I want to share with you 9 stories about ML development which illustrate how confusing and counterintuitive ML can sometimes be. Some of them fun, some... have learning value. I hope they will help you avoid some of the mistakes and achive significant results faster.Key Takeaways:
- Always start with a heuristics.
- Reliable fallback is a good foundation for innovation.
- Model = Code + Data.
- Online/offline parity.
- Why models always fail quitely.
- Do customers care about the precision and recall?
- Experimenting 100x faster, are you kidding me?
- Where models go to die.
- If you want to eat well invest in modelling approaches, if you want to sleep well invest in features.