How Do You Measure Success as a Data Scientist?
Our weekly selection of must-read Editors’ Picks and original features
We’re feeling festive this week—Towards Data Science recently hit a major milestone when we waved an excited “Hello!” at our 600,000th follower on Medium.
There are many ways to measure the success of a publication, but none of them matter if your audience doesn’t show up. So, to everyone reading the Variable: thank you for showing up, keeping our community supportive and welcoming, and helping us spread the word. If you’d like to support us and our contributors in more direct ways, consider becoming a Medium member. It will not only give you unlimited access to every single post on TDS (and to all paywalled posts on Medium), but also help us and our writers continue to publish top-notch data science articles.
In case you wondered how we celebrated: we’re a distributed editorial team, so many champagne-bottle emoji were involved. (Maybe some real ones, too! Who knows?) Few things give us more joy, though, than publishing really excellent posts, and we certainly did that, too.
Unlike a digital publication, where readers continuously vote with their eyes and their time, many software companies require nuanced tools to measure success. Chris Dowsett shared the process the Spaceship team developed to implement a fully automated NPS (Net Promoter Score) program—read it if you’d like to collect meaningful user feedback but don’t have a Google-sized budget (or data team).
In many industry contexts, defining success is an iterative process, a key element of which is avoiding the most common and harmful mistakes. For someone like Ying Li, who manages data science for IBM’s corporate headquarters HR team, having a robust framework in place is crucial; in her latest post, she details the seven-stage lifecycle she recommends for analytics management, from formulating the right problem to sunsetting obsolete solutions. Fabrizio Fantini, for his part, focuses on building the right guardrails to keep teams focused on the right goals; read his post to learn how to avoid four common pitfalls in data-science project design. Tackling a similar problem from a different angle, Aparna Dhinakaran explains ML observability, and why it matters for teams as they take their models from the drawing board into practice: it’s “the key difference between a team that flies blind after deploying a model and a team that can iterate and improve their models quickly.”
Sometimes, gaining a more nuanced understanding of the problem at hand is itself a measure of success—especially in a field as young and ever-shifting as machine learning. Julia Nikulski took a deep dive into the notoriously complex topic of toxicity in AI text generation, and if you follow her along you, too, will get a clearer idea of the factors and stakeholders that make an easy fix impossible. Likewise, Blake VanBerlo and the team at London, Ontario’s Municipal AI Applications Lab are working towards addressing chronic homelessness in their city; it’s still early days for the predictive model they’re building. What they’ve learned so far is already valuable, though, and will shape the options available to policymakers in their work to support marginalized people in their community.
Finally, our own team had occasion to reflect on process, success, and iteration this past week too—and not just because we share our posts with no fewer than 600,000 subscribers. (Did we mention that already? We may have.) In our latest Author Spotlight Q&A, featured writer Eugene Yan walked us through the path that took him from psychology major to Applied Scientist at Amazon, and offered insights on goal-setting, learning, and challenging oneself as a data scientist and public writer. On the TDS Podcast, host Jeremie Harris welcomed Brian Christian, author of The Alignment Problem, to discuss what might be the biggest challenge we—data scientists, machine learning engineers, and, well, humans in general—face today: ensuring that the AI systems we build don’t, well, destroy us.
Thank you once again for being part of our community—this week, and every week. If you’re celebrating an achievement or milestone this week too, let us know. We can never have enough champagne-bottle emoji in our lives.
Until the next Variable,
TDS Editors
Recent additions to our curated topics:
Getting Started
- Interpretable or Accurate? Why Not Both? by Parul Pandey
- Making Decisions with Data Science by Josh Taylor
- Interviewing for Your First Data Scientist Job: What to Expect and How to Prepare by Emma Ding and Hide Kato
Hands-On Tutorials
- Machine Learning for Credit Card Fraud Detection by Yann-Aël Le Borgne
- ConnectorX: The Fastest Library for Loading Your Python DataFrame by Xiaoying Wang
- Getting Started with Graph Embeddings by CJ Sullivan
Deep Dives
- Space Partitioning and KD Trees by Max Miller
- Engineering Essentials for Data Science by Matt Sosna
- Build a Machine Learning Model to a Practical Use Case from Scratch by Yasas Sandeepa
