Skip to content

How to Survive Your Data Science Interview

Interviews are scary as shit. You sit across the table from someone who has the power to grant you income and measure of security. They hold your future in their hands. You have to make them like you, trust you, think you are smart. There are fewer situations in life that induce greater anxiety. Luckily, there are some things you can do to make it a little gentler on you.


Preparing for the interview

Programming language textbooks

There are many wonderful things about data science. It’s extreme breadth is not one of them. The title of data scientist means something different at every company. To some it means PhD statistician. To some it means proficiency in Excel. To some it means machine learning generalist. To some it means being handy with Spark and Hadoop. Read job postings carefully for specific skills, tools and languages. A well written posting will give you a lot of insight into what they are looking for.

You have a limited amount of time to prepare. This is your budget, and you want to spend it so as to get the biggest bang for your buck. You have a few options:

  • BEST: If you have a gut feeling for what you need to prepare, go with that.
  • SECOND BEST, BUT STILL AWESOME: If the recruiter has given you advice, follow that.
  • LAST RESORT: If you have no idea where to start, I recommend one of the following.


Pocket Plan (10 hours)

This is only enough time to dust off things you already know. Compare your current level of comfort to the job posting. Are there some skills you haven’t used in five years? Some terms you don’t recognize? Where are your biggest gaps? Spend more of your time on these.

Here’s how you can loosen your foundational skills in their holster:

  1. Coding
    • Flip through a Cookbook (for example). Read through 6 entries in foundational areas like data structures and string manipulation.
    • Write out and explain programs that solve two practice problems.
  2. Data engineering
    • Flip through a SQL cookbook (for example). Read through 6 entries on foundational areas like conditional WHEREs and ways to combine tables.
    • Write out and explain SQL queries for three practice problems.
  3. Probability and Statistics
    • Flip through the first half of an introductory textbook. Read all the key definitions and equations.
    • Write out and explain the answers to two practice problems.
  4. Machine learning and algorithms

If you need ideas for practice problems, GlassDoor and LeetCode are helpful resources.

You will get the biggest benefit if you do all of your practicing OUT LOUD. Explain your answers to your cat, or to an empty chair. Use a pen and paper, or better yet, a whiteboard. These help recreate common interviewing environments in small ways so that when you get there it feels a bit more familiar.

Don’t panic if you encounter topics and tools you’ve never heard of before. Many job postings are written as wishlists. They sound like a fifteen year old describing their perfect mate—a billionaire celebrity-lookalike winner of the Nobel Prize in physics and the Peace Prize. Those are all fine attributes in a partner, but most of us would be pretty excited about finding one or two of them. No one has them all. In my experience, the candidates hired are strong in some of the points listed in the post, but not necessarily all of them.

Also, don’t forget to leave time for a fifth area: learning about the company. Visit the their web page. Get a sense of how they make their money and who their customers are. Read their engineering blog to learn what tools they use and how their infrastructure is built. Learn the name of the CEO. If you are lucky enough to get your interviewers’ names, Google their professional activities. Learn about their research interests. Get a feeling for what matters to them.


Standard Plan (40 hours)

If Pocket Plan preparation is a single pancake, Standard Plan is the tall stack – the same process, repeated, each pass going a little deeper than the last. You work more of the same example problems and read through more of the same references. Where the Pocket Plan only gives you enough to time to dust off your skills, the Standard lets you put a sharp edge on them. You can learn about things that you’ve heard of but never absorbed.

After your first pass through the Pocket Plan, take a step back and look at where you feel the least prepared. For instance, if the job posting says “SQL required” and the recruiter told you that you will be asked a couple of SQL questions, but you don’t know any SQL, that’s a gap. Make a wishlist of things you would like to dig into deeper and start with the biggest gap. Then go back to the Pocket Plan and allocate your time to practice problems and reference reading accordingly.


Deluxe Plan (100+ hours)

If you have the luxury of buying the deluxe option, congratulations! You have enough time to acquire new skills from scratch. The way to do this is to start with a Standard Plan-style gap analysis, but instead of working practice problems you get to bring out the big guns: microprojects and practice interviews.

Microprojects. A microproject is a small-scale project, no more than 10 hours, start-to-finish. They will set you on the road to mastering entirely new skills and concepts. Here’s how they work.

  1. Selection. Choose a project that requires you to use one or two of your gap skills. This is harder and it might seem, but try not to overthink it. The use case can be ridiculous. The result can be trivial. The only requirement is that it demands the skills you are trying to learn
  2. Execution. This is the fun part. In order to build your tiny project, you will have to gain a working knowledge of some small slice of whatever it is you’re trying to learn. There will be a lot of Internet searches, trial and error, and possibly tears. This is all very good. Self directed learning can be painful, but it is the fastest way to grow new skills. Don’t give up.
  3. Stopping. Projects have a tendency to take on a life of their own and are easy to fall in love with. Force yourself to stop after about ten hours. Otherwise you run the risk of neglecting your other gaps. If you can’t make yourself kill it, at least put it on hold until after your interview. Then you can come back and adopt it as a passion project.
  4. Reporting. You’re not quite done until you explain your project out loud. What problem are you trying to solve? What approach did you use? How did it work? What did you learn? What would you do differently next time? This is the best practice for an interview that you are ever going to get.


