search

median-research-group / LibMTL

A PyTorch Library for Multi-Task Learning
MIT License
1.98k stars 185 forks source link

Inconsistency between formula and implementation in count_improvement function #65

Closed Tyqfat closed 10 months ago

Tyqfat commented 10 months ago

Dear Maintainers, I have encountered a potential issue while reading and using the project. In the LibMTL.utils, I believe there is an inconsistency between the mathematical formula for the countimprovement function and its implementation. The formula is as follows: image However, I noticed that certain calculations in the code implementation do not entirely align with this formula. Specifically, here is where I think the inconsistency occurs: the denominator in the formula is N{t,m}, while the denominator in the code when calculating the formula is B_{t,m} image

Baijiong-Lin commented 10 months ago

There is a typo in the formula. The denominator $N{t, m}$ should be $B{t, m}$. Thanks for pointing out it.

Tyqfat commented 10 months ago

You're welcome. I'm glad to contribute to the project.