Data Science Manager Interview Experience

Complete four notebook-style tasks cleaning messy data, summarizing it, and building a basic model in 90 minutes.

I moved quickly through messy data cleaning and manipulation, basic summary statistics, and a simple model with a train test split. I did really well on it, but the main challenge was pacing because 90 minutes goes fast and you cannot get stuck on any one notebook.

Do you want to evaluate a linear regression or a classification and regression tree, and can you walk me through what is wrong with this airline delay model?

I chose linear regression because I felt more comfortable with it. I walked through the airline delay example and pointed out that time of day should be categorical, not treated like a straight numeric variable, because 11:55 p.m. and 12:05 a.m. are close in reality but look far apart numerically. I also said if the real output is delay probability, I would switch to logistic regression. They also showed scatterplots and a correlation matrix, so I talked through multicollinearity and why I would revisit the feature setup.

How does this amusement park make money?

I started with ticket sales, then added premium rides, food and drinks, and event rentals as other revenue streams. From there I translated their assumptions about visit days, season passholders, and per-visit customers into basic algebra to estimate total revenue. After that they had me work down an income statement with sales, cost of goods sold, fixed costs, and things like depreciation to get to net income. The math itself was not hard, but you had to stay organized and keep track of a lot of numbers.

Write a short shell script to download a package, unzip it, and install it into an Anaconda environment.

I wrote a very short bash script using basic commands like curl or wget, unzip, and the install step into an Anaconda environment. It was only a few lines, so they were not looking for anything fancy, just whether I could handle straightforward command-line tasks.

Walk me through this Python script and class, explain what each function is doing, tell me what __init__ is, and say whether these pytest unit tests are sufficient.

They gave me a well-written Python script with a class and several methods, and I explained it more function by function than line by line. I answered the basic Python question that __init__ is the constructor method, then reviewed the pytest examples and said whether they covered enough. Where they did not, I proposed and wrote a simple extra unit test or two to cover the missing cases.

Review this poorly written script and point out the problems.

I walked through the script and called out the kinds of issues they wanted to see: repeated code, hardcoded values, weak variable names, and other basic maintainability problems. I did not have to rewrite it, just identify the issues clearly and explain why they were problems.