nicklashansen/dmcontrol-generalization-benchmark

cnn-encoder part of Actor in sac.py would never be updated?

Closed this issue · 1 comments

Hi, here

_, pi, log_pi, log_std = self.actor(obs, detach=True)

As the detach=True, all the cnn-encoder parts of the actor would not be updated? is it right? or you do not want to update the cnn encoder of the actor?

or am I missing something?

Hi, the cnn layers of the encoder (shared_cnn in the code) are shared between actor and critic, and only the critic updates those layers. We largely follow the implementation of pytorch_sac.