We provide the code to learn to solve 16 standard combinatorial optimization problems.

The multi-task model is pre-trained on eight problems:

• the Asymetric Traveling Salesman Problems (ATSP)
• the Capacitated Vehicle Routing Problem (CVRP)
• the Capacitated Vehicle Routing Problem with Time Windows (CVRPTW)
• the Orienteering Problem (OP)
• the Knapsack Problem (KP)
• the Minimum Vertex Covering (MVC)
• the Unrelated Parallel Machihe Scheduling (UPMS)
• the Job-Shop Scheduling Problem (JSSP)

Fine-tuning (supervised and unsupervised) is possible on eight new problems:

• the Traveling Repairman Problem (TRP)
• the Price Collecting Traveling Salesman Problem (PCTSP)
• the Split-Delivery Capacitated Vehicle Routing Problem (SDCVRP)
• the Open Capacitated Vehicle Routing Problem (OCVRP)
• the Sequential Ordering Problem (SOP)
• the Maximal Independent Set (MIS)
• the Maximum Covering Location Problem (MCLP)
• the Open-Shop Scheduling Problem (OSSP)

See GOAL: A Generalist Combinatorial Optimization Agent Learner for the paper associated with this codebase. If you find this code useful, please cite our paper as:

@article{
   drakulic2025goal,
   title={GOAL: A Generalist Combinatorial Optimization Agent Learner},
   author={Darko Drakulic and Sofia Michel and Jean-Marc Andreoli},
   booktitle={International Conference on Learning Representations (ICLR)},
   year={2025},
   url={https://openreview.net/forum?id=z2z9suDRjw},
}

python3 train.py 
  --problems [list_of_problems] 
  --train_datasets [list_of_train_datasets] 
  --val_datasets [list_of_val_datasets] 
  --test_datasets [list_of_test_datasets]

For all training arguments

python3 train.py -h

python3 test.py 
  --problems [tsp|cvrp|cvrptw|op|kp|upms|jssp] 
  --test_datasets [list_of_test_datasets]
  --pretrained_model pretrained/multi.best

For all test options

python3 test.py -h

python3 finetune.py
  --problems [trp|pctsp|ocvrp|sdcvrp|sop|mis|mclp|ossp]
  --train_datasets [train_dataset]
  --test_datasets [test_dataset]
  --pretrained_model pretrained/multi.best

python3 finetune.py
  --problems [trp|pctsp|ocvrp|sdcvrp|sop|mis|mclp|ossp]
  --test_datasets [test_dataset]
  --pretrained_model pretrained/multi.best

Test and fine-tuning data are provided in the data/ directory.

Training datasets can be downloaded from here.

Pretrained models are provided in the pretrained/ directory.