Practice interviews. Unfortunately, the best way to learn what a company wants in a data scientist is to interview with them. It’s not uncommon to interview with the same employer several times over the course of a few years. This is admittedly an expensive data gathering process, but it pays handsome rewards to the patient candidate.

If you want a faster version of this process, use interviews with other companies for practice. This may not tell you everything you need to know about your target company, but it will build out your interviewing skills faster than anything else I know.


Interview Day

A Special Forces Assessment and Selection candidate conducts training at the Nasty Nick obstacle course Camp Mackall in Hoffman, N.C., September 2009 091009-A-GV060-075

I wish I could guarantee that if you prepare thoroughly enough, you’ll nail your interview and get an offer. Unfortunately the truth is bleaker than that. But don’t let it get you down! Having your eyes wide open will help you survive your data science interview with style and grace.

Bleak truth number one: You probably won’t get an offer. On average, we each interview more than once before getting an offer. More than likely any given interview will result in a rejection.

SolutionNever do a “real” interview. If every interview is practice for the next one, you never have to deal with the additional pressure of “I must land this,” and it’s no big deal if you don’t get an offer. When you approach it this way, a strange thing can happen. Unexpectedly one of your practice interviews goes so well that you find yourself holding an offer to work with a bunch of great people on a fun problem.

Bleak truth number two: your interviewers are biased. It’s because they’re humans. There are common biases like gender, race, age and sexual orientation. There are also idiosyncratic biases such as voice timbre, alma mater and which text editor you use. When they are conscious and intentional, biases are a moral failure, but unconscious biases are unavoidable. We are hard wired for them.

Solution: There’s nothing you can do about this, so don’t try. You can’t anticipate it, you can’t compensate for it, and you don’t want to.

Bleak truth number three: Your own brain will try to sabotage you. It is hard not to get tripped up by the voices in your own head. They’ll chant things like “Everyone here looks smarter than me”, “I shouldn’t have said that”, “I should have learned more about Bayes Theorem”.

Solution: You can’t ignore your voices (I’ve tried) or make them go away (I’ve tried that too). Just notice them. They are not telling you the truth or helping you, but, for better or worse, they are part of you. Let them clamor in the background and go about your business.

Bleak truth number four: You probably don’t want to work for this company. Any group of people, whether it is a football team, a company, a rock band or a family, is quirky (not to say dysfunctional) in its own way. The trick is to find a company whose quirks are compatible with our own. Unfortunately, quirks are by definition unusual, and finding a complementary set is tough. It’s like finding a climbing partner or a good hat. You have try on a few to find one that works.

Solution: When you’re interviewing, watch for things you like and things you don’t. Listen to how you feel. If you look around and find it easy to picture yourself in this place, working with these people, that’s a good sign. If you find yourself thinking “I could make this work”, proceed with caution. If you can’t wait to get out, that’s a strong no.

Heads up: Wherever there is a great power disparity, there will always be a few who exploit it. After your interview, you should feel like you just completed a workout where a personal trainer pushed you to your limits. You should not feel like you were made to dance and grovel for someone else’s pleasure. If this was the case, that is a clear signal. You do not want to be connected to a culture that permits small abuses like this. They fester and foster large abuses as well. Luckily, this isn’t common, but it is worth keeping an eye out for.

Be you

MCCC 15 - University of Michigan Stormtrooper (17473982183)

On the bright side, by the time you get to your interview, most of your work is done. All you need to do is take a shower and show up on time. Get some sleep (if you can), eat a good breakfast (if you can keep it down) and you are good to go.

Forget about checklists of things to say. Don’t bother with tricks like mirroring and power poses. Don’t pretend to be hyper enthusiastic. Just be you. If you are thirsty, ask for a drink. If you are curious about something, ask about it. If something is funny to you, laugh. If you are confused, say so. You will be asked questions you don’t know how to answer. That’s on purpose. Feel free to admit it and to ask for clarification. Once you let go of trying to remember all the things you are supposed to be, you’ll have more mental energy for working out code on the whiteboard.

It’s in your interest to give your potential employer an accurate impression of who you are. Truth in advertising means you don’t have to keep up false pretenses. If you are a passionate Yankees fan, it’s OK to mention it. If you have an abiding fondness for image processing, share that. If you are an unrepentant asshole, let your interviewer see it. You will want to be with people who know how to deal with that. If you would rather have your fingernails pulled off than pair program, tell your interviewer. If you’ve only written two lines of C++ code in your life, it’s OK to say so. Avoid trash talking anyone, including your past self, but be straightforward. Pretending to be anything other than you are doesn’t help you and it doesn’t help your interviewer.

