Open wolf2127 opened 3 months ago
Yes, you should first train the teacher model (the black-box), then generate the dataset within the code, and finally train the SAFE-RL model. For your convenience, I have just released the pretrained teacher and SAFE-RL models for some DRL agents across various environments. You only need to generate the dataset now. Please check the Releases repository for instructions.
I would like to know how to be able to train or test your code? When I try to run SAFE_RL.ipynb, it prompts me with FileNotFoundError: [Errno 2] No such file or directory: 'teacher_models/A2C_pong_teacher_model.zip' I think this is caused by the lack of training.