The Unknown Benefits of using a Soft-F1 Loss in Classification Systems
How to not maintain a decision threshold and still optimize directly for what you care about!
Published in
14 min readDec 4, 2019
Deploying machine learning systems in production and ensuring their day-to-day efficiency can be a very hard challenge. During the last five years, I have seen and increasing number of technologies that are becoming more and more complex (algorithms, coding libraries, cloud services,…). This is all good news to reshape our world with useful AI applications.