BOPO

Official Implementation of paper: BOPO: Neural Combinatorial Optimization via Best-anchored and Objective-guided Preference Optimization

JSP

Implementation of BOPO for JSP.

First, unzip the training set and validation set.

unzip JSP/dataset5k/dataset5k.zip
unzip JSP/benchmarks/validation/validation.zip

Then, you can start the project quickly:

# Quick Testing
python JSP/test.py -B 32 -benchmark LA 

# Training
python JSP/train.py -B 256 -K 16 -tag test -epochs 20 

TSP

Implementation of BOPO for TSP.

You can start the project directly:

# Quick Testing
python TSP/test_n20.py
python TSP/test_tsplib.py

# Training
python TSP/train_n20.py 

FJSP

Implementation of BOPO for FJSP.

First, unzip the training set and validation set.

unzip FJSP/dataset/dataset.zip
unzip FJSP/benchmarks/validation/validation.zip

Then, you can start the project quickly:

# Quick Testing
python FJSP/test.py -B 32 -benchmark LA-e 

# Training
python FJSP/train.py -B 256 -K 16 -tag test -epochs 20 

Requirements

Please see the requirements.txt

Cite

@inproceedings{
liao2025bopo,
title={{BOPO}: Neural Combinatorial Optimization via Best-anchored and Objective-guided Preference Optimization},
author={Zijun Liao and Jinbiao Chen and Debing Wang and Zizhen Zhang and Jiahai Wang},
booktitle={Forty-second International Conference on Machine Learning},
year={2025},
}