Using the WeakLearn algorithm from the original RankBoost paper, how do you set the optimal threshold to maximize AU-RPC (instead of AUC)? And, once that threshold is set, how do you calculate the AU-RPC for use in calculating alpha?

This paper suggests that the "area under the recall-precision curve [is] a much better metric [than AUC] when working with highly skewed data sets", which is my current application. The paper proposes a modification of RankBoost.B that purports to maximize AU-RPC. However, the authors are using learners based on first-order clauses, while I need to use the originally proposed WeakLearn learners. How can I apply the same idea with the original learner algorithm?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.