Optimization-AI / LibAUC

LibAUC: A Deep Learning Library for X-Risk Optimization
https://libauc.org/
MIT License
273 stars 37 forks source link

The difference on AUC margin loss V1 & V2 #28

Closed wxr99 closed 1 year ago

wxr99 commented 1 year ago

Great job by the authors. We are now optimizing the AUC margin loss on a new federated optimization scheme, but we found two versions of AUC margin loss methods in the /libauc/losses, respectively named AUCMLoss_V1 & AUCMLoss_V2 in auc.py and losses_v1.py. The main difference is about the positive ratio p (V1 includes but v2 didn't), and it directly influences the optimization. Then, why or which should I concentrate on?

wxr99 commented 1 year ago

And both are used in the given examples, like federated examples and basic examples given by authors, so it really made me confused.

yzhuoning commented 1 year ago

Please use V1 (by default) for now. We are still testing V2 (to remove p in the loss). Thanks!

wxr99 commented 1 year ago

It will be a great job if no prior ratio is introduced, we hope to see a more simple version of AUC loss