This is the GitHub repository for 'DGL: Dynamic Global-Local Information Aggregation for Scalable VRP Generalization with Self-Improvement Learning'.

We use the same dataset as INViT, Please prepare your dataset under the data directory in the project root directory first.

train

cd CVRP
python train.py

cd ..
cd TSP
python train.py

test

cd CVRP
python test.py

cd ..
cd TSP
python test.py

We would like to thank the following repository, which is the baseline of our code:

https://github.com/Kasumigaoka-Utaha/INViT

https://github.com/CIAM-Group/NCO_code/tree/main/single_objective/LEHD

⚠️ Notice on Experimental Results (Updated: [2025/10/24])

We have identified that the ELG configuration was unfair and have now corrected its results(pomo_size=50).