Solving TSP using Reinforcement Learning and Optimal Transport

This repository contains a PyTorch implementation of the paper: Combining Reinforcement Learning and Optimal Transport for the Traveling Salesman Problem.

Requirements

python -m pip install -r requirements.txt

Training a new model

First change the parameters on args.yml and then run the following command:

python train.py

Once the model is trained, run:

python -m jupyter nbconvert --to python demo.ipynb
python demo.py

References

• Goh, Yong Liang, et al. Combining Reinforcement Learning and Optimal Transport for the Traveling Salesman Problem. arXiv, 2 mars 2022. arXiv.org, https://doi.org/10.48550/arXiv.2203.00903.
• Kool, Wouter, et al. Attention, Learn to Solve Routing Problems! arXiv, 7 février 2019. arXiv.org, https://doi.org/10.48550/arXiv.1803.08475.
• Bresson, Xavier, et Thomas Laurent. The Transformer Network for the Traveling Salesman Problem. arXiv, 4 mars 2021. arXiv.org, https://doi.org/10.48550/arXiv.2103.03012.