points short of advancing. Armed with your mighty coding powers, the first three problems fall quickly, but problem 4 is proving a tough nut to crack. After four incorrect attempts, your time is running short, and you’re searching desperately for an off-by-one error, an edge case you haven’t considered.

You read the problem statement one more time, and at last, you find it. An integer overflow bug. With a wide grin, you fix the bug with a few quick keystrokes. You upload the source code file…


Accepted! You sit back, feeling both relieved and elated. With the addition of 25 points, your advancement to the next round is now guaranteed.

It’s not hard to see why programming contests are so appealing — this is programming distilled to its essence. No need to figure out how to integrate an unfamiliar API, or refactor code to be unit testable, or make sense of vague business requirements. In competitive programming, each problem has a self-contained, algorithmic solution, and an automated judge instantly determines if it’s correct or not.

2Above: Programming contests contain the same types of problems as technical interviews at companies like Google.

Competitive programming promises even more glory. Win enough contests, and you get an interview at Facebook or Google, where they ask you… you guessed it… more algorithm coding questions!

By doing programming contests, you gain an intimate understanding of data structures and algorithms and their complexities. While your colleagues vaguely know the difference between a depth-first versus a breadst-first search, you develop a much deeper intuition. You will never forget that one contest where you used DFS instead of BFS, causing your solution to time out.

Unlike the real world, competitive programming offers an arena of pure meritocracy. As long as you solve difficult problems fast, you will surely rise through the ranks. In the Google Code Jam, thousands of programmers try their luck in the qualifying round, but this number is reduced to 3000 by Round 2, and 500 by Round 3. Then for the grand finale, the top 25 elite coders are flown in to compete on-site in the world finals.

3Above: 25 of the world’s best compete in the Google Code Jam world finals.

I used to look up in awe at red coders (the highest rated users have red usernames). By the time I solved the first problem, they would have not only solved it in 10 minutes, but also solved 2-3 even harder ones. What was it like to think at that level, to possess that much coding wizardry?

And for some time, I strove to be like them. I studied my dynamic programming and graph algorithms in my spare time, and practiced on SPOJ and Hackerrank and Codeforces. I competed in my university’s ACM tryouts, and three times failed to qualify for the ACM team. So it goes.

In the last few years, I got to talk to a lot of competitive programmers, most of whom were far better than myself. Perhaps I was searching for some magical quality that gave them coding superpowers, but none was to be found. Instead, the key to a high rating was simply many years of dedication and hard work.

It’s illuminating to read the answers on this Quora question: “How does it feel to finally be red at TopCoder after years of hard work?” The short answer: nothing much really.

4Above: Rating graph of Codeforces user netman. Getting to red takes years of practice.

Given the amount of time it takes to master competitive programming, one naturally wonders: is this really a good use of time? In a contest, you are ultimately solving problems that other people have solved already, so nothing new is being produced. Although solving a contest problem is satisfying, I find it a lot more rewarding to build projects or apps with my novel ideas.

Recently, Google found that among its engineers, being good at programming competitions is negatively correlated to being good at software engineering.

In the video, Peter Norvig notes that competitive programmers are “used to going really fast, cranking the answer out and moving to the next thing, but you do better if you’re more reflective and go slowly and get things right”.

Ironically, the same thing that makes programming contests so attractive is its own downfall. Contests focus on data structures and algorithms, which are just a small part of software engineering. Other skills like UI design, databases, network architecture, distributed systems, etc, are not touched in programming contests.

Even if you only look at algorithmic problems, competitive programming is still not representative of reality. Due to limitations in automated judging, contest problems are limited to exact, deterministic algorithms that have a single correct answer. This rules out entire classes of randomized and approximate algorithms. Algorithms now rely more and more on data and machine learning, and less on combinatorial approaches, which further renders competitive programming less relevant.

Now, are programming contests useful? Yes, but only up to a point. Solving contest problems is an excellent way to familiarize yourself with a programming language and its data structures, as well as get better at converting procedural ideas to code. These are very useful skills for a coding interview. However, even the most difficult Facebook/Google interview questions are maybe around a Codeforces Div2 C (or Div1 A) difficulty, which is a long way from the hardest contest problems.

5Above: Beyond a certain point, skills learned in programming contests are only useful for programming contests.

I would put the inflection point at about 1700 Codeforces rating (enough to solve Div2 A and B consistently). Beyond that, you continue to improve, but be aware that you’ll be studying things solely for contests that have little relevance anywhere else, for example, Fenwick trees, max flow algorithms, bitmask DP, and other increasingly obscure topics.

So far, I’ve been fairly critical of competitive programming, but rather than deride it as a waste of time, I think it’s best to view it as a sport. Like soccer or basketball, the function of sports in society is to inspire excellence, and above all, to have fun. Terry Tao wrote a similar article on math competitions; I’d agree with him.

My advice to you: do programming contests if you find them fun and you enjoy tackling hard problems. But don’t take it too seriously: it takes years of dedicated effort to get extremely good at it, dedication that very few people have. Unless you’re at or aiming to compete at the World Final level, you definitely shouldn’t be spending a significant amount of time getting better at contests. Your time is better spent studying machine learning, or statistics, or compilers, or distributed systems, or just about anything else in computer science.

I have accounts on Hackerrank and Codeforces if you want to follow me. As per my own advice, I’m no longer actively trying to get better, but I still do problems for fun occasionally.

Edit: This article has some interesting discussion on Reddit and Codeforces.

5 thoughts on “Are programming competitions a good use of time?

  1. For what it’s worth, the study Peter Norvig is citing in the video is about a decade old, and interpreting the results in this way is a mistake. Google hired heavily based on competitive programming results, and then checked among its employees to see whether, among those people, competitive programming was a predictor of success. It wasn’t, because the other people they’d hired were exceptional in other ways.

    Another problem with interpreting that study this way is that the feature they used was “did this person participate in a contest?” and not “did this person do well in contests?”

    There are also problems with using machine learning for attribution like this. Suppose that people who did well in contests also did well in coding questions on interviews. Then much of the “contestants are good” signal would be captured by the “people who did well in interviews” feature in the model.

    Peter Norvig has been citing this example for years. I don’t believe he has actually taken the time to understand why the results were the way they were.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s