alstn12088 / Sym-NCO

Sym-NCO: Leveraging Symmetricity for Neural Combinatorial Optimization

Sym-NCO is deep reinforcement learning-based neural combinatorial optimization scheme that exploits the symmetric nature of combinatorial optimization.

Before reading our code, we strongly recommend reading the code of AM and POMO which is base of our code.

Sym-NCO-POMO

Sym-NCO is an extended method of POMO. POMO will give powerful performances in TSP. Sym-NCO can improve POMO at CVRP and slightly at TSP.

TSP

Firstly, go to folder:

cd Sym-NCO-POMO/TSP/

Test

Sym-NCO test

python test_symnco.py

You can change "test_batch_size", "aug_factor" (related to sample width), "aug_batch_size".

POMO baseline test

python test_baseline.py

Training

Sym-NCO training

python train_symnco.py

POMO training

python train_baseline.py

CVRP

Firstly, go to folder:

cd Sym-NCO-POMO/CVRP/

Test

Sym-NCO test

python test_symnco.py

You can change "test_batch_size", "aug_factor" (related to sample width), "aug_batch_size".

POMO baseline test

python test_baseline.py

Training

Sym-NCO training

python train_symnco.py

POMO training

python train_baseline.py

Sym-NCO-AM

Sym-NCO can be also applied to vanilla AM model.

AM is more expandable to solve various problems including TSP,CVRP,PCTSP and OP.

We provide pretrained Sym-NCO based AM model for PCTSP and OP.

Firstly, go to folder:

cd Sym-NCO-AM/

Test

General

python eval.py --dataset_path [YOUR_DATASET] --model [YOUR_MODEL] --eval_batch_size [YOUR BATCH SIZE] -- augment [SAMPLE WIDTH]

PCTSP reproduce

python eval.py --dataset_path 'data/pctsp100_test_seed1234.pkl' --model pretrained_model/pctsp_100/epoch-99.pt 

OP reproduce

python eval.py --dataset_path 'data/op_dist100_test_seed1234.pkl' --model pretrained_model/op_100/epoch-99.pt 

Train

General

python run.py --problem [Target Problem ('tsp', 'cvrp', 'pctsp_det', 'op')] --N_aug [L: problem symmetric width]

Example

python run.py --problem 'tsp' --N_aug 10 

Dependencies (Same with AM)

• Python>=3.8
• NumPy
• SciPy
• PyTorch>=1.7
• tqdm
• pytz
• Matplotlib (optional, only for plotting)

Further Work

Symmetric learning gives powerful benefits for combinatorial optimization. However, there are no remaining rooms at DRL. Instead, imitation learning with sparse data setting can be an alterative benchmark for further work. If you want to do further work of Sym-NCO, please check out Sym-NCO-IL.

Paper

This is official PyTorch code for our paper Sym-NCO: Leveraging Symmetricity for Neural Combinatorial Optimization which has been accepted at [NeurIPS 2022].

@article{kim2022sym,
  title={Sym-NCO: Leveraging Symmetricity for Neural Combinatorial Optimization},
  author={Kim, Minsu and Park, Junyoung and Park, Jinkyoo},
  journal={arXiv preprint arXiv:2205.13209},
  year={2022}
}