How Data Scientists Deal with the Unknown
2022 is just around the corner, but the uncertainty most of us have had to cope with these past couple of years is still with us (it looks like it might stick around for a while longer). While we can’t do much to help with many of the unknowns you may be facing, a few of them are very much up our alley.
This week, we turn our attention to (model) drifts and (career) transitions—two topics that data scientists are never too tired to discuss, and which feel especially timely during this period of change.
- What are the best metrics for analyzing model drift? As models run in production, data changes, input/output dynamics evolve, and performance can often deteriorate. Piotr (Peter) Mardziel dives deep into the problem in his recent post, and shows how solutions vary greatly depending on context and a model’s features—which is why it’s so important to have a wide range of metrics through which to look at your model.
- You found out your data is drifting. Now What? Once you’ve identified the symptoms of data drift in your model, you need a plan. Fortunately, Elena Samuylova’s roadmap is here to save the day: it’s detailed, pragmatic, and accessible, and approaches the issue from enough distance to be applicable to many real-world scenarios.
A drifting model is one thing. You drifting? A much bigger problem, clearly. If you’re thinking about a career change, feeling stuck, or just looking for inspiration from people who’ve recently made moves in their career, here are three recent posts you shouldn’t miss.
- A brand-new data scientist shares key takeaways from the journey thus far. If you’re mulling a switch to data science, Amanda West has recently made it to the six-month milestone and generously shares some insights (technical and nontechnical) about life as a freshly minted data scientist, from the importance of hardware to the need to tackle bad habits.
- Learn about the challenges and rewards of discipline-hopping. For Danny Kim, who has an academic background in communications, the leap to data science felt daunting; it made him question whether there was a place for someone like him in the field. His helpful post focuses on the hurdles people with social-sciences training face—and on the benefits of entering data science with a different skill set and knowledge base.
- Is the jump from in-house data scientist to analytics consultant worth it? After a year as an Associate Consultant at Slalom, Shravankumar Hiregoudar has a good idea of the advantages (and, sure, the drawbacks) of the consultant role, which he lays out in great, patient detail. If that’s a transition you’ve been considering, you’ll want to read his overview.
Looking to add a few more reads to your list? We hope so, because our almost-end-of-the-year crop of recent articles was fantastic, and covered a lot of ground. Here are some highlights:
- Valerie Carey explored the always-important topic of integrity and how data scientists should deal with their own mistakes.
- How strongly should you hold on to your beliefs? It’s a big, crucial question that transcends our daily grind, and Vishesh Khemani, Ph.D. does a great job approaching it with curiosity and level-headedness.
- Raise your hand if you love great-looking plots! Or better yet: click away to read Aruna Pisharody’s handy guide to creating crisp, publication-ready plots with LaTeX.
- Hungry for some cutting-edge research? Check out Leonardo Tanzi’s engaging writeup based on his primary PhD research, which focuses on using Vision Transformers to classify femur fractures.
Wherever you are, we hope you’re safe, relaxed, and ready to wave 2021 goodbye in a couple of weeks. We’re grateful for the time you spend with us and for your support of our authors’ work.
Until the next Variable,
TDS Editors
