issues
search
Denys88
/
rl_games
RL implementations
MIT License
848
stars
142
forks
source link
Update for tacsl release: CNN tower processing, critic weights loading and freezing.
#298
Closed
iakinola23
closed
1 month ago
iakinola23
commented
2 months ago
enable loading the weights of the critic network from a PPO checkpoint, without the actor weights
add flag to freeze critic while training actor
adding ability to post-process the output of a conv tower with the spatial soft argmax or flatten layer