Preparing for the transition - data science as a student vs in the industry
0 likes43 views
The document discusses the differences between data science as a student and in the industry, highlighting how problems are approached differently in academia and industry settings. It contrasts academic environments with well-defined, clean problems and industry where issues are often product requirements with dirty data and real user impacts. The author shares personal experiences, detailing ineffective strategies and helpful practices to thrive in data science careers.
Preparing for the transition - data science as a student vs in the industry
- 1. Ning Zhou 2018.03.08 Preparing for the Transition: Data Science as a Student v.s. in the Industry
- 2. 2 How I got started with data science Bachelor Master PhD Industry Research Industry Product Development Classroom Project Academic Project Industry Project
- 3. Classroom vs Academic vs Industry 3 Problem Data User Peer
- 4. Problem Academic ● Technical problems abstracted from business scenarios ● Prefer challenging problems that are difficult to solve 4 Industry ● Problem often arrives as a product requirement instead of a technical problem ● Prefer low-hanging fruits that can bring large impact with relatively small efforts Classroom ● Well-defined problems with clear metrics to measure success ● “Solved” problems with known solutions
- 5. Data Academic ● Open dataset with some quality assurance ● Mid to large volume ● Work with industry datasets too, but often pre-collected 5 Industry ● “Dirty” data ● Can be from very small to very large volume ● Data collection takes time Classroom ● Clean data ● Relatively “small” volume
- 6. User Academic ● Limited opportunities to test with real users ● Offline test is still the most common way to measure performance 6 Industry ● Impact on real users (no matter good or bad…) ● Online test is considered as “final” Classroom ● No real user impact ● Mostly offline test only
- 7. Peer Academic ● Smart peers from the broader research community working on similar topics 7 Industry ● Smart colleagues, but they normally work on different projects Classroom ● Smart classmates who work on the same project
- 8. Five Things I tried that didn’t help ● Team up with smart and hardworking classmates and then just be lazy ● Try random open-source models without thinking through ● “Tune metrics” instead of tuning models ● Manipulate data manually to throw out bad or difficult samples ● Procrastinate until the deadline approaches 8
- 9. Five Things I tried that helped ● Be curious about what other people are working on ● Keep cost and performance in mind ● Stay updated with the latest progress from both academia and industry ● Try things hands-on ● Write papers / technical sketches / blogs 9
- 10. Tusen takk! Questions? You can also contact me at ning@duoja.com. 10