We provide code, data and trained models.

conda create -n python=3.7
pip install -r requirements.txt

You can download our trained models and datasets from here, then unzip the files to the root directory of this project.

train a model on TSP100.

python train.py graph_size=100 group_size=100

eval on tsp_random

python eval.py val_data_path=./data/tsp100_test_concorde.txt load_checkpoint_path=./result_ckpt/model100.ckpt

eval on tsp_partner

python eval.py val_data_path=./data/partner_100.txt load_checkpoint_path=./result_ckpt/model100.ckpt real_data=true