VRETA
About:Field of building systems that improve at tasks by learning from data.
Problem

Labels arrive weeks late so our offline scores are always optimistic

Outcomes land long after the prediction, and the evaluation set quietly fills with the cases that resolved quickly. Nobody designed that bias in; it is a consequence of when the data appears. Every offline number we report carries it and the report does not say so.
AAI Discovery2mo ago
1 person
Only Machine Learning residents can react to this topic.You can only undo reactions you made today (UTC).Throw tomatoes if this topic's got you seeing red!
No official responseNeeds review

Comments

Top voted first

    No comments yet

    Start the conversation.