A Typical Day for a Data Scientist with Dr. Andrew Graczyk

Can you describe a data scientist’s typical day at work?

In my experience, it’s hard to say what even counts as a “typical” day. I have had the opportunity to work on a wide variety of projects at NNData, but even within projects the day-to-day can vary greatly. Some days are spent coding, some spent testing/debugging, some spent researching new/different approaches to problems, some on coordinating project deliverables with other teams. Generally, days will be mixtures of those things (and more). A data science team (and even each individual data scientist) should always have a plan of action for a day or a week, but be prepared for that plan to be disrupted by unexpected problems, changes in client desires, or any of a thousand other things that can change priorities. It is important for a data scientist to think and plan dynamically.

What skills do you find yourself using the most in your day-to-day job?

There are two skills that I use daily: mathematics and communication. Programming skills are good, data engineering skills and knowledge help me to coordinate projects more effectively, but the basic skills of mathematics and communication are even more important on a near-constant basis. A data scientist needs to have a strong understanding of mathematics and statistics to understand what models and methods are appropriate for a given problem, quickly master new models and methods, determine what machine learning techniques are best suited to any task, and know whether machine learning is appropriate for a task at all. Perhaps more importantly, mathematical reasoning can help a data scientist to interpret results from any analytics they apply and understand how results could have come from a data set. Communication skills are vital to sharing and expanding ideas with team members, managers, and clients. Data analytics and models are only useful if you can communicate why they are useful and what actionable insight can be gained from their results.

Can you describe your workplace? Do you work in an office, from home, or in a hybrid environment?

My work is entirely remote, and I think that is an excellent model for data science. It is much easier to be productive on the job and live a fulfilling personal life without the need for a commute. Also, not having to move to a new city with a new job is fantastic. For data scientists, there is really no aspect of work that can’t be done remotely just as effectively as in person, if not more so.

A data scientist needs to have a strong understanding of mathematics and statistics to understand what models and methods are appropriate for a given problem, quickly master new models and methods, determine what machine learning techniques are best suited to any task, and know whether machine learning is appropriate for a task at all.

Is there a lot of collaboration in your role, or is it mostly independent work?

My role has a mixture of collaboration and independent work. Over the last year and a half, I have been involved with half a dozen projects, some of which required more independent problem solving, and some were more explicitly team-oriented. The team with which I work at NNData is very supportive; no matter the project or situation, the team can pool our resources to help come up with solutions to problems that might be creating roadblocks for any one of us, whether it be an elusive bug in a code, or difficulty implementing a particular technology in a system. It is always good to be able to talk through ideas with other data scientists, since no one has exactly the same perspective. The important thing to remember is that you are a team. No one expects one person to solve every problem on their own, so communicating problems as well as successes to your colleagues is key to success. That can be a tough thing for people who are more accustomed to academia to do!

What’s your favorite part of your job? The most challenging part of your job?

My favorite part of the job is creating a plan of attack for a new problem: envisioning the steps necessary to take raw data and put it into a form that gives meaningful, actionable knowledge to the client. That’s when I usually discover the most interesting new technologies, techniques, and models, and get to come up with the most creative ways to use those methods to process data given available resources. This is also the second most challenging part of the job, since there are often so many potential ways to solve a problem that determining which one is best can be very difficult, and sometimes even subjective. The most challenging part of the job, of course, is finding the source of a bug that is causing all of your model outputs to be nonsensical. You search your entire codebase for several hours, only to discover that you misplaced a comma somewhere two weeks ago that somehow never caused a problem until now!