How curriculum learning is implemented?
We thank the authors for publishing an open-source implementation of their paper. Looking at the repo, it is not immediately clear how curricula are implemented. Can the authors please clarify this point for work on a successor paper? Thank you very much.
It is done manually. We run the commands shown here in the README in sequence, waiting for each step to finish before starting the next one. Each corresponding configuration file contains the key model_path, which either is False (for random initialization) or points to a file that stores policy network parameters obtained from the previous curriculum step. We use those parameters to initialize the network. Please let me know if you have further questions.