ADVERSPARSE

This is the original pytorch implementation of ADVERSPARSE in the following paper:

[ADVERSPARSE: AN ADVERSARIAL ATTACK FRAMEWORK FOR DEEP SPATIAL-TEMPORAL GRAPH NEURAL NETWORKS, ICASSP 2022]

Data prepration

Download METR-LA and PEMS-BAY data from Google Drive or Baidu Yun links provided by DCRNN. You also need to download the file sensor_graph provided by DCRNN.

Create data directories

mkdir -p data/{METR-LA,PEMS-BAY}

mkdir garage

METR-LA

python generate_training_data.py --output_dir=data/METR-LA --traffic_df_filename=data/metr-la.h5 --seq_length_x=12 ----seq_length_y=12

PEMS-BAY

python generate_training_data.py --output_dir=data/PEMS-BAY --traffic_df_filename=data/pems-bay.h5 --seq_length_x=12 ----seq_length_y=12

Note that the readings of the sensors are aggregated into 5-minutes windows, therefore 15-minute-, 30-minute-, and 60-minutes-ahead predictions correspond to the seq_length equal to 3, 6, 12 resepctively. You need to set the seq_length_x and seq_length_y based on your prediction when processing raw data.

Pretrained step

python train.py

Note that you may need to change some settings about the parameters in train.py based on the dataset you are running. For example, --data (data/PEMS-BAY or data/METR-LA), --seq_length(3, 6, or 12), ----num_nodes(PEMS-BAY:325 and METR-LA:207), --save(./garage/metr or ./garage/pems).

ADMM training step

python train_x.py

Note that you may need to change some settings about the parameters in train.py based on the dataset you are running. For example, --data (data/PEMS-BAY or data/METR-LA), --seq_length(3, 6, or 12), ----num_nodes(PEMS-BAY:325 and METR-LA:207), --save(./garage/metr or ./garage/pems), --best_pretrained_model(load the pretrained model whose file name has "best"), --keep(vary from 10% to 90%), --rho, --ADMMtimes, --epoch