Spatiotemporal Multi-Graph Convolution Network for Ride-Hailing Demand Forecasting
Authors: Xu Geng, Yaguang Li, Leye Wang, Lingyu Zhang, Qiang Yang, Jieping Ye, Yan Liu3656-3663
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate the proposed model on two real-world large scale ride-hailing demand datasets and observe consistent improvement of more than 10% over state-of-the-art baselines. We conduct experiments on two real-world large scale ride-hailing datasets collected in cities: Beijing and Shanghai. |
| Researcher Affiliation | Collaboration | 1Hong Kong University of Science and Technology, 2University of Southern California, 3Peking University, 4Didi AI Labs, Didi Chuxing |
| Pseudocode | No | The paper includes architectural diagrams (Figure 2, Figure 4) but no explicit pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any statement about releasing source code or a link to a code repository for the described methodology. |
| Open Datasets | Yes | The road network data used for transportation connectivity evaluation is provided by Open Street Map (Haklay and Weber 2008). |
| Dataset Splits | Yes | For data split, we use the data from Mar 1st 2017 to Jul 31st 2017 for training, data from Aug 1st 2017 to Sep 30th 2017 as validation, and the data from Oct 1st 2017 to Dec 31st 2017 is used for testing. |
| Hardware Specification | Yes | The training of ST-MGCN takes 10GB RAM and 9GB GPU memory. The training process takes about 1.5 hour on a single Tesla P40. |
| Software Dependencies | No | All neural network based approaches are implemented using Tensorflow (Abadi and others 2016), and trained using the Adam optimizer (Kingma and Ba 2015) for minimizing RMSE. |
| Experiment Setup | Yes | Following the practice in (Zhang, Zheng, and Qi 2017), the input of the network consists of 5 historical observations, including 3 latest closeness components, 1 period component and 1 latest trend component. In the experiment, f(A; θi) in Equation 6 is chosen to be the Chebyshev polynomial functionl (Defferrard, Bresson, and Vandergheynst 2016) of the graph Laplacian with the degree K equals to 2, and F is chosen to be the sum aggregation function. The number of hidden layers is 3, with 64 hidden units each and an L2 regularization with a weight decay equal to 1e-4 is applied to each layer. Specially, the graph convolution degree K in CGRNN equals to 1. We use Re LU as the activation in the graph convolution network. The learning rate of ST-MGCN is set to 2e-3, and early stopping on the validation dataset is used. All neural network based approaches are implemented using Tensorflow (Abadi and others 2016), and trained using the Adam optimizer (Kingma and Ba 2015) for minimizing RMSE. |