OMNI¶

Omni focuses on the generalization challenge of neural methods in Vehicle Routing Problems (VRPs) and proposes a universal meta-learning framework to achieve “full generalization” across problem scales and distributions.

Usage¶

You can run the following command lines to execute the code.

Omni train in diverse scale [50,200]

python train.py settings=omni_settings mode=train problem={tsp/cvrp}
python eval.py settings=omni_settings mode=eval problem={tsp/cvrp} scale={scale}

Tasks¶

Supported Tasks: TSP, CVRP

Required Data Generator:

TSPGenerator
CVRPGenerator

Training¶

Omni applies autoregressive reinforcement learning class ARREINFORCELightning() for training.