The source code of the "Graph neural networks and deep reinforcement learning in job-shop scheduling"

Edit: Organised the configuration files for better reproducibility of results. Added only configurations from the final steps of the experiments!

To perform any of the experiments you must define the configuration file following examples in configuration/experiments/**/emperiment.yml files. Then to execute the experiment just call cli.py, i.e.

python cli.py --configuration PATH_TO_EXPERIMENT_FILE

The results of the tournament runs are stored in the pickled version of the class from environment/statistics.py. To load the file use the load method of Statistics class.

from environment.statistics import Statistics

statistics = Statistics.load(PATH_TO_PICKLED_FILE)

The requirements of the work are present in requirements.txt.

Trained models and experiment results are available here. Additionally, in evaluation folder you can find the results of the tournaments of the trained models.

If you want to launch any of the experiments from the evaluation archive, then you must follow the next procedure

1. Copy the configuration folder to the 'diploma_thesis/configuration/experiments/jsp' folder
2. Update reference paths in either 'experiment.yml' or 'experiment/0.yml' files from the copied folder. For instance, in the following yml slice, the path ''configuration/experiments/jsp/BEST/experiments/1/mr_machine.yml' must be updated to 'configuration/experiments/jsp/YOUR_FOLDER/experiments/1/mr_machine.yml'

dqn_2: &dqn_2
  base_path: 'configuration/experiments/jsp/BEST/experiments/1/mr_machine.yml'
  template: '../../../../../../mods/machine/model/marl_dqn/baseline'
  mod_dirs:
    - 'configuration/mods/machine/mods'
  mods:
    - *default_mods