search

danijar / dreamerv3

Mastering Diverse Domains through World Models
https://danijar.com/dreamerv3
MIT License
1.28k stars 219 forks source link

Offline Training #80

Closed mrsamsami closed 1 year ago

mrsamsami commented 1 year ago

Hey, I want to express my gratitude for this fantastic code repo!

I'm currently seeking advice on the most straightforward approach to training an online agent, gathering a dataset, and subsequently training an offline agent with the dataset. I've been unable to locate the "offline mode" for training. Is it possibly embedded somewhere in the codebase?

danijar commented 1 year ago

Hi, there is no purely offline training implemented, although you could just set the train ratio to a very large number. It's also quite straightforward to adapt one of the run scripts, for example run/train_eval.py or similar. Hope that helps a little bit.