Meta-Learning-based Deep Reinforcement Learning for Multiobjective Optimization Problems

Dependencies

• Python>=3.6
• NumPy
• PyTorch>=1.4

Meta-Learning

For training meta-model on MOTSP-20 instances:

python run.py --graph_size 20 --CUDA_VISIBLE_ID "0" --is_train --meta_iterations 10000

For training meta-model on MOTSP-50 instances:

python run.py --graph_size 50 --CUDA_VISIBLE_ID "0" --is_train --meta_iterations 5000

You can initialize or resume a run using a pretrained meta-model by using the --load_path option, e.g.:

python run.py --graph_size 50 --is_load --load_path "meta-model-MOTSP50.pt" --CUDA_VISIBLE_ID "0" --is_train --meta_iterations 10000 --start_meta_iteration 5000

Fine-tuning

For fine-tuning the trained meta-model on MOTSP-50 instances with 10-step per subproblem:

python run.py --graph_size 50 --is_load --load_path "meta-model-MOTSP50.pt" --CUDA_VISIBLE_ID "0" --is_test --update_step_test 10

For fine-tuning the trained meta-model on MOTSP-30 instances with 100-step per subproblem:

python run.py --graph_size 30 --is_load --load_path "meta-model-MOTSP50.pt" --CUDA_VISIBLE_ID "0" --is_test --update_step_test 100

For fine-tuning the random-model on MOTSP-50 instances with 10-step per subproblem:

python run.py --graph_size 50 --CUDA_VISIBLE_ID "0" --is_test --update_step_test 10

Transfer-Learning

For training all the submodels with transfer-learning by loading the well trained 1st-submodel on MOTSP-50 instances with 10-step per subproblem:

python run.py --graph_size 50 --is_load --load_path "model-0.pt" --CUDA_VISIBLE_ID "0" --is_transfer --is_test --update_step_test 10

Acknowledgements

Thanks to wouterkool/attention-learn-to-route for getting me started with the code for the Attention Model.