search

wuzhy66 / ML-DAM

Meta-Learning-based Deep Reinforcement Learning for Multiobjective Optimization Problems
25 stars 16 forks source link

a question on the maml implementation #1

Open YJPQQ opened 2 years ago

YJPQQ commented 2 years ago

According to the MAML paper,in line 268 of the file 'run.py', meta_loss should not be divided by task_num, right?

wuzhy66 commented 2 years ago

I referred to MAML-Pytorch and pytorch-maml, where meta_loss is divided by task_num. I think it can normalize the loss.