Spatio-Temporal Graph Structure Learning for Traffic Forecasting
Authors: Qi Zhang, Jianlong Chang, Gaofeng Meng, Shiming Xiang, Chunhong Pan1177-1185
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensively comparative experiments on six real-world datasets demonstrate our proposed approach significantly outperforms the state-of-the-art ones. |
| Researcher Affiliation | Academia | Qi Zhang,1,2 Jianlong Chang,1,2 Gaofeng Meng,1 Shiming Xiang,1,2 Chunhong Pan1 1National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences 2School of Artificial Intelligence, University of Chinese Academy of Sciences {qi.zhang2015, jianlong.chang, gfmeng, smxiang, chpan}@nlpr.ia.ac.cn |
| Pseudocode | No | The paper does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any statement or link indicating that the source code for the methodology described is publicly available. |
| Open Datasets | Yes | Our model is verified on six traffic datasets. Three of them (Pe MS-S, Pe MS-BAY and METR-LA) are public datasets released by previous work (Li et al. 2018; Yu, Yin, and Zhu 2018; Jagadish et al. 2014) and others (BJF, BRF, BRFL) are generated by ourselves. |
| Dataset Splits | No | Grid search strategy is executed to choose hyper-parameters on validation. |
| Hardware Specification | Yes | All experiments are conducted on a 64-bit Linux Server with 2.40 GHz CPU and NVIDIA Titan GPU. |
| Software Dependencies | No | The paper does not specify version numbers for any software dependencies or libraries used for the experiments. |
| Experiment Setup | Yes | Hyper-parameters Setting. The two kernel ranges (Cs, Cd) of global SLC are set to 6 and the kernel size (K) of local SLC is set to 8. The temporal depth of P3D is set to 3. For the sake of fairness, GCNN methods adopt the architecture consisting 3 layers (blocks), where the output channels of three layers are 32, 32, Q, respectively. Q is the number of time intervals of predicted results. We train our models by minimizing MAE and use Adma as optimizer. |