Home

Awesome

ASTGCN

Attention Based Spatial-Temporal Graph Convolutional Networks for Traffic Flow Forecasting (ASTGCN)

model architecture

References

Shengnan Guo, Youfang Lin, Ning Feng, Chao Song, Huaiyu Wan(*). Attention Based Spatial-Temporal Graph Convolutional Networks for Traffic Flow Forecasting. The 33rd AAAI Conference on Artificial Intelligence (AAAI'19) 2019.

Datasets

We validate our model on two highway traffic datasets PeMSD4 and PeMSD8 from California. The datasets are collected by the Caltrans Performance Measurement System (PeMS) (Chen et al., 2001) in real time every 30 seconds. The traffic data are aggregated into every 5-minute interval from the raw data. The system has more than 39,000 detectors deployed on the highway in the major metropolitan areas in California. Geographic information about the sensor stations are recorded in the datasets. There are three kinds of traffic measurements considered in our experiments, including total flow, average speed, and average occupancy.

We provide two dataset: PEMS-04, PEMS-08

  1. PEMS-04:

    307 detectors
    Jan to Feb in 2018
    3 features: flow, occupy, speed.

  2. PEMS-08:

    170 detectors
    July to Augest in 2016
    3 features: flow, occupy, speed.

Requirements

To install MXNet correctly, you should follow the instruction provided by this page.

To run mxboard, you have to install tensorboard.

Other dependencies can be installed using the following command:

pip install -r requirements.txt

If you are using docker, install nvidia-docker and run the commands below:

# build image
docker build -t astgcn/mxnet:1.4.1_cu100_mkl_py35 -f docker/Dockerfile .

# training model in background
docker run -d -it --rm --runtime=nvidia -v $PWD:/mxnet --name astgcn astgcn/mxnet:1.4.1_cu100_mkl_py35 python3 train.py --config configurations/PEMS04.conf --force True

Usage

train model on PEMS04:

python train.py --config configurations/PEMS04.conf --force True

train model on PEMS08:

python train.py --config configurations/PEMS08.conf --force True

visualize training progress:

tensorboard --logdir logs --port 6006

then open http://127.0.0.1:6006 to visualize the training process.

Improvements

  1. We use convolutional operation to map the output of ASTGCN block to the label space because that can help the model achieve a better performance.

Configuration

The configuration file config.conf contains three parts: Data, Training and Predict:

Data

Training