How to Tackle Take-Home Data Scientist Interviews? ✍
Happy new year!
And, I’m hiring (MLE role and Data Scientist role)!
Even if I fill these roles, I’d love to hear from you, make connections and offer help as I can: Linkedin and Twitter.
As a data leader, I have two ultimate goals:
1. Build efficient data organizations.
2. Hire and grow the best talents.
As a hiring manager, I’ve reviewed thousands of resumes and interviewed hundreds of candidates. When it comes to finding good data science talents through technical assessment, I’ve used most of the tools including:
- Knowledge testing style live interview (Such as my interview challenges)
- LeetCode style live interview
- Live coding (SQL or modeling)
- Short time-span take home test (Such as a SQL test in 1–3 hours)
- Long time-span take home test (Such as building a more complex NLP model in 24 hours)
The difficulty level is very much commensurate to the time given to the candidate. With more time, you’ll be given a tougher problem to solve.
My most preferred approach is the longer time-span take home test and will be the focus of this post, and my least favorite is LeetCode — it can be a great tool to filter down the candidate pool, but by itself not reliable to make any decisions without other tests.
What am I looking for in your Take-Home Deliverables?
In short, here are the requirements from my most recent hiring effort:
The setup of the test is quite simple. The test provides synthetic data as well as a detailed explanation of the context of a real-world business problem. The task is to build a machine learning model to solve this business problem within 24 hours. The format of delivery is a final demo and discussion with a powerpoint deck (45 minutes), and submission of your python code.
The above is a copy of the expectations on the output of this test that the interviewer is looking for.
In fact, most of the take-home tests will not tell you any of the above. More commonly you’ll see a list of questions that you would be graded on. The questions break down the problem into smaller tasks and give you hints of what needs to be done. There can be very strict rubrics for each question and you should answer each question thoroughly.
Here, I choose to leave it open to the candidate to decide the scope of his/her solution; and instead of giving questions as hints, I publish the expectations explicitly to the candidates.
What am I really looking for in YOU?
To answer this question, let me take one step back and answer the following first:
What am I NOT looking for?
I’m not looking for a “Fast-Food Data Scientist [1]”.
A “Fast-Food Data Scientist” is someone who learned some level of the baselines (tools, packages, models, concepts) but doesn’t have enough experience or understanding of what’s going on. I’ve briefly touched upon this in a previous post contrasting statistical and machine learning.
Symptoms can be one or more of the following:
- Lack of curiosity to really understand the business context
- Does not fully understand the business problem and naively fits models without justification
- Does not perform exploratory data analyses (EDA) when building models
- Cannot provide rationale for the engineered features or model structure
- Do not understand or verify model assumptions as needed
- Cannot interpret the results or explain the model
- Works as a lone wolf and deliver once at the end of the project
Why do I like open-ended questions?
Because they mimic real-world situations. Your manager or stakeholder will come to you with a vaguely defined problem. They require a more well-rounded skill set.
That is not equivalent to saying that I’m looking for jack-of-all-trades type of candidates. On the contrary, I’m looking for someone not just with technical abilities but also with a thoughtful mind.
That is also not saying that you need a PhD to be qualified and proven. I’ve known people who switched careers, self-taught, or learned on the job (etc.), and became renowned machine learning practitioners. A role model who you must know is Jay Alammar .
Again, a thoughtful and curious mind is needed to not be a “Fast-Food Data Scientist”. And it doesn’t take five-year graduate training to achieve that.
What are the hiring managers thinking?
I’ve talked with many hiring managers to get a sense of what they are looking for in every step of your applications. I’ve also observed my peer hiring managers’ actions as well as my own in the past. There are a lot of myths (or all kinds of biases) about what your hiring managers may be thinking, from when they first see your resume to the on-site interview.
- Do they prefer a certain style of resumes?
- Do they prefer PhD degrees?
- Do they prefer big company experiences?
- Do they really look at the resumes in the applicants pool?
- Or as what this question was asking from LinkedinGroup about pedigrees
A lot more to cover in a future post! Stay tuned for the next myth buster!
Happy practicing!
Thanks for reading my newsletter. You can follow me on Linkedin or Twitter @Angelina_Magr !
Reference:
[1]”Fast-Food Data Scientist” — a concept coined by the author.