It’s hard to overstate to power of being genuinely you. It is magnetic. It gives you an air of strength and makes you interesting. It’s kind of scary, so few of us do it. As a result, it tends to make you stand out in the minds of your interviewers. They may or may not make you an offer, but whatever the case, it will be the best outcome for you.

I wish you luck as you travel along your path. May it take you somewhere even better than you planned.

Source: How to Survive Your Data Science Interview


One does not discover new lands without consenting to lose sight of the shore for a very long time.

André Gide

It takes something more than intelligence to act intelligently.

Fyodor Dostoevsky

Contradictions do not exist. Whenever you think you are facing a contradiction, check your premises. You will find that one of them is wrong.

Ayn Rand

The ability to quote is a serviceable substitute for wit.

W. Somerset Maugham

How Formula One Teams Are Using Big Data To Get The Inside Edge

Since joining the Formula One circuit in 2005, Red Bull Racing has won four championships. (Photo credit: Wikipedia)

The back room in the Infiniti Red Bull Racing garage at the Formula One U.S. Grand Prix looks more like a NASA mission control room than it does a garage.

There are more computer monitors than there are eyeballs in the room, where three days before the race, several engineers were already in their seats, running simulations and studying its stream of data.

If the weekend’s race was going to be anything like the Brazilian Grand Prix two years ago, they knew that they had to be prepared, as their performance could mean winning or losing a championship.

The data that the engineers were scouring over at the time was just a fraction of the hundreds of gigabytes of data produced from each car over race weekend – the exact number, a team secret.

What isn’t a secret, however, is the use of big data analytics in Formula One, a sport as technically demanding as building a rocket ship.

Race teams at the U.S. Grand Prix collected more than 243 terabytes of data earlier this month according to AT&T T -0.46%, a few terabytes more data than there are in the Library of Congress.

Using the data, teams are trying to predict where they’ll finish the race before it even starts, based off the data they’ve amassed over the season from their competition and from their own car.

Hundreds of sensors provide thousands of data points, from tire pressure to fuel burn efficiency, in real-time that are then analyzed by race engineers onsite. Live race data is also transferred in under a quarter of a second through thousands of miles of fiber optic cables on AT&T’s backbone to another fully staffed operations room at team headquarters in the United Kingdom.

“We’re looking two hours into the future and trying to predict where we’ll finish the race before it starts,” said Al Peasland, head of technical partnerships at Infiniti Red Bull Racing. But things rarely ever stick to plan.

Formula One cars generate terabytes of data at each race.

In 2012, Red Bull Racing and its driver, Sebastian Vettel, entered the final race of the season, needing only a top-3 finish to win their third straight world championship. But on the first lap, Vettel was clipped from behind, damaging his car and sending him into a spin that put him in last place.

“Immediately, we thought race over,” said Peasland.

But within a second of the accident, the engineers snapped into their action plan, first checking the car’s telemetry data to determine if the car would even have a chance of finishing the race.

By the end of the first lap, the crew knew the extent of the damage. The car, while drivable, had become unbalanced and a failure to address the issue would result in tire and engine damage.

It took a few additional laps for Red Bull’s engineers to finish their simulations and come up with a temporary fix. And by their first scheduled pit stop, 10 laps into the 71-lap race, the crew was not only able to rebalance the car, but they were also able to devise a new race strategy that was just enough to secure the minimum number of points to win the world championship.

“To win with such a fine margin and to have an accident like that at the very beginning and for the whole team to come together, it’s the race that films are made of,” said Peasland.

Two years later, Infiniti Red Bull Racing is collecting three times as much data as they were in Brazil. The team has also expanded their operations room to include more engineers and data analysts at their headquarters in Milton Keynes, U.K.

“Data is so good for drivers that the pit crew could literally coach a driver to the most optimal racing conditions based on a live data feed of everything imaginable,” said Tony Jardine.

Currently a Formula One analyst for Sky Sports in the U.K., Jardine has almost 20 years of experience in the garage, including as an engineer and a team manager. He’s seen Formula One technology evolve from drawing boards to basic computers in the 1990s to the analysis and massive data collection today.

“It’s a data driven business and always has been,” said Matt Cadieux, chief technology officer for Infiniti Red Bull Racing. “But the sophistication and speed of race simulations have improved just in the last few years, whereas races used to rely more on gut feel and experience.”

Now those simulations take hours not weeks and simpler simulations take seconds not minutes.

“It’s about making educated decisions efficiently,” said Peasland, head of technical partnerships. “And that comes from measuring the right information.”

Source: How Formula One Teams Are Using Big Data To Get The Inside Edge

Those who appeal to you to hold onto the past do so at the cost of denying you your future

Lyndon Johnson