Gated Residual Recurrent Graph Neural Networks for Traffic Prediction

Authors: Cen Chen, Kenli Li, Sin G. Teo, Xiaofeng Zou, Kang Wang, Jie Wang, Zeng Zeng485-492

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on two traffic datasets have demonstrated that the proposed MRes-RGNN outperforms state-of-the-art methods significantly.
Researcher Affiliation Collaboration Cen Chen, Kenli Li, * Sin G. Teo, Xiaofeng Zou, Kang Wang Jie Wang, Zeng Zeng * College of Information Science and Engineering, Hunan University, China Institute for Infocomm Research, Singapore
Pseudocode No The paper provides equations and architectural diagrams but does not include explicit pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement about the release of open-source code or a link to a code repository for the described methodology.
Open Datasets Yes Two large real-world datasets, i.e., METR-LA and PEMS-BAY datasets, are used in the experiment: METR-LA: Traffic data are collected from observation sensors in the highway of Los Angeles County. We use 207 sensors and 4 months of data dated from 1st Mar 2012 until 30th Jun 2012 in the experiment. PEMS-BAY: Traffic data are collected by California Transportation Agencies Performance Measurement System (Pe MS). We use 325 sensors in the Bay Area and 6 months of data dated from 1st Jan 2017 until 31th May 2017 in the experiment.
Dataset Splits Yes In each dataset, 70%, 20% and 10% of its dataset are split into training, validation and testing datasets, respectively.
Hardware Specification Yes We configure a Linux server and the other configurations to run the experiment as follows: 8 Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHZ; 256GB RAM; 4 NVIDIA P100 GPUs.
Software Dependencies No The paper mentions 'Linux server' but does not provide specific software dependencies with version numbers (e.g., Python, TensorFlow/PyTorch versions).
Experiment Setup Yes In our implementation, two Res-RGNN layers are utilized, and the graph convolution kernel size is set to 64. To have a fair comparison, these parameters are set same in DCRNN. The proposed framework utilizes one Res-RGNN branch for near time, and one hop1 Res-RGNN branch for daily period. In the above Res-RGNN branch, 6 observed data points are used to forecast traffic conditions. Another branch, the hop Res-RGNN for daily period, 4 historical data points are utilized in the experiments.