Generalize Learned Heuristics to Solve Large-scale Vehicle Routing Problems in Real-time

Authors: Qingchun Hou, Jingwei Yang, Yiqiang Su, Xiaoqing Wang, Yuming Deng

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experiments on synthetic and real-world large-scale VRPs show our method could generalize the learned heuristics trained on datasets of VRP 100 to solve VRPs with over 5000 nodes in real-time while keeping the solution quality better than data-driven heuristics and competitive with traditional heuristics.
Researcher Affiliation Collaboration Qingchun Hou, Jingwei Yang, Yiqiang Su, Xiaoqing Wang, Yuming Deng Alibaba Group houqingchun.hqc@alibaba-inc.com, hqc16@tsinghua.org.cn
Pseudocode Yes Algorithm 1 in Appendix A.3.1 reports the implementation details of our training algorithm... The final training algorithm for TAM is shown in Algorithm 1.
Open Source Code No The paper does not include an explicit statement or link indicating that the code for their proposed method is open-source or publicly available.
Open Datasets No We sample 1280000 CVRP100 instances on the fly as training datasets.
Dataset Splits No The paper specifies training and testing sets, stating: 'We sample 1280000 CVRP100 instances on the fly as training datasets. And then we test our TAM and benchmark algorithms on 100 instances for CVRP 100, 400, 1000, 2000, 5000, and 7000.' However, it does not explicitly define a separate validation dataset split with specific percentages or counts.
Hardware Specification Yes All models are trained on a single GPU Tesla V100.
Software Dependencies No The paper mentions specific software tools like 'LKH3' and 'Ortools' and architectural components like 'Transformer' and 'Graph neural network', but it does not specify exact version numbers for any of these software dependencies.
Experiment Setup Yes For a fair comparison, all our training hyperparameters are the same as AM. We trained TAM for 100 epochs with batch size 512 on the generated dataset. We choose 3 layers transformer in the encoder with 8 heads. The learning rate is constant η = 10 4. All models are trained on a single GPU Tesla V100. The rest of the parameters such as dimension of hidden state is listed in Kool et al. (2019).