AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
How to Balance Machine Learning Metrics and Business Metrics in a System
Ludwig: The system was doing multiple tasks. One of the tasks was identifying what type of issue it was related to from a list of, I don't remember, but probably 6,000 different classes. And then there was the defining what template, or among the different template for responding to the user, the customer support represented should use. We decided that figuring out if the correct prediction was within the first three of the ones suggested was actually the most valuable thing. So we wanted to make sure that both things were improving or at the very least that if one was improving, then the one was not decreasing.