RL_VRP

This implementation is similar to this implementation https://github.com/OptMLGroup/VRP-RL

We added the local neighborhood algorithm 2opt, to improve the final solution.

Dependencies

• Numpy
• tensorflow>=1.2
• tqdm

How to Run

Train

By default, the code is running in the training mode on a single gpu. For running the code, one can use the following command:

python with2opt.py --task=vrp10

Inference

For running the trained model for inference, it is possible to turn off the training mode. For this, you need to specify the directory of the trained model, otherwise random model will be used for decoding:

python with2opt.py dataset_name --task=vrp10 --is_train=False --model_dir=./path_to_your_saved_checkpoint

where dataset_name is the input dataset

Logs

All logs are stored in result.txt file stored in ./logs/task_date_time directory.

Cite: Nazari, M., Oroojlooy, A., Snyder, L. V., and Tak ́ac, M.Deep reinforcement learning for solving the vehiclerouting problem.CoRR, abs/1802.04240, 2018. URL: http://arxiv.org/abs/1802.04240, https://github.com/OptMLGroup/VRP-RL