search

YyzHarry / imbalanced-regression

[ICML 2021, Long Talk] Delving into Deep Imbalanced Regression
http://dir.csail.mit.edu
MIT License
826 stars 130 forks source link

关于FDS、LDS相关参数在自己数据集上的调整 #8

Closed Breeze-Zero closed 3 years ago

Breeze-Zero commented 3 years ago

老师您好，我想将FDS、LDS应用到自己的数据上，请问FDS、LDS示例里的kernel、 ks、sigma等参数需要根据什么情况去设定呢？

YyzHarry commented 3 years ago

Hi - thanks for your interets. The choice of kernel, ks, sigma should be related to the task of interest. For example, in age estimation, with the minimum bin size of 1 (the minimum age resolution you care is 1 year), you would not expect a very large kernel size considering similar nearby ages. In Appendix E.3 of our paper, we studied some choices of ks and sigma, and it might give you some sense on what values are good for different tasks.

Breeze-Zero commented 3 years ago

Well, I applied LDS and FDS to my age estimation model and found that MAE was hovering at 10 after the introduction of FDS and could not be reduced. This problem did not occur when LDS was used only without FDS. Do you have any idea what might have caused this?

YyzHarry commented 3 years ago

I'm not sure --- we did not see such phenomenon in the provided datasets (e.g., IMDB-WIKI-DIR, etc.). Maybe you need to tune the hyper-parameters of FDS a bit (still, kernel, ks, sigma), and see if the problem still exists.