Skip to content

在仿真easyrl书中的ddpg算法遇到的问题 #175

Closed Answered by johnjim0816
yxz777 asked this question in 问题求助
在仿真easyrl书中的ddpg算法遇到的问题 #175
@yxz777 yxz777
Nov 7, 2023 · 1 comment
Answered by johnjim0816 Return to top
Discussion options

yxz777
Nov 7, 2023

You must be logged in to vote
Answered by johnjim0816 Nov 13, 2023

按照ddpg算法伪码以及代码中的定义，actorloss应该是-q，criticloss为目标网络与实际网络的差值。 在训练的过程中，actorloss在不断的上升，criticloss也上下飘忽不定 那这样也就意味着actor网络输出动作的q值是越来越小的，criticloss也无法达到一个稳定的loss 所以评判agent学习的好坏要看奖励的增减还是loss的收敛呢？

看奖励，loss根据具体算法不同会有不同特性，用动作熵评估更好

View full answer

Replies: 1 comment

Comment options

johnjim0816
Nov 13, 2023

You must be logged in to vote
0 replies
Answer selected by Sm1les
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
2 participants
@yxz777 @johnjim0816