search

drawbridge / keras-mmoe

A TensorFlow Keras implementation of "Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts" (KDD 2018)
MIT License
681 stars 217 forks source link

How to solve the both regression task and classification task by MMOE model #17

Closed zoe218 closed 2 years ago

zoe218 commented 3 years ago

Hi~

I want to combine regression task and classification task in one model, but I found the regression loss and classification loss are so different, so it's too hard to optimize?

Can you give me some advice to handle this problem? Thx~~

alvin319 commented 3 years ago

Hello - one piece of advice that I would give is that multi-task learning is most useful when there is some correlation between your classification and regression tasks. You might want to try to inspect its Spearman/Pearson correlation on your labels to see if that's the case.