liangxinedu / MDAM

Multi-Decoder Attention Model with Embedding Glimpse for Solving Vehicle Routing Problems

Liang Xin, Wen Song, Zhiguang Cao, Jie Zhang. Multi-Decoder Attention Model with Embedding Glimpse for Solving Vehicle Routing Problems, Proceedings of the AAAI Conference on Artificial Intelligence, pp.12042-12049, 2021. [pdf]

To train the model (e.g. TSP with 20 nodes)

python run.py --graph_size 20 --baseline rollout --run_name 'tsp20' --problem="tsp" --batch_size=512 --epoch_size=1280000 --kl_loss=0.01 --n_EG=2 --n_paths=5 --val_size=10000

To generate the test dataset for TSP20

mkdir data
python generate_data.py --filename=data/tsp_20 --problem="tsp" --graph_sizes=20 --dataset_size=100000

To evaluate the test dataset with pretrained model for TSP20

CUDA_VISIBLE_DEVICES="0" python -u search.py data/tsp_20.pkl --model=pretrained/tsp_20/epoch-99.pt --beam_size=1 --eval_batch_size=1024

To Download all the pretrained models, generate the test datasets, evaluate with greedy strategy, 30-width and 50-width beam searching

bash eval.sh download
bash eval.sh generate_data
bash eval.sh tsp
bash eval.sh cvrp
bash eval.sh sdvrp
bash eval.sh op
bash eval.sh pctsp
bash eval.sh spctsp

Dependencies

• Python==3.6
• NumPy
• PyTorch>=1.4.0
• tqdm
• tensorboard_logger

MDAM is developed based on https://github.com/wouterkool/attention-learn-to-route (Kool, W.; van Hoof, H.; and Welling, M. 2019. Attention, Learn to Solve Routing Problems! In Proceedings of International Conference on Learning Representations (ICLR).)