An improvement-based Deep Reinforcement Learning Algorithm presented in paper https://arxiv.org/abs/1912.05784v2 for solving the TSP problem.
This repo implements paper Wu et. al., Learning Improvement Heuristics for Solving Routing Problems, IEEE Transactions on Neural Networks and Learning Systems, 2021, which solves TSP with the improvement-base Deep Reinforcement Learning method.
For more details, please see the paper Wu et. al., Learning Improvement Heuristics for Solving Routing Problems, IEEE Transactions on Neural Networks and Learning Systems, 2021.
@article{wu2021learning,
title={Learning improvement heuristics for solving routing problem},
author={Wu, Yaoxin and Song, Wen and Cao, Zhiguang and Zhang, Jie and Lim, Andrew},
journal={IEEE Transactions on Neural Networks and Learning Systems},
year={2021}
}
You may also be interested in our new approaches:
We provide a Jupyter notebook to help you get started and understand our code. Please open the notebook here for more details.
Note: due to the 100MB limit of Github, please download the logs folder for the pre-trained model via Google Drive and put it under './logs/pre_trained/' folder.
For the exception below from package tensorboard_logger,
AttributeError: module 'scipy.misc' has no attribute 'toimage'
Please refer to issue #27 to fix it.
CUDA_VISIBLE_DEVICES=0 python run.py --graph_size 20 --seed 1234 --n_epochs 100 --batch_size 512 --epoch_size 5120 --val_size 1000 --eval_batch_size 1000 --val_dataset './datasets/tsp_20_10000.pkl' --no_assert --run_name training
--eval_only --load_path '{add model to load here}'
Note: A pre-trained model can be found at './outputs/tsp_20/tsp_20200714T212735/epoch-99.pt'
The code is based on the repo wouterkool/attention-learn-to-route and the paper Wu et. al., Learning Improvement Heuristics for Solving Routing Problems, IEEE Transactions on Neural Networks and Learning Systems, 2021.