Save model and train on multiple instances about rl-job-shop-scheduling HOT 2 CLOSED

Hello,

Happy to hear you find our work useful :)

1. In the current version of the code, we don't save the hyperparameter as we haven't explored pre-training approaches. Mainly because the instances we tested our approach on were generated randomly, so pre-training was not yelling good results.
   It shouldn't be hard to implement however using RLLib API (export_model function if I remember correctly): https://docs.ray.io/en/master/rllib-package-ref.html?highlight=export#module-ray.rllib.policy

2. Yes the environment takes only one instance as an argument. The easiest way to implement multi-instances IMO would be to use an environment wrapper that loads multiple environments with the different instances and let him select the desired environment depending on some conditions.

Thank you,
Have a nice day too :)

from rl-job-shop-scheduling.

Thank you!

from rl-job-shop-scheduling.