search

GeminiLight / drl-sfcp

[ICC'21 - DRL-SFCP] Implementation of our paper "DRL-SFCP: Adaptive Service Function Chains Placement with Deep Reinforcement Learning", accepted by ICC 2021.
Apache License 2.0
28 stars 11 forks source link

为什么actor_loss不收敛呢 #8

Closed swqsbssddqs closed 1 year ago

swqsbssddqs commented 1 year ago

你好，作者大大，为什么actor_loss不收敛呢，全部参数都是virne里面默认的？ image

GeminiLight commented 1 year ago

强化学习中actor loss和critic loss的趋势与agent收敛的关系并不紧密，更重要的是累积回报有没有提升、entroy_loss有没有下降。或许这两个指标可以更好的帮助你判断agent学习的情况。