Practice doesn’t make perfect. Long hours of hard work can easily go to waste if not structured wisely. Athletes have long known this, which is why the best coaches are so intensely sought after. The psychological field of Expert Performance has mapped out some requirements for allowing time spent in practice to actually improve performance. Taken together, the application of these principles is known as ‘deliberate practice.’
Most of the studies were conducted on athletes, musicians, and a small number of other professions. How does it generalize to those who work with data? Here are some of the core principles, as applied to the daily life in a real business:
- Build in tight feedback loops. When you screw up (which is common—otherwise you’re not challenging yourself), you should see it immediately. Preferably, it should be fairly clear where you screwed up, as well. We’ve all tried to debug a frustrating SQL query, analysis pipeline or spreadsheet, where the numbers are clearly absurd, but it’s hard to trace back exactly to where the system is breaking down. You are taken completely out of your ability to think of the bigger picture and dragged into the weeds of one particular bug. Once you’ve caught the bug, it’s important to step back and make sure that you understand the true root of the issue before moving on. Keep practicing the fundamentals until you’re doing them right the first time, and insist on an environment that helps you do so. As Bill Dimm of Hot Neuron suggests, “When learning about a new area, try to invent your own algorithm, experiment with it to see how tweaking helps/hurts, and only turn to research by others after you’ve really invested time thinking for yourself.”
- Get objective feedback. There are many pursuits in life where you can too easily fool yourself into believing that you’re doing a great job, because there’s no clear outcome that would tell you that you’re not. Many people are overconfident in their own abilities, particularly when they are too inexperienced to have a gut sense of what quality looks like. Other people are under-confident, taking the lack of comparative metrics to mean that they are behind the pack. Either way, if your self-perception is based solely on a combination of your personality and whether you woke up on the right side of the bed, it will be hard to make the slow but consistent progress that true mastery requires. Regularly setting aside time for explicit feedback, whether with a peer or stakeholder, will give you the necessary clarity.
- Join the community. If you feel like you don’t have any data science peers or mentors, Twitter can be a fantastic place to build community. Renee Teate had great success building connections with her DS-focused account @becomingdatasci, and has even assembled four fantastic lists of accounts (including 600+ women in Data Science) to help you get started. Being part of the community helps you understand what’s really important and what’s noise. It also gives you a place to turn when you go beyond the introductory tutorials and need some specific advice.
- Account for non-deterministic feedback. Stock brokers have very objective and tangible feedback on the performance of their portfolios. It’s easy to assume that they also have such great bearings on their skill, but for both wins and losses, they need to constantly wonder, “Was this just the result of luck?” When such a large component of your feedback is random noise, it’s tough to see the signal. This is well understood in a mathematical sense, but they are less well appreciated in a psychological sense. Human beings are just really bad at it, particularly when the signal only appears over the time span of months and years. While your workplace isn’t as variable as the stock market (hopefully!) it’s important to understand whether a project’s outcome was primarily due to factors within your control, or outside of it.
- Focus. Training is an intense mental activity, and it requires your full attention. If you’re in an environment where you’re being dragged away from your practice for emails, meetings and office politics, there’s no way that you can see the bigger patterns or bring the fundamentals into muscle memory where they belong. Get a good pair of headphones, or claim an empty conference room for a few hours. You don’t need to be deliberately practicing all day, but you do need to be 100% on when you are. And it’s important to structure your practice around what’s relevant to your specific business. As Sandro Saitta, creator of the Swiss analytics association, says, “In my opinion, you grow your data science skills by practicing. Not Kaggle competitions of course, but by solving business problems in industry.”
- Don’t forget broader skills. Kevin Hillstrom, former Vice President of Database Marketing at Nordstrom, realized that being a data scientist requires reaching across fields for best practices: “For me, technical skills were important to a point. The best training I ever received was Dale Carnegie sales training. I learned how to sell my message, and that turned out to be far more important than learning additional technical skills.”
One particularly difficult technique to practice is exploratory analysis. How can you define whether an exploration was successful? It’s difficult, but absolutely possible. You need to focus on the smallest possible chunks, the individual questions that you asked as you stepped along path towards understanding. If you were able to effectively identify, specify, analyze and communicate those solutions, you had success. Whether that step ended up being part of the broader answer is immaterial (for the purposes of Deliberate Practice, at least). What matters is that you’ve got the basic skill set of an effective data explorer.
These same principles can be applied to all of the other core skills in your role. Before you know it, you’ll be the Tiger Woods of your company!