Awesome

CTVI-master

For ICDM 2021. Accepted as short paper.

The implementation of the CTVI model(Citywide Traffic Volume Inference).

Paper: Temporal Multi-view Graph Convolutional Networks for Citywide Traffic Volume Inference

Usage:

Install dependencies

pip install -r requirements.txt

Clone this repo

git clone https://github.com/dsj96/CTVI-master.git

Function

args.py defines some necessary parameters.

attention.py defines the multi-head temporal attention and position encoder model.

extract_city_volume_info.py is adpoted to extract and process raw city traffic volume data.

FNN.py is the implemention of three layers MLP.

jinan_optuna.py is the implemention of our model on Jinan dataset. You can run and evaluate the model by executing this code file. And if you want to change the range of optuna hyperparameters, you can modify the objective function in jinan_optuna.py file.

metrics.py is used to evaluate our model and print log information.

utils.py file is mainly used to implement some data processing functions.

walk.py file is mainly used to generate random walk sequence on affinity graph.

jinan.zip file is mainly to provide a toyset to run.

Data

We conduct our experiments on Hangzhou and Jinan cities in China. Due to privacy issues, we public part of the Jinan traffic vloume data in an anonymous form. The preprocessed data is included in jinan.zip file (just unzip the file under the current file path).

Split Data

We randomly split the road segments with traffic volume data into training (80%) and testing (20%), respectively. We further select 20% of the training randomly as validation.

Note that for the selected road segments used for testing, we completely masked its traffic volume information in all time slice. Afterwards we use CTVI model to inference the traffic volume values in each time slice.

Roadnet Data Format

roadnet.txt: intersection0_intersection1, num_of_lanes, speed limit, road segment name

cams_attr.txt: sensor ID, intersection0_intersection1, num_of_lanes, road grade, speed limit, road segment name

Volume File Format

Each file contains the real time traffic volume information in the following form:

File name 8_1_8_0_4.volume denotes the traffic volume values at August 1, 8:00:00 to 8:04:59.

In each file: 5_6,172: intersection0_intersection1, traffic volume

Training and Evaluate

You can train and evaluate the model by run jinan_optuna.py file.

Cite

Please cite our paper if you find this code useful for your research:

@inproceedings{dai2021temporal,
  title={Temporal Multi-view Graph Convolutional Networks for Citywide Traffic Volume Inference},
  author={Dai, Shaojie and Wang, Jinshuai and Huang, Chao and Yu, Yanwei and Dong, Junyu},
  booktitle={ICDM},
  pages={1048--1053},
  year={2021}
}