You’ve made it past the initial phone screen for a data science job (congrats!), and now they’ve given you a take-home project. It’s your chance to blow their minds, knock their socks off, convince them that you’re a slam-dunk hire—but where to start? When tackling a take-home project, it’s important to be strategic so that you can deliver your best work in the time allotted.
Not everybody is a fan of take-home projects, especially if you won’t be paid for the time you spend. I attended a panel recently where one presenter said he always declines to do take-home projects and instead offers to discuss over the phone how he would approach the project. How you respond to a request to do a take-home project is up to you, of course. Personally, I have found them to be a good opportunity to show that I really know how to do the things I say I know how to do. And luckily, I have had the time to spend on them.
In this post I’ll walk you through the process of planning and executing a take-home project. Ready, set…read on!
Before you start
Whether the project prompt asks you to do something you do every day or something you’ve never even heard of, you may be tempted to start working right away. STOP! Before you dive in, it’s very important to a) make sure you understand exactly what you’re being asked to do and b) think about what you can reasonably accomplish in the time allowed. You wouldn’t want to reach the deadline only to realize you’ve been trying to answer a question that nobody asked!
Read the prompt. Read it again. Make notes on the requested deliverables, any specific questions you need to answer, or any techniques you’re being asked to demonstrate. Determine the intended audience for your finished product, whether it’s a technical team, non-technical business stakeholders, executives, etc. Check that you are able to access the dataset, if one is provided. If you were given, say, a week to do the project, reach out to the company within the first day or so with any questions or concerns about the prompt. This gives them time to respond while also leaving you time to do your work.
Once you’re sure that you understand what you’re supposed to do (and that you have the data and tools to do it), move on to making a plan.
Make a plan
Now is the time to put on your project manager hat and carefully define a scope for this project. Let’s assume you were given a dataset and asked to come up with some business insights from it. That’s very open-ended! Having taken a peek at the data already, you can now lay out a plan of attack.
Your plan can take a couple of forms. Here are some examples:
- You could come up with a few business questions you want to answer with the data. These questions should have some relationship to one another so that together they tell a story. You could start with a big-picture question and drill down, for instance.
- You could organize your project according to analytical techniques you want to use. For instance, if the dataset lends itself to some kind of predictive modeling, you might plan to do some exploration, fit a few models, and evaluate the best one.
If, on the other hand, the project prompt asked you to do something very specific, plan for that plus some bonus analyses or visualizations if you have time.
Your main goal at this stage should be to give yourself steps to follow and a time limit for each so that you are sure to complete a nice-looking project in the given time. Time spent planning is time well spent!
Do your best work
There’s only so much a person can do in the hours/days/week allotted for a take-home project. Your interviewers know that. But that doesn’t mean you shouldn’t try to give them the most polished product you can.
Even if they won’t see your project code (for instance, if they just want a slide deck with your findings), make that code clean and commented. Put it on your GitHub with some clear but concise documentation so you can refer folks there to see your work in detail.
If your project includes visualizations, make them easy to read for the appropriate audience. Label those axes. Indicate units if necessary. Make sure the colors you choose are colorblind-friendly. If your audience is non-technical, it is especially important to make sure that each visualization conveys a small amount of information with maximum clarity. You don’t want your audience distracted by weird design choices or an unusual format.
Think through what questions people might ask about the project, and as you work, prepare answers. What are you assuming about the data? How does your model work? Make sure you’re not doing anything you wouldn’t be comfortable explaining to technical or non-technical folks.
Of course, the best-laid plans of data scientists sometimes go awry. Maybe you made some assumptions about your data that didn’t hold up, or you later think of some business questions that are much more interesting that the ones you originally chose. As you work through your plan, it’s all right to make adjustments as needed, but always with an eye on the clock and on the quality of the finished product.
Make a pretty package
You’re nearing the end! Now is the time to make sure that whatever you’re submitting makes a strong positive impression of you and the kind of work you can do.
If the deliverable is something like a Jupyter Notebook, make sure the code is clean and commented, with plenty of narrative interspersed so that your work is self-explanatory. Run the notebook top-to-bottom to make sure everything outputs as expected. Consider having alternative formats handy (e.g., by exporting an HTML or PDF version of your notebook) in case there are any problems using your file.
If you’re supposed to hand in a slide deck, take some time to make sure it is impressive all on its own. I highly recommend using a template in PowerPoint or Google Slides, since these give a professional look (and help ensure that your slides are accessible to screen-readers). Personally, I think Google Slides has some more modern-looking template options, but you do you.
If people will be reviewing your deck on its own, make sure it has a clear narrative structure and enough text to be self-explanatory. If you’re going to be presenting it live, go for less text so your audience doesn’t read instead of listening to you. You might go the extra mile and match any colors in your visualizations to the theme colors of your slide template. You might even go the extra extra mile and match your colors to the company’s visual branding. (I’ve done this, and while I’m not sure anyone even noticed, it made me feel like a total pro. The confidence boost alone might be worth it to you!)
Don’t forget to include your name and contact info in whatever you hand in, just in case it gets passed around inside the company. Make sure you spelled your name right—and the company’s! If you have time, it’s always beneficial to get someone to take a look at your finished product and give feedback. Then submit your project and take a break!
I hope this is helpful to anyone facing a take-home data science project for the first time. Be strategic, do your best work, and you’ll be fine!