NeuroMLR: Robust & Reliable Route Recommendation on Road Networks
Authors: Jayant Jain, Vrittika Bagadia, Sahil Manchanda, Sayan Ranu
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through in-depth experiments on realworld datasets, we establish that NEUROMLR imparts significant improvement in accuracy over the state of the art. More importantly, NEUROMLR generalizes dramatically better to unseen data and the recommended routes reach the destination with much higher likelihood than existing techniques. and 5 Experiments In this section, we benchmark NEUROMLR against DEEPST and CSSRNN and establish that: Accuracy: NEUROMLR is more accurate in terms of precision and recall when compared to the state-of-the-art algorithms of DEEPST [10] and CSSRNN [27]. Reachability: NEUROMLR, with its greedy route search mechanism, is more efficient, and achieves significantly higher reachability than DEEPST and CSSRNN. Inductive Learning: Due to its inductive learning capability, NEUROMLR learns more effectively and generalizes significantly better to unseen/lesser seen parts of the road network. Scalability: NEUROMLR generates high quality routes on large road networks. The performance of CSSRNN, on the contrary, deteriorates heavily with increase in road network size(Fig. 1c). |
| Researcher Affiliation | Academia | Jayant Jain , Vrittika Bagadia , Sahil Manchanda, Sayan Ranu Indian Institute of Technology Delhi {jayantjain100,vrittikabagadia}@gmail.com {sahil.manchanda,sayanranu}@cse.iitd.ac.in |
| Pseudocode | Yes | The pseudocode for this search algorithm is provided in App. A. and The pseudocode of the greedy approach can be found in Alg. 2 in App. B. and The pseudocode of the training procedure can be found in Alg. 3 in App. D. |
| Open Source Code | Yes | Our code-base is available at https://github.com/idea-iitd/Neuro MLR. |
| Open Datasets | Yes | Datasets: We use publicly available real datasets from five different cities. Table 1 summarizes the statistics of the datasets. The first four cities namely Chengdu4, Porto[15], Harbin [11] and Beijing[12] are taxi datasets. The fifth dataset is a publicly available food delivery dataset[7]. |
| Dataset Splits | Yes | Before splitting, we sort the trajectories on the basis of the start timestamp. Unless specifically mentioned, we use the first 60% of the trajectories for training, next 20% for validation and remaining 20% for inference. |
| Hardware Specification | Yes | The system configuration details are present in App. F. (Appendix F: All experiments are conducted on a single server equipped with an Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz, 2 NVIDIA GeForce RTX 2080 Ti GPUs (11 GB memory), and 512 GB RAM.) |
| Software Dependencies | Yes | The codebase of CSSRNN, shared by the authors, is implemented in Tensor Flow 1.15. DEEPST and NEUROMLR are implemented in Py Torch 1.6.0. |
| Experiment Setup | Yes | Parameters: The default parameters for NEUROMLR are provided in App. G. (Appendix G provides specific parameters such as "Number of Anchors (k): 100", "GCN Layers (L): 2", "MLP Layers (LM): 2", "GCN Hidden Dimensions: 128", "MLP Hidden Dimensions: 256", "Learning Rate: 0.001", "Batch Size: 256", "Dropout Rate: 0.5".) |