rll-research / url_benchmark

MIT License
331 stars 51 forks source link

Why no target network for DDPG actor? #14

Open fdeng18 opened 2 years ago

fdeng18 commented 2 years ago

Hi, thank you all for this great work!

I'm a bit curious about why you decided not to use a target network for the actor when computing the target Q values, as this differs from the original DDPG (and also TD3 and D4PG). Did you have some ablation study?

seolhokim commented 2 years ago

This repository Inherits drq-v2. That algorithm doesn't use target network. You'd better contact the author of drq-v2 (or drq-v1).