Machine Learning/8 Guidelines: Data Science Initiative Excellence
1. Liberally use time-boxed spikes for exploring ideas and tracking open-ended work.
2. Always build naive baseline models first, and then incrementally improve and revise them, adding complexity as you go.
3. Demonstrate your progress just like you would any other software initiative, even if the results are underperforming .
4. There is a tendency with data science engagements to have a somewhat irrational focus on just the numbers. Remember that the point of these initiatives is to drive to an outcome, so telling the complete story about how the work is leading toward the outcome is as important as the numbers that show it.
5. Kickoff every major data-driven feature with robot empathy maps to help set guidelines about how non-deterministic, intelligent algorithms should work.
6. Have frequent (at least daily) check-ins, desk checks, or collaboration sessions focused on the longer tasks to promote healthy ideation and to prevent frequently check-in (daily) on longer tasks to prevent those pesky rabbit-hole scenarios that can present themselves. 7. Track experiments and knowledge in a common repository and share them as you go and certainly during retrospectives.
8. Allocate time liberally based on the level of uncertainty of the task at hand. Uncertainty is natural with data science initiatives. Embrace it, consider the first seven guidelines, and plan accordingly